Paul M Churchland A Neurocomputational Perspective The Nature of Mind and The Structure of Science 1989

A
BRADFORD BOOKS
A NEUROCOMPUTATIONAL PERSPECTIVE
The Nature of Mind and the Structure of S cie n ce
Paul M. Churchland NEUROCOMPUTATIONAL
"O ccasionally s o m e o ne does something genuinely creative in philosophy. Such a c
complishments are rare enough that w hen they do o ccur they are not too hard to spot
If doubts had remained, the publication of Paul Churchland’s A Neurocomputational
Perspective: The Nature of Mind and the Structure of Science surely s e c u r e s his
P e r s p e c tiv e
position a s one of today's most creative philosophers."
— Robert N. McCauley, Philosophical Psychology The Nature of Mind and
If w e are to solve the central problems in the philosophy of science, Paul Churchland the Structure of Science
argues, w e must d raw heavily on the resou rce s of the emerging s c ie n c e s of the mind-
brain. A Neurocomputational Perspective illustrates the fertility of the con cepts and
data drawn from the study of the brain and of artificial networks that model the brain.
T hese co n cepts bring unexpected coh ere n ce to scattered issues in the philosophy
of s cie n c e, n ew solutions to old philosophical problems, and n ew possibilities for the
enterprise of s c ie n c e itself.
Paul M. Churchland is P rofessor of Philosophy and a member of the Cognitive
S c ie n c e Faculty at the University of California at San Diego.
The MIT Press

M a ssa c h u s e tts Institute of Technology
Cambridge, M a s sa c h u s e tts 02142
CHUPP 978026253106190000
0-262-53106-2
PAUL M. CHURCHLAND
A Neurocomputational Perspective
The Nature of Mind and the
Structure of Science
Paul M. Churchland
A Bradford Book
The MIT Press
C am bridge, M assachusetts
L on d on , England
First МIT Press paperback edition, 1992
© 1989 Massachusetts Institute of Technology
All rights reserved. No part of this book may be reproduced in any form by any electro
nic or mechanical means (including photocopying, recording, or information storage
and retrieval) without permission in writing from the publisher.
This book was set in Palatino by Asco Trade Typesetting Ltd., Hong Kong, and printed
and bound in the United States of America.
Library of Congress Cataloguing-in-Publication Data
Churchland, Paul M ., 1942-

A neurocomputational perspective: the nature of mind and the
structure of science/Paul M. Churchland.
p. cm.
Bibliography: p.
Includes index.
ISBN 0-262-03151-5 (HB), 0-262-53106-2 (PB)
1. Intellect. 2. Consciousness. 3. Cognition. 4. Neural
computers. 5. Science— Philosophy. 6. Psychology—
Philosophy. 7. Science and psychology. I. Title.
BF431.C48 1989
128'. 1— dc20 89-9398
CIP
For M ark and A nne,
tw o lam ps against the darkness
Contents
Preface xi
PART I
The N atu re of Mind
C h ap ter 1
Elim inative M aterialism and the Prepositional A ttitudes 1
1 W hy F olk Psychology Is a Theory 2
2 W hy F olk P sychology M ight (Really) Be False 6
3 A rgum ents against Elim ination 9
4 The Conservative N ature o f Functionalism 12
5 B eyond F olk P sychology 17
C h ap ter 2
Functionalism , Qualia, and Intentionality 23
1 F ou r P roblem s concerning Q ualia 23
2 The Problem o f N onstandard R ealizations 34
3 Functionalism and M ethodology 44
C h ap ter 3
R eduction, Qualia, and the Direct Introspection of Brain States 47
1 Intertheoretic Reduction 47
2 T heoretical Change and P erceptual Change 52
3 Thom as N agel's A rgum ents 56
4 Jackson's K now ledge A rgum ent 61
C h ap ter 4
K now ing Qualia: A Reply to Jackson 67
1 The P ersisten t Equivocation 67
2 O ther Invalid Instances 71
3 A G enuinely N onequivocal Know ledge A rgum ent 72
4 C onverting a Third-Person A ccount into a First-Person A ccount 74
viii Contents
C h apter 5
Som e Reductive Strategies in C ognitive N eurobiology 77
1 Introduction 77
2 L am inar Cortex, Vertical C onnections, and T opographic M aps 79
3 Sensorim otor Coordination 82
4 C oordinate T ransform ation: Its Physical Im plem entation 88
5 Cortex w ith M ore than Two Layers 96
6 Beyond State-Space Sandwiches 97
7 The R epresen tation al P ow er o f State Spaces 102
8 C oncluding R em arks 108
C h apter 6
Folk Psychology and the Explanation of H um an Behavior 111
1 O bjections to the T heoretical V iew 112
2 A n A lternative Form o f K now ledge R epresentation 122
3 A dden du m : C om m entary on D ennett 125
C h apter 7
R eductionism , C onnectionism , and the Plasticity of H um an
C on sciou sness 129
1 The P lasticity A rgum ent 131
2 The C ulturaTE m bedding O bjection 132
3 Conclusion 134
PART II
The Structure of Science
C h ap ter 8
The O ntological Status of O bservables: In Praise of the
Superem pirical Virtues 139
1 H ow van F raassen ’s Problem C ollapses into H ume's Problem 140
2 The Prim acy o f the Superem pirical Virtues 145
3 T ow ard a M ore R ealistic Realism 149
C h apter 9
O n the N atu re of Theories: A N eurocom putational Perspective 153
1 The C lassical View o f Theories 153
2 Problem s an d A lternative A pproaches 154
3 E lem entary Brainlike N etw orks 159
4 R epresentation and Learning in B rainlike N etw orks 163
5 Som e F unctional P roperties o f Brainlike N etw orks 171
6 H ow F aithfu lly D o These N etw orks D epict the Brain? 181
7 C om putational N euroscience: The N aturalization o f Epistem ology 188
8 C oncluding R em arks 194
Contents
C h apter 10
On the N atu re of Explanation: A PDP A p p roach 197
1 Introduction 197
2 C onceptual O rganization in P D P N etw orks 200
3 R ecognition and U nderstanding 206
4 P rototype A ctivation : A U nified Theory o f Explanation 209
5 Inference to the Best Explanation 218
6 Com parison w ith Earlier M odels 223
C h ap ter 11
Learning and C onceptual C hange 231
1 Introduction 231
2 M u ltiple C onceptual Com petence 234
3 C onceptual Change versus C onceptual R edeploym ent 236
4 W hat D rives Conceptual Change? 241
5 A utom ated Science 250
C h ap ter 12
P erceptual Plasticity and Theoretical N eutrality: A Reply to Jerry
F od or 255
1 The Etiology o f P erceptual B elief 257
2 The Sem antics o f O bservation Predicates 271
3 Conclusion 278
C h ap ter 13
C on ceptu al P rogress and W ord-W orld Relations: In Search of the
E ssen ce of N atural Kinds 281
1 N atu ral Kinds and Scientific P rogress: The P utnam -K ripke View 282
2 N atu ral Kinds as Law -B ound K inds: Som e Virtues, Consequences, and
D ifficulties 287
C h ap ter 14
M oral F acts and M oral K now ledge 297
1 The E pistem ology and O ntology o f M orals 297
2 M oral Prototypes and M oral D evelopm ent 298
3 P raxis, T heoria, and Progress 300
References 305
Index 315
г
Preface
The single m ost im portant developm ent in the philosophy of m ind

during the p ast forty years has been the em erging influence of the
philosophy of science. Issues that had been the special province of
m etaphysicians and philosophers of language cam e increasingly to be
seen as issues of the sam e sort as those found elsew here in our sci
entific history. The questions surrounding the nature of m ind and
consciousness have slowly come to be represented as empirical ques
tions, high level perhaps, but solvable eventually by the m eth od s of
theoretical science. The sem inal thinkers w ho launched the philoso
p hy of m ind on this dram atic new cou rse— Feigl, Sm art, Sellars, and
Feyerabend spring im m ediately to m ind— w ere also and w ithout ex
ception first-rate philosophers of science. Since then it has hardly
been possible to do any system atic w ork in the philosophy of m ind,
or even to un derstand the debates, w ithout draw ing heavily on
them es, com m itm ents, or an tecedent expertise draw n from the phi
losophy of science.
E xam ples of this infusion are close at hand. The natu re of explana
tion in general and of psychological explanations in particular becam e
an issue central to philosophers of m ind because the conceptual
fram ew ork w ith w hich w e conceive of ourselves (as creatu res with
beliefs, desires, and other m ental states) cam e to be seen as a
com m on -sen se explanatory theory— "folk p sy ch o lo g y "— a theory
that enables each of us to predict and explain the behavior of the
o ther h u m ans that surround us. This m ade an issue in turn of the
status of theoretical entities (Are m ental states real?), the nature of
cross-theoretic identities (Are m ental states identical w ith brain
states?), and the criteria for intertheoretic reduction (Could folk
p sych ology itself eventually be explained in turn by a deep er and
m ore neurobiological theory of h um an nature?).
O ntological concerns w ere joined by epistem ological concerns.
W h eth er our noninferential know ledge m u st alw ays be laden w ith
prior theoretical com m itm ents becam e a further focus of dispute b e
xii Preface
cause the ap paren t authority of introspection w as by this m eans called

into question (C an one know by sheer introspection that the mind is
distinct from the brain?), and because it held out the intriguing possi
bility of conceptual change even at the introspective level (M ight one
learn to introspect, to be "directly aw are" of, o n e's ow n neurochem ic
al and neurocom putational states?).
E ven the highly abstract issue of the desirability of an ultim ately
unified science of N ature has becom e a bone of fierce contention
am ong philosophers of m ind. This is b ecause the p ressure on folk
psychology to find an integrated place within such a unified exp lan a
tory schem e, or else be throw n on the scrap heap of history, will be a
function of how high a value w e place on the unity of science. Philo
sop hers such as Jerry Fod or, Donald D avidson, and D an D ennett, in
their quite different w ays (nam ely, positive disunity in science, co n
ceptual dualism , and selective instrum entalism ), place a sharply
restricted value on unity and thus feel able to claim a separate and
au ton om ou s place for folk p sychology safe from the advancing tide of
natural science. O ther philosophers, such as W . V. Q uine, Patricia
C hurchland, Cliff H ooker, and m yself, place a very high value on
unity, and thus insist that folk p sychology cannot be p rotected in
definitely from having to m eet the obligations that all acceptable
theories m u st ultim ately m eet. Either it m u st be successfully reduced
(to a m atured cognitive neurobiology, for exam ple), p erhaps u n d er
going som e m odification in the process. O r it m u st be displaced by a
b etter theory, one that does cohere w ith the rest of our grow ing scien
tific corpus.
In these w ays, and in others, problem s in the philosophy of mind
have found them selves system atically recon stru cted as problem s in
the philosophy of science. This infusion of inform ation from a highly
developed area of philosophy has transform ed the philosophy of
m ind by placing it w ithin a new and m ore general con text of im p or
tant questions and plausible answ ers, and m ore im portant still, by
placing it firmly in contact w ith ongoing developm ents in the adja
cen t natural sciences. The essays in part 1 of this volum e constitute
one journey through the m ajor problem s in the philosophy of m ind,
one attem p t to bring the m ost im portant results in the philosophy of
science to b ear on them . They also attem p t to apply the m ost recent
results in com putational neuroscience and connectionist AI to the
m any problem s at hand. The problem of subjective sensory qualia is
salient h ere, as is the the problem of m ental representation. Both
h ave been central to philosophical discussions about the natu re of
m ind, and both have been sharply illuminated by recent develop
m en ts in physiological p sychology and cognitive neurobiology.
Preface xiii
T hough published independently over a period of seven years, the

essays of part 1 constitute an unfolding argum ent for a unified m ater
ialist position concerning the m ind. The first and earliest of the essays
here assem bled, "Elim inative M aterialism and the Propositional A tti
tu d e s" (1981a), picks up on a them e that concluded m y 1979 book.
That them e urges the need for a new cognitive kinem atics and cor
relative dynam ics, one that is radically different from the venerable
senten ce-crun chin g paradigm com m on to folk p sych ology, orthodox
cognitive science, and conventional AI. The new kinem atics and
dynam ics, I argued, should be draw n from , or inspired by, the
em erging sciences of the brain, and I there presu m ed to sketch som e
general features that the new cognitive paradigm should display
(pp. 1 7 -1 8 ). A t the time I w ould have guessed that a new p ara
digm m eeting the stated conditions w as at least tw enty-five years
aw ay, and probably m ore like fifty.
In this I w as w rong, for in fact it already existed and had existed, at
least in stick-figure form , since the late fifties. By 1959 F. Rosenblatt
had developed the Percep tron paradigm of vector-to-vector tran s
form ations in a parallel netw ork of neuronlike p rocessing units.
These artificial netw orks could learn to recognize p attern s by the re
p eated presentation of similar exam ples. U nfortunately, that p ara
digm did not catch on, and for tw o decades it w as alm ost forgotten—
its resou rces pursued by only a small handful of research ers: a few
w ithin neuroscience and even few er within AI. It stum bled as it left
the starting blocks partly because of the limited success of its earliest
netw ork s, partly because of som e slightly overstated b ut w idely in
fluential criticisms by M. Minsky and S. Pap ert (1969), but m ostly
b ecau se of the sim ultaneous eclipse of this im m ature paradigm by the
very real virtues of conventional general-purpose com p u ters and the
m ore im m ediate rew ards of program -w riting AI. The AI profession
flocked enthusiastically to this nonbiological alternative, and the
neural-n etw ork approach faded to invisibility.
W ithin tw o years of the 1981a p aper Patricia C hurchland and I w ere
fortunate to stumble across this obscure vector-transform ation-in-a-
parallel-netw ork m odel during our m ean d ers through the field of
n eu roscien ce. Specifically, A ndras Pellionisz and Rodolfo Llinas had
developed a theoretical m odel of the cerebellum that sh ow ed us how
such a m assively parallel system could solve com plex problem s in
m otor control and sensorim otor coordination. But w e, as they, could
see th at its potential applications w ere m u ch b roader. Som e com puter
sim ulations of sensorim otor coordination on m y ow n m odest
m achine deepened m y acquaintance w ith its virtues, and the paper
xiv Preface
"S o m e Reductive Strategies in C ognitive N eurobiology" w as w ritten

w ith som e excitem ent in the spring of 1984.
That p aper provides a detailed articulation of som ething I had been
arguing for and seeking since 1971: a genuine alternative cognitive
paradigm , one firmly grounded in neuroscience, one that m ight
eventually be developed so as to reduce or displace the sentential
paradigm of folk p sychology. M ost of m y w ork since 1984 has been
aim ed at exploring that prom ise. But the account in that p aper also
has a m ajor lacuna. It outlines no p roced u res by w hich the relevant
netw orks can arrive at the specific configurations n ecessary to execute
their cognitive achievem ents. It contains, that is, no accou n t of func
tional plasticity, or learning.
That piece of the puzzle w as still m issing as, in the fall of 1 9 8 4 ,1 left
the U niversity of M anitoba to join the philosophy dep artm ent at the
U niversity of California, San D iego, and becam e a m em ber of the
cam pu s cognitive-science faculty. This interdisciplinary group of
p erh ap s tw o dozen scholars held a w eekly sem inar organized m ainly
by David R um elhart. There I discovered som ething I had n ot appreci
ated before arriving. The vector-transform ation-by-parallel-netw ork
paradigm had been a topic of research am on g U C SD 's cognitive sci
entists for several years, and suspicion of the orth od ox sentential
ap p roach es to m odeling the m ind w as already w idespread in that
grou p. M oreover, during that first year or so Rum elhart, H inton, and
Williams (1986a) perfected an im p ortan t generalization of R osen
blatt's original learning algorithm , the sim ple delta rule. Their g en eral
ized delta rule, or back propagation algorithm as it is often called, w ould
train m ultilayered artificial netw orks to perform alm ost any vector-to-
vector transform ation, given repeated exp osu re to m any instances of
the transform ation desired.
The full significance of that algorithm w as only dimly appreciated
at its initial articulation (late in 1980), even by its authors. It w as bio
logically unrealistic for starters. It w as com putationally intensive and
h en ce rath er slow in execution. A nd for a w ide variety of problem s
it w as not expected to be reliable at getting anyw here close to an o p
timal solution. O n the first tw o points the w orries w ere well founded.
But on the third point the w orries p roved to be exaggerated . E x
perience in training diverse netw orks on m an y different problem s
show ed the back-propagation algorithm to be a relentlessly efficient
and highly reliable m eans of finding excellent solutions over a wide
range of problem s. Though it w as not the only learning algorithm
that had been developed, it quickly becam e the technology of choice
nationw ide for training various netw orks to see w hat they could and
could not do.
Preface xv
They could do, it began to seem , just about anything. They could
perform sensorim otor coordination, direct gaze accurately despite
ch an ges in head position, recognize subtle similarities am ong sonar
retu rn s, pronou n ce w ell-articulated speech from printed text, recog
nize three-dim ensional shapes independently of the angle of illu
m ination, p arse sentences into gram m atical types, recognize voiced
p h on em es, predict the folding of protein m olecules, correctly recog
nize colors across changes in illumination, and so forth. A n d they did
all of this by learning from a store of relevant exam ples. The practical
necessity of sim ulating the relevant parallel netw orks w ithin con ven
tional serial com puters im posed limits on how large the m odel n et
w orks could be, since the num ber of com putationally active synaptic
connections explodes as roughly the square of the num ber of n eu ron
al units in the netw ork. A nd a further limit w as im posed by the n e
cessity of com puting thousands of adjustm ents for each of these
m an y connections during the course of training w ith the back-
p ropagation algorithm . Even so, netw orks of a few hun dred units
w ere readily sim ulated, and they proved to be capable, after training,
of som e quite arresting displays of cognitive sophistication.
The questions then becam e, Just how are these netw orks doing all
this? H ow do they m anage to find unity and structure in the chaos
and diversity of their sensory inputs? W h at form s of rep resentation
are being generated inside them ? W hy do som e netw orks learn better
than others on the sam e training corpus? Fortunately, sim ulated n et
w orks, unlike biological brains, can easily be tw eaked into giving up
their hidden secrets. O ne need not insert m icroelectrodes one by one
and blindly into a m ass of fragile cells. O ne can w rite the sim ulation
p ro gram to read out on com m and the sim ultaneous and undegrad ed
m icrobehavior of entire populations of artificial n eurons.
H ere the results are highly intriguing, and to the eye of a philo
sop h er of science they are very revealing. A lm ost im m ediately one
could begin to see lessons that bear directly and system atically on
issues central to the philosophy of science: the nature of theories, the
theory-laden ness of perception, the nature of conceptual unification,
the virtues of theoretical simplicity, the nature of parad igm s, the
kinem atics of conceptual change, the character of abductive infer
ence, and the nature of explanatory understanding (see especially the
discussions in ch apters 9, 10, and 11).
This brings m e to the second m ajor them e of this volum e. I began
this preface by detailing how it is no longer possible to d o m ajor w ork
in the philosophy of m ind w ithout draw ing on them es from the phi
losophy of science and the several sciences of the m ind-brain. I w ish
now to su ggest that the instruction and inform ation h as begun to
xvi Preface
flow vigorously in the opposite direction. Very shortly it will no

longer be possible to do m ajor w ork in the philosophy of science
w ithout draw ing on them es from the philosophy of mind and from
the related disciplines of com putational n euroscience, cognitive p sy
chology, and connectionist AI.
The detailed grounds for this claim are to be found in the essays of
p art 2, especially in chapters 9 to 11, b ut the basic sketch is quickly
given. First, the relevant them e from the philosophy of m ind is the
view that the sentential kinem atics of folk p sychology is but a com -
m on sen se theory, and alm ost certainly a fa ls e theory, at least as an
accou nt of the basic kinem atics of cognitive creatures generally.
Second, the relevant them e from cognitive neurobiology and connec-
tionst AI is that the basic kinem atics of cognitive creatures is a
kinem atics not of sentences but of high-dim ensional activation v ec
tors being transform ed into other such v ectors by passing through
large arrays of synaptic connections. A new kinem atics for cognitive
activity is already here. D ynam ical alternatives (for learning) are
u n d er active exploration, and results directly relevant to the natu re of
scientific inquiry can already be discerned. It is the prim ary aim of
this volum e to bring those results to the attention of the philosophy-
of-science com m unity.
Though the results are un exp ected and to som e extent destructive
of orth od oxy, I exp ect them to be given a fair hearing by the relevant
com m unity, since a crucial p reced en t h as already been set. The rele
vance, to theories about the nature of science, of the actual history of
h um an scientific inquiry is som ething that h as been conceded by the
bulk of the com m unity for well over a d ecad e. On the w hole, this has
b een a healthy developm ent. The m ore em pirical constraints w e can
place on any theoretical enterprise, the better, and the philosophy of
science is no exception. The infusion I am recom m ending is m ore of
the sam e. But the empirical data are draw n not from history or sociol
ogy: they are draw n from the brain and from those artificial system s
that m odel the salient m icrostructural and m icrofunctional features of
the brain. M oreover, the data are accom panied by som e prom ising
new theory about the nature of cognition. Suitably developed, that
theory m ay help us find a new coheren ce in the scattered issues that
m ake up the philosophy of science, new solutions to old philosophi
cal problem s, and new possibilities for the enterprise of science itself.
T here is no guarantee that this will h ap p en , but the gathering case is
assem bled in p art 2 of this volum e.
A lm ost all of the essays here collected w ere w ritten as independent
p ieces, and alm ost all should be intelligible as such. C ertain strong
threads conn ect them , how ever, and certain sequences are salient.
Preface xvii
The quick tour through the m ajor them es of the volum e w ould focus
on ch apters 1, 5, and 6 from part 1, and chapters 9, 10 and 11 from
part 2.
The research here presen ted w as supported by a gran t from the
Institute for A d vanced Study, by a grant from the Social Sciences and
H um anities R esearch Council of C an ada, by a sabbatical leave from
the U niversity of M anitoba, and by the U niversity of California, San
D iego. Turning finally to individuals, I am pleased to be able to thank
L arry Jord an, David Zipser, Terry Sejnowski, David R um elhart, F ran
cis Crick, Rodolfo Llinas, A nd ras Pellionisz, Philip K itcher, Stephen
Stich, and Patricia C hurchland. W ithout their w ise and generou s
com p an y to draw on, I could never have w ritten the later p ap ers in
this volum e, and w ould never have been able to com plete the argu
m ent begun in the earlier papers. Thanks in large m easure to them , I
h ave a coh erent story to tell.
La Jolla, California, 1988

PA RT I
The Nature of Mind

Chapter 1
Eliminative Materialism and the
Propositional Attitudes
Elim inative m aterialism is the thesis that our co m m on sen se con cep
tion of psychological p h enom ena constitutes a radically false theory,
a theory so fundam entally defective that both the principles and the
ontology of that theory will eventually be displaced, rath er than
sm oothly red u ced , by com pleted neuroscience. O ur m utual u n d er
standing and even our introspection m ay then be reconstituted w ith
in the conceptual fram ew ork of com pleted neuroscience, a theory
we m ay exp ect to be m ore pow erful by far than the com m on sen se
p sych ology it displaces, and m ore substantially integrated w ithin
physical science generally. M y purpose in this p aper is to explore
these projections, especially as they bear on (1) the principal elem ents
of co m m onsense psychology: the propositional attitudes (beliefs,
desires, e tc.), and (2) the conception of rationality in w hich those
elem ents figure.
This focus rep resents a change in the fortunes of m aterialism .
Tw enty years ago em otions, qualia, and "ra w feels" w ere held to be
the principal stum bling blocks for the m aterialist program . W ith these
barriers dissolving (Feyerabend 1963a; Rorty 1965; C hurchland 1979),
the locus of opposition has shifted. N ow it is the realm of the inten
tional, the realm of the propositional attitude, that is m o st com m only
held up as being both irreducible to and ineliminable in favor of any
thing from w ithin a m aterialist fram ew ork. W h eth er and w hy this is
so, w e m u st exam ine.
Such an exam ination will make little sense, h ow ever, unless it is
first ap p reciated that the relevant netw ork of com m on sen se concepts
does indeed constitute an empirical theory, w ith all the functions,
virtues, and perils entailed by that status. I shall therefore begin w ith a
brief sketch of this view and a sum m ary rehearsal of its rationale. The
resistance it en cou nters still surprises m e. A fter all, com m on sense
has yielded up m any theories. Recall the view that space has a pre-
T h is p a p e r first ap p eared in th e Journ al o f P hilosophy 78 (1981), n o . 2.

2 The Nature of Mind
ferred direction in w hich all things fall, that w eight is an intrinsic

feature of a body, that a force-free m oving object will prom ptly return
to rest, that the sphere of the heavens turns daily, and so on. These
exam ples are clear, perh ap s, but people seem willing to concede a
theoretical com ponent w ithin com m on sense only if (1) the theory
and the com m on sense involved are safely located in antiquity, and
(2) the relevant theory is now so clearly false that its speculative
nature is inescapable. Theories are indeed easier to discern under
these circum stances. But the vision of hindsight is alw ays 20/20.
Let us aspire to som e foresight for a change.
1 W hy Folk Psychology Is a Theory
Seeing our com m onsense conceptual fram ew ork for m ental p h e

n om ena as a theory brings a simple and unifying organization to m ost
of the m ajor topics in the philosophy of m ind, including the explana
tion and prediction of behavior, the sem antics of m ental predicates,
action theory, the problem of oth er m inds, the intentionality of
m ental states, the nature of introspection, and the m ind-body
problem . A n y view that can pull this lot togeth er deserves careful
consideration.
L et us begin w ith the explanation of h um an (and anim al) behavior.
The fact is that the average person is able to explain, and even p re
dict, the behavior of p ersons w ith a facility and success that is rem ark
able. Such explanations and predictions standardly m ake reference to
the desires, beliefs, fears, intentions, percep tion s, and so forth, to
w hich the agents are presum ed subject. But explanations presu p p ose
law s— rough and ready ones, at least— that connect the explanatory
conditions w ith the behavior explained. The sam e is true for the m ak
ing of predictions, and for the justification of subjunctive and coun-
terfactual conditionals concerning behavior. Reassuringly, a rich
netw ork of com m onsense laws can indeed be recon stru cted from
this quotidian commerce of explanation and anticipation; its principles
are familiar hom ilies; and their sundry functions are tran sp aren t. W e
un d erstan d others, as well as w e do, b ecause w e share a tacit com
m and of an integrated body of lore concerning the lawlike relations
holding am ong external circum stances, internal states, and overt be
havior. Given its nature and functions, this body of lore m ay quite
aptly be called "folk p sy ch o lo g y ." (I shall exam ine a handful of these
law s presently. For a m ore com prehensive sam pling of the law s of
folk psychology, see P. M. Churchland 1979. For a detailed exam ina
tion of the law s that underw rite action explanations in particular, see
P. M. C hurchland 1970.)
Eliminative Materialism and the Propositional Attitudes 3
This ap p roach entails that the sem antics of the term s in our familiar
m entalistic vocabulary is to be u nderstood in the sam e m an n er as the
sem antics of theoretical term s generally: the m eaning of an y theoret
ical term is fixed or constituted by the netw ork of laws in w hich it
figures. (This position is quite distinct from logical behaviorism . I
deny that the relevant law s are analytic, and it is the lawlike con n ec
tions generally that carry the sem antic w eight, not just the con n ec
tions w ith overt behavior. But this view does accou n t for w hat little
plausibility logical behaviorism did enjoy.)
M ore im portant, the recognition that folk psychology is a theory
provides a simple and decisive solution to an old skeptical problem ,
the problem of other m inds. The problem atic conviction that another
individual is the subject of certain m ental states is not inferred de
ductively from his behavior, nor is it inferred by inductive analogy
from the perilously isolated instance of o n e's ow n case. R ather, that
conviction is a singular explanatory hypothesis of a perfectly straight
forw ard kind. Its function, in conjunction with the background
laws of folk psychology, is to provide explanations/predictions/
u n d erstan din g of the individual's continuing behavior, and it is credi
ble to the d egree that it is successful in this regard o ver com peting
h yp oth eses. In the m ain, such h ypotheses are successful, and so the
belief that others enjoy the internal states com p reh en d ed by folk
p sychology is a reasonable belief.
K now ledge of oth er m inds thus has no essential d ep en den ce on
know ledge of on e's own m ind. A pplying the principles of our folk
p sych ology to our behavior, a M artian could justly ascribe to us the
familiar run of m ental states, even though his ow n p sych ology w ere
very different from ours. H e w ould not, therefore, be "gen eralizin g
from his ow n c a s e ."
A s well, introspective judgm ents about o n e's ow n case turn out not
to h ave any special status or integrity anyw ay. O n the p resen t v iew , a
sp on tan eous introspective judgm en t is just an instance of an acquired
habit of conceptual respon se to one's internal states, and the integrity
of any particular response is alw ays contingent on the integrity of the
acquired conceptual fram ew ork (theory) in w hich the respon se is
fram ed. A ccordingly, on e's introspective certainty that o n e 's mind is
the seat of beliefs and desires m ay be as badly m isplaced as w as the
classical m an 's visual certainty that the star-flecked sphere of the
heaven s turns daily.
A n o th er con un drum is the intentionality of m ental states. The
"p rop ositional attitu d es," as Russell called them , form the system atic
core of folk psychology, and their uniqueness and anom alous logical
p roperties have inspired som e to see here a fundam ental contrast
w ith anything that m ere physical p hen om en a m ight conceivably dis

play. The key to this m atter lies again in the theoretical nature of folk
psychology. The intentionality of m ental states here em erges not as a
m ystery of n ature, but as a structural feature of the concepts of folk
psychology. Ironically, those sam e structural features reveal the very
close affinity that folk psychology bears to theories in the physical
sciences. Let m e try to explain.
C onsider the large variety of w hat m ight be called “ num erical atti
tu d e s" ap p earing in the conceptual fram ew orks of physical science:
. . has a masskg of n', . . has a velocity™^ of n', ‘. . . h as a tem
p era tureK of n', and so forth. These expressions are predicate-
form ing expressions: w hen one substitutes a singular term for a n u m
ber into the place held by 'и', a determ inate predicate results. M ore
interesting, the relations betw een the various "n u m erical attitu d es"
th at result are precisely the relations betw een the num bers "c o n
tain ed " in those attitudes. M ore interesting still, the argu m en t place
that takes the singular term s for num bers is open to quantification.
All this perm its the expression of generalizations concerning the law
like relations that hold betw een the various num erical attitudes in
nature. Such law s involve quantification over num bers, and they ex
ploit the m athem atical relations holding in that dom ain. Thus, for
exam ple,
(1) (x )(f)(m )[((x has a m ass of m) & (x suffers a net force o f /) )

з (x accelerates at f/m )].
C onsider now the large variety of propositional attitudes: '. . .b e

lieves that p', '. . . desires that p', . . fears that p', '. . . is happy
that p', etc. These expressions are predicate-form ing expressions
also. W h en one substitutes a singular term for a proposition into the
place held by ‘p ’, a determ inate predicate results, e .g ., . .believes
that Tom is tali'. (Sentences do not generally function as singular
term s, but it is difficult to escap e the idea that w hen a sentence occurs
in the place held by ‘p ’, it is there functioning as or like a singular
term . M ore on this below .) M ore interesting, the relations betw een
the resulting propositional attitudes are characteristically the relations
that hold betw een the propositions "co n ta in e d " in them , relations
such as entailm ent, equivalence, and m utual inconsistency. M ore in
teresting still, the argum ent place that takes the singular term s for
propositions is open to quantification. All this perm its the expression
of generalizations concerning the lawlike relations that hold am ong
propositional attitudes. Such laws involve quantification over p rop
ositions, and they exploit various relations holding in that dom ain.
Thus, for exam ple,
(2) (x)(p)[(x fears that p) (x desires that ~ p )]

(3) (x)(p )[((x h op es that p) & (x discovers that pj)
з (x is pleased that p)]
(4) (x)(p)(q)[((x believes that p) & (x believes that (if p then q)))
zd (barring confusion, distraction, etc., x believes that q)]
(5) (x)(p)(q)[{(x desires that p) & (x believes that (if q then pj)
& (x is able to bring it about that q))
з (barring conflicting desires or preferred m ean s,
x brings it about that q)].
(If w e stay w ithin an objectual interpretation of the quantifiers,

perh ap s the sim plest w ay to make system atic sense of expressions
like rx believes that p* and closed sentences form ed therefrom is just
to con stru e w h atever occurs in the nested position held by 'p', ‘q’,
etc. as there having the function of a singular term . A ccordingly,
w hen the standard connectives occur betw een term s in that nested
position, they m ust be construed as there functioning as operators
that form com p oun d singular term s from other singular term s, and
not as sen tence operators. The com pound singular term s so form ed
denote the appropriate com pound propositions. Substitutional quan
tification will of course underw rite a different interpretation, and
there are other approach es as well. Especially appealing is the pro-
sentential approach of G rover, C am p, and Belnap 1975. But the re
solution of these issues is not vital to the p resen t discussion.)
Finally, the realization that folk p sychology is a theory puts a new
light o n the m ind-body problem . The issue becom es a m atter of how
the ontology of one theory (folk psychology) is, or is not, going to be
related to the ontology of another theory (com pleted neuroscience);
and the m ajor philosophical positions on the m ind-body problem
em erge as so m any different anticipations of w hat future research will
reveal about the intertheoretic status and integrity of folk psychology.
The identity theorist optimistically expects that folk psychology will
be sm oothly reduced by com pleted neuroscience, and its ontology p re
served by dint of transtheoretic identities. The dualist exp ects that it
will prove z'rreducible to com pleted neuroscience, by dint of being a
non red u n d an t description of an au ton om ou s, nonphysical dom ain of
natural p h en om ena. The functionalist too expects that it will prove
irreducible, but on the quite different grou n d s that the internal econ
om y ch aracterized by folk p sychology is n ot, in the last analysis, a
law -governed econom y of natural states but an abstract organization
of functional states, an organization instantiable in a variety of quite
different m aterial substrates. It is therefore irreducible to the princi

ples peculiar to any one of them .
Finally, the eliminative m aterialist too is pessim istic about the
prospects for reduction, but his reason is that folk p sychology is a
radically inadequate account of our internal activities, too confused
and too defective to win survival throu gh intertheoretic reduction.
O n his view , it will simply be displaced by a better theory of those
activities.
W hich of these fates is the real destiny of folk p sychology, w e shall
attem pt to divine presently. For now the point to keep in m ind is that
w e shall be exploring the fate of a theory, a system atic, corrigible,
speculative theory.
2 W hy F olk Psychology M ight (Really) Be False
Since folk psychology is an em pirical theory, it is at least an abstract

possibility that its principles are radically false and that its ontology is
an illusion. W ith the exception of eliminative m aterialism , h ow ever,
n on e of the m ajor positions takes this possibility seriously. N on e of
them doubts the basic integrity or truth of folk psychology (FP), and
all of them anticipate a future in w hich its law s and categories are
conserved. This conservatism is not w ithout som e foundation. After
all, FP does enjoy a substantial am ount of explanatory and predictive
su ccess. A n d w hat better grou nd s for confidence in the integrity of its
categories?
W hat b etter groun d s indeed. Even so, the presum ption in favor of
FP is spurious, born of innocence and tunnel vision. A m ore search
ing exam ination reveals a different picture. First, w e m u st reckon
not only w ith the successes of FP but also w ith its explanatory failures
and w ith their extent and seriousness. Second, w e m ust consider the
long-term history of FP, its grow th , fertility, and current prom ise of
future developm ent. A nd third, w e m u st consider w hat sorts of
theories are likely to be true of the etiology of our behavior, given
w hat else w e have learned about ourselves in recent history. That is,
w e m ust evaluate FP w ith regard to its coh eren ce and continuity with
fertile and w ell-established theories in adjacent and overlapping
d om ain s— w ith evolutionary theory, biology, and neuroscience, for
exam p le— because active coherence w ith the rest of w hat w e p re
sum e to know is p erhaps the final m easure of any hypothesis.
A serious inventory of this sort reveals a very troubled situation,
one w hich w ould evoke open skepticism in the case of any theory less
familiar and d ear to us. Let m e sketch som e relevant detail. W h en one
cen ters on e's attention not on w hat FP can explain, but on w hat it
can n ot explain or fails even to address, one discovers that there is a

great deal. A s exam ples of central and im portant m ental p h en om en a
that rem ain largely or w holly m ysterious within the fram ew ork of FP,
consider the nature and dynam ics of m ental illness, the faculty of
creative im agination, or the ground of intelligence differences be
tw een individuals. C onsider our utter ignorance of the nature and
psychological functions of sleep, that curious state in w hich a third of
on e's life is spent. Reflect on the com m on ability to catch an outfield
fly ball on the run, or hit a m oving car w ith a snowball. C onsider the
internal construction of a three-dim ensional visual im age from subtle
differences in the tw o-dim ensional array of stim ulations in on e's re
spective retinas. C onsider the rich variety of p erceptual illusions,
visual and otherw ise. O r consider the m iracle of m em o ry , w ith its
lightning capacity for relevant retrieval. O n these and m an y other
m ental p h en om en a, FP sheds negligible light.
O ne particularly outstanding m ystery is the nature of the learning
p rocess itself, especially w here it involves large-scale conceptual
ch an ge, and especially in its prelinguistic or entirely nonlinguistic
form (as in infants and anim als), w hich is by far the m o st com m on
form in n atu re. FP is faced with special difficulties here, since its con
ception of learning as the m anipulation and storage of propositional
attitudes founders on the fact that how to form ulate, m anipulate, and
store a rich fabric of propositional attitudes is itself som ething that is
learned, and is only one am ong m any acquired cognitive skills. FP
w ould thus ap pear constitutionally incapable of even addressing this
m ost basic of m ysteries. (A possible respon se here is to insist that the
cognitive activity of anim als and infants is linguiform al in its ele
m en ts, structu res, and processing right from birth. J. A . Fod or [1975]
has erected a positive theory of thought on the assum ption that the
innate form s of cognitive activity have precisely the form h ere denied.
For a critique of Fod or's view , see P. S. C hurchland 1978.)
Failures on such a large scale do not yet show that F P is a false
theory, but they do m ove that p rospect well into the ran ge of real
possibility, and they do show decisively that FP is at best a highly
superficial theory, a partial and u npenetrating gloss on a deep er and
m ore com plex reality. H aving reached this opinion, w e m ay be for
given for exploring the possibility that FP provides a positively m is
leading sketch of our internal kinem atics and dynam ics, one w hose
su ccess is ow ed m ore to selective application and forced interp reta
tion o n o u r part than to genuine theoretical insight on F P 's part.
A look at the history of FP does little to allay such fears, once
raised. The story is one of retreat, infertility, and d ecad en ce. The p re
sum ed dom ain of FP used to be m uch larger than it is n ow . In prim i
tive cultures, the behavior of m ost of the elem ents of nature w ere
un derstood in intentional term s. The w ind could know anger, the
m oon jealousy, the river generosity, the sea fury, and so forth. These
w ere not m etaphors. Sacrifices w ere m ade and auguries undertaken
to placate or divine the changing passions of the gods. D espite its
sterility, this anim istic approach to nature has dom inated our history,
and it is only in the last tw o or three thou san d years that w e have
restricted F P 's literal interpretation to the dom ain of the higher
anim als.
Even in this preferred dom ain, h ow ever, both the con ten t and the
success of FP have not advanced sensibly in tw o or three thousand
years. The FP of the G reeks is essentially the FP w e use today, and w e
are negligibly better at explaining h um an behavior in its term s than
w as Sophocles. This is a very long period of stagnation and infertility
for any theory to display, especially w hen faced w ith such an en or
m ou s backlog of anom alies and m ysteries in its ow n explanatory d o
m ain. Perfect theories, perh ap s, have no need to evolve. But FP is
profoundly im perfect. Its failure to develop its resources and extend
its range of success is therefore darkly curious, and one m u st query
the integrity of its basic categories. To use Imre Lakotos's term s, FP
is a stagnan t or degenerating research p rogram and has been for
millennia.
Exp lan atory success to date is, of course, n ot the only dim ension in
w hich a theory can display virtue or p rom ise. A troubled or stagnant
theory m ay m erit patience and solicitude on oth er grounds, for exam
ple, on groun d s that it is the only theory or theoretical ap proach that
fits well w ith oth er theories about adjacent subject m atters, or the
only one that prom ises to reduce to, or to be explained by, som e
established background theory w hose dom ain encom p asses the do
m ain of the theory at issue. In sum , it m ay rate credence b ecause it
holds prom ise of theoretical integration. H ow does FP rate in this
dim ension?
It is just here, perh ap s, that FP fares p oorest of all. If w e approach
H om o sapiens from the perspective of natural history and the physical
sciences, w e can tell a coh erent story of the species' constitution, d e
velop m ent, and behavioral capacities that encom p asses particle p h y
sics, atom ic and m olecular theory, organic chem istry, evolutionary
theory, biology, physiology, and m aterialistic neuroscience. That
story, though still radically incom plete, is already extrem ely p ow er
ful, outperform ing FP at m any points even in its ow n dom ain. A nd it
is deliberately and self-consciously coh eren t w ith the rest of our d e
veloping w orld picture. In short, the greatest theoretical synthesis in
the history of the hum an race is currently in our hands, and parts of it
already provide searching descriptions and explanations of hum an

sensory input, neural activity, and m otor control.
But FP is no p art of this grow ing synthesis. Its intentional cate
gories stand m agnificently alone, w ithout visible p ro sp ect of red u c
tion to that larger corpus. A successful reduction cannot be ruled out,
in m y view , but the explanatory im potence and long stagnation of
FP inspire little faith that its categories will find them selves neatly
reflected in the fram ew ork of neuroscience. O n the con trary, one is
rem inded of how alchem y m u st have looked as elem ental chem istry
w as taking form , how Aristotelian cosm ology m ust h ave looked as
classical m echanics w as being articulated, or how the vitalist con cep
tion of life m u st have looked as organic chem istry m arch ed forw ard.
In sketching a fair sum m ary of this situation, w e m u st m ake a spe
cial effort to abstract from the fact that FP is a central part of our
cu rrent Lebensw elt, and serves as the principal vehicle of o u r interper
sonal com m erce. For these facts provide FP with a conceptual inertia
that goes far beyond its purely theoretical virtues. Restricting
ourselves to this latter dim ension, w hat w e m ust say is th at FP suffers
explanatory failures on an epic scale, that it has been stagnan t for at
least tw enty-five centuries, and that its categories appear (so far) to be
incom m ensurable w ith, or orthogonal to, the categories of the back
ground physical science w hose long-term claim to explain h um an be
havior seem s undeniable. A ny theory that m eets this description
m ust be allowed a serious candidate for outright elim ination.
W e can , of course, insist on no stronger conclusion at this stage.
N or is it m y con cern to do so. W e are here exploring a possibility, and
the facts dem and no m ore, and no less, than that it be taken serious
ly. The distinguishing feature of the eliminative m aterialist is that he
takes it very seriously indeed.
3 A rgum ents against E lim ination
Thus the basic rationale of eliminative m aterialism : FP is a theory,

and quite probably a false one; let us attem pt, therefore, to transcend
it.
The rationale is clear and simple, but m any find it uncom pelling. It
will be objected that FP is not, strictly speaking, an em pirical theory;
that it is not false, or at least not refutable by empirical considerations;
and th at it ought not or cannot be transcended in the fashion of a
defunct physical theory. In w hat follows I shall exam ine these objec
tions as they flow from the m ost popular and best founded of the
com peting positions in the philosophy of m ind: functionalism .
An antipathy tow ard eliminative m aterialism arises from tw o dis
tinct threads running through contem porary functionalism . The first

thread co n cern s the norm ative character of FP, or at least of that cen
tral core of FP that treats of the propositional attitudes. FP, som e
will say, is a characterization of an ideal, or at least a praisew orthy,
m ode of internal activity. It outlines n o t only w hat it is to have and
process beliefs and desires, but also (and inevitably) w hat it is to be
rational in their adm inistration. The ideal laid dow n by FP m ay be
im perfectly achieved by empirical h u m an s, but this does not im pugn
FP as a norm ative characterization. N or need such failures seriously
im pugn FP even as a descriptive characterization, for it rem ains true
that our activities can be both usefully and accurately u n d erstood as
rational except for the occasional lapse due to noise, interference, or
other breakdow n, w hich defects em pirical research m ay eventually
unravel. A ccordingly, though neuroscience m ay usefully au gm ent it,
FP has no pressing need to be displaced, even as a descriptive theory;
nor could it be replaced, qua norm ative characterization, by any de
scriptive theory of neural m echanism s, since rationality is defined over
propositional attitudes like beliefs and desires. FP, therefore, is here
to stay.
Daniel D ennett has defended a view along these lines. (He d e
fended it m ost explicitly in 1981, but this them e of D ennett's goes all
the w ay back to his 1971.) A n d the view just outlined gives voice to
a them e of the p roperty dualists as well. Karl Popper and Joseph
M argolis both cite the norm ative natu re of m ental and linguistic
activity as a bar to their p enetration b y any descriptive/m aterialist
theory (Popper 1972; P op p er and Eccles 1978; M argolis 1978). I hope
to deflate the appeal of such m oves below .
The secon d thread con cern s the abstract nature of FP. The central
claim of functionalism is that the principles of FP characterize our
internal states in a fashion that m akes no reference to their intrinsic
nature or physical constitution. R ather, they are characterized in
term s of the netw ork of causal relations they bear to one another
and to sen sory circum stances and overt behavior. Given its abstract
specification, that internal econom y m ay therefore be realized in a
nom ically h eterogen eou s variety of physical system s. All of them
m ay differ, even radically, in their physical constitution, and yet at
anoth er level they will all share the sam e nature. This view , says
Fod or, "is compatible w ith v ery stron g claims about the inelimina-
bility of m ental language from behavioral th eories" (1968, p. 116).
Given the real possibility of m ultiple instantiations in heterogen eou s
physical substrates, w e can not elim inate the functional ch aracter
ization in favor of any theory peculiar to one such substrate. That
w ould preclude our being able to describe the (abstract) organization

that any one instantiation shares with all the others. A functional
characterization of our internal states is therefore here to stay.
This second them e, like the first, assigns a faintly stipulative ch ar
acter to F P , as if the onus w ere on the em pirical system s to instantiate
faithfully the functional organization that FP specifies, instead of the
onus being on FP to describe faithfully the internal activities of a
naturally distinct class of empirical system s. This im pression is
en h anced by the standard exam ples used to illustrate the claim s of
functionalism : m ousetrap s, valve lifters, arithm etical calculators,
com puters, robots, and the like. These are artifacts, con stru cted to
fill a p reconceived bill. In such cases, a failure of fit betw een the p h y
sical sy stem and the relevant functional characterization im pugns
only the form er, not the latter. The functional characterization is thus
rem oved from em pirical criticism in a w ay that is m ost unlike the
case of an em pirical theory. O ne prom inent functionalist, H ilary P ut
nam , h as argued outright that FP is not a corrigible em pirical theory
at all (Pu tn am 1964, pp. 675, 681ff). Plainly, if FP is con stru ed on
these m odels, as regularly it is, the question of its em pirical integrity
is unlikely ever to pose itself, let alone receive a critical answ er.
A lthough fair to som e functionalists, the preceding is n ot entirely
fair to Fod or. O n his view the aim of psychology is to find the best
functional characterization of ourselves, and w hat that is rem ains an
empirical question. Also, his argum ent for the ineliminability of m en
tal vocabulary from psychology does not pick out cu rren t FP in p ar
ticular as ineliminable. It need claim only that som e abstract functional
characterization m u st be retained, som e articulation or refinem ent of
FP perh ap s.
His estim ate of eliminative m aterialism rem ains low , how ever.
First, F od or plainly thinks that there is nothing fundam entally or in
terestingly w ron g w ith FP. O n the contrary, F P 's central conception
of cognitive activity— as consisting in the m anipulation of proposi
tional attitu d es— turns up as the central elem ent in F o d o r's ow n
theory of the nature of thought (1975). A nd second, there rem ains
the point that, w hatever tidying up FP m ay or m ay n ot require, it
cannot be displaced by any naturalistic theory of our physical sub
strate, since it is the abstract functional features of the internal states
that m ake a p erson, not the chem istry of his substrate.
All of this is appealing. But alm ost none of it, I think, is right. Fu n c
tionalism h as too long enjoyed its reputation as a daring and avant-
garde position. It needs to be revealed for the shortsighted and re
actionary position it is.
4 The C onservative N ature o f Functionalism
A valuable perspective on functionalism can be gained from the fol

low ing story. To begin w ith, recall the alchem ists' theory of inanim ate
m atter. W e have here a long and variegated tradition, of course, not a
single theory, but our purp oses will be served by a gloss.
The alchem ists conceived the "in a n im a te " as entirely continuous
w ith anim ated m atter in that the sensible and behavioral properties of
the various substances are due to the en soulm ent of b aser m atter by
various spirits or essen ces. These nonm aterial aspects w ere held to
u n dergo developm ent, just as w e find grow th and d evelopm ent in
the various souls of plants, animals, and hum ans. The alchemist's pecu
liar skill lay in know ing how to seed, nourish, and bring to m aturity
the desired spirits en m attered in the appropriate com binations.
O n one orth od oxy, the four fundam ental spirits (for "in an im ate"
m atter) w ere nam ed " m e rc u ry ," "s u lfu r," "yellow a rse n ic," and "sal
am m o n iac." Each of these spirits w as held responsible for a rough
but characteristic syndrom e of sensible, com binatorial, and causal
properties. The spirit m ercury, for exam ple, w as held responsible
for certain features typical of m etallic substances: their shininess,
liquefiability, and so forth. Sulfur w as held responsible for certain
residual features typical of m etals, and for those displayed by the
ores from w hich running m etal could be distilled. A ny given m etallic
substance w as a critical orchestration principally of these tw o spirits.
A similar story held for the oth er tw o spirits, and am ong the four of
them a certain dom ain of physical features and transform ations w as
rendered intelligible and controllable.
The degree of control w as alw ays limited, of course. O r b etter, such
prediction and control as the alchem ist possessed w as ow ed m ore to
the m anipulative lore acquired as an apprentice to a m aster than to
any genuine insight supplied by the theory. The theory followed,
m ore than it dictated, practice. But the theory did supply som e rhym e
to the practice, and in the absence of a developed alternative it w as
sufficiently com pelling to sustain a long and stubborn tradition.
The tradition had becom e faded and fragm ented by the tim e the
elem ental chem istry of Lavoisier and D alton arose to replace it for
good. But let us suppose that it had h u n g on a little longer— p erhaps
because the four-spirit orth od oxy had becom e a thum b-w orn p art of
ev erym an 's com m on sen se— and let us exam ine the nature of the
conflict betw een the tw o theories and som e possible avenu es of
resolution.
N o doubt the sim plest line of resolution, and the one that histor
ically took place, is outright displacem ent. The dualistic interp reta
tion of the four essen ces— as im m aterial spirits— will ap p ear both
feckless and u nn ecessary given the p ow er of the corpuscularian tax
onom y of atom ic chem istry. A nd a reduction of the old taxon om y to
the new will appear impossible, given the extent to w hich the com
paratively toothless old theory cross-classifies things relative to the
new . Elim ination w ould thus appear the only alternative— unless
som e cunning and determ ined defender of the alchem ical vision had
the w it to suggest the following defense.
Being "en sou led by m ercu ry ," or "s u lfu r," or either of the other
two so-called spirits, is actually afu n ctio n a l state. The first, for exam
ple, is defined by the disposition to reflect light, to liquefy u n d er heat,
to unite w ith oth er m atter in the sam e state, and so forth. A n d each of
these four states is related to the others, in that the syndrom e for each
varies as a function of w hich of the other three states is also instanti
ated by the sam e substrate. Thus the level of description com
p reh ended by the alchemical vocabulary is abstract: various m aterial
substances, suitably "e n s o u le d ," can display the features of a m etal,
for exam ple, or even of gold specifically. For it is the total syndrom e
of occu rren t and causal properties that m atters, not the corpuscula
rian details of the substrate. Alchemy, it is concluded, com prehends
a level of organization in reality that is distinct from , and irreducible
to, the organization found at the level of corpuscularian chem istry.
This view m ight have had considerable appeal. After all, it spares
alchem ists the burden of defending im m aterial souls th at com e and
go; it frees them from having to m eet the v ery strong d em an d s of a
naturalistic reduction; and it spares them the shock and confusion of
outright elim ination. A lchem ical theory em erges as basically all right!
N or need the alchem ists appear too obviously stubborn or dogm atic
in this. A lchem y as it stands, they con cede, m ay need substantial
tidying up , and experience m ust be our guide. But w e n eed not fear
its naturalistic displacem ent, they rem ind us, since it is the peculiar
orchestration of the syndrom es of occu rrent and causal properties
that m akes a piece of m atter gold, not the idiosyncratic details of its
corpuscularian substrate. A further circum stance w ould have m ade
this claim even m ore plausible. For the fact is, the alchem ists did
know h ow to m ake gold, in this relevantly w eakened sense of 'gold',
and th ey could do so in a variety of w ays. Their "g o ld " w as never as
perfect, alas, as the "g o ld " n u rtured in n atu re's w om b , but w hat
m ortal can exp ect to m atch the skills of nature herself?
W hat this story show s is that it is at least possible for the constella
tion of m oves, claims, and defenses characteristic of functionalism to
constitute an outrage against reason and tru th , and to do so w ith a
plausibility that is frightening. A lchem y is a terrible theory, well de
serving of its com plete elimination, and the defense of it just explored
is reactionary, obfuscatory, retrograd e, and w rong. But in historical
con text, that defense m ight have seem ed wholly sensible, even to
reasonable people.
The alchem ic exam ple is a deliberately transparent case of w hat
m ight be called "th e functionalist stra ta g e m ," and other cases are
easy to im agine. A cracking good defense of the phlogiston theory of
com bustion can also be constructed along these lines. C on stru e being
highly phlogisticated and being dephlogisticated as functional states
defined by certain syndrom es of causal dispositions; point to the
great variety of natural substrates capable of com bustion and calx-
ification; claim an irreducible functional integrity for w hat has proved
to lack any natural integrity; and bury the rem aining defects u n d er a
pledge to contrive im provem ents. A similar recipe will provide new
life for the four hum ors of m edieval m edicine, for the archeus or vital
essen ce of prem od ern biology, and so forth.
If its application in these other cases is any guide, the functionalist
stratagem is a sm oke screen for the preservation of error and confu
sion. W h en ce derives our assurance that in con tem porary journals
the sam e charade is n ot being played out on behalf of FP? The parallel
w ith the case of alchem y is in all other respects distressingly com
plete, right dow n to the parallel betw een the search for artificial gold
and the search for artificial intelligence!
Let m e not be m isunderstood on this last point. Both aim s are
w orth y aim s: thanks to nuclear physics, artificial (but real) gold is
finally w ithin our m ean s, if only in subm icroscopic quantities, and
artificial (but real) intelligence eventually will be. But just as the care
ful orchestration of superficial syndrom es w as the w ron g w ay to p ro
duce genuine gold, so m ay the careful orchestration of superficial
syndrom es be the w ron g w ay to produ ce genuine intelligence. Ju st as
w ith gold, w hat m ay be required is that our science p enetrate to the
underlying natural kind that gives rise to the total syndrom e directly.
(See chapters 5 and 9 to 11.)
In sum m ary, w hen confronted w ith the explanatory im potence,
stagnant history, and system atic isolation of the intentional idiom s of
FP, it is not an adequate or responsive defense to insist that those
idioms are abstract, functional, and irreducible in character. For one
thing, this sam e defense could have b een m ou nted with com parable
plausibility no m atter w hat hayw ire netw ork of internal states our
folklore had ascribed to us. A nd for an oth er, the defense assum es
essentially w hat is at issue: it assum es that it is the intentional idioms
of FP, plus or m inus a bit, that express the im portant features shared
by all cognitive system s. But they m ay not. Certainly it is w ron g to
assum e that they do, and then argue against the possibility of a
m aterialistic displacem ent on grounds that it m u st describe m atters at
a level that is distinct from the im portant level. This just begs the
question in favor of the older fram ew ork.
Finally, it is very im p ortant to point out that eliminative m aterial
ism is strictly consistent with the claim that the essence of a cognitive
system resides in the abstract functional organization of its internal
states. The eliminative m aterialist is not com m itted to the idea that
the correct accou n t of cognition m ust be a naturalistic accou n t, though
he m ay be forgiven for exploring the possibility. W hat h e does hold
is that the correct accoun t of cognition, w heth er functionalistic or
naturalistic, will bear about as m uch resem blance to FP as m odern
chem istry bears to four-spirit alchem y.
Let us now try to deal with the argu m en t, against eliminative
m aterialism , from the norm ative dim ension of FP. This can be dealt
w ith rath er swiftly, I believe.
First, the fact that the regularities ascribed by the intentional core of
FP are predicated on certain logical regularities am ong propositions is
not by itself grounds for claiming anything essentially norm ative
about F P. To draw a relevant parallel, the fact that the regularities
ascribed by the classical gas law are predicated on arithm etical rela
tions b etw een num bers does not imply anything essentially n or
m ative about the classical gas law. A nd logical relations betw een
propositions are as m uch an objective m atter of abstract fact as are
arithm etical relations betw een num bers. In this respect, the law
(4) {x){p)(q)[((x believes that p) & (x believes that (if p then q)))
=3 (barring confusion, distraction, e tc., x believes that q)]
is entirely on a p ar with the classical gas law
(6) (x)(P)(V )(/x)[((x has a pressure P)

& (x has a volum e V ) & (x has a quantity ju))
(barring very high pressure or density,
x has a tem peratu re of PV /ц К )].
A norm ative dim ension enters only because w e happen to value m ost
of the p attern s ascribed by FP. But w e do not value all of them .
Consider
(7) (x)(p)[((x desires w ith all his heart that p)

& (x learns that ~p ))
(barring unusual strength of character,
x is shattered that ~ p )].
M oreover, and as w ith norm ative convictions generally, fresh insight

m ay m otivate m ajor changes in w hat w e value.
Second, the law s of FP ascribe to us only a very minimal and tru n
cated rationality, not an ideal rationality as som e have suggested. The
rationality characterized by the set of all FP laws falls well short of an
ideal rationality. This is not surprising. W e have no clear or finished
conception of ideal rationality an yw ay; certainly the ordinary m an
does not. A ccordingly, it is just not plausible to suppose that the
explanatory failures from w hich FP suffers are due prim arily to
hu m an failure to live up to the standards it p rovides. Quite to the co n
trary, the conception of rationality it provides appears limping and
superficial, especially w hen com pared w ith the dialectical com plexity
of our scientific history or w ith the ratiocinative virtuosity displayed
by any child.
Third, even if our current conception of rationality— and m ore
generally, of cognitive virtu e— is largely constituted w ithin the
sentential/propositional fram ew ork of FP, there is no guarantee that
this fram ew ork is adequate to the deep er and m ore accurate account
of cognitive virtue that is clearly n eed ed . Even if w e concede the
categorial integrity of FP, at least as applied to language-using hu
m an s, it rem ains far from clear that the basic p aram eters of intellec
tual virtue are to be found at the categorial level com p reh en d ed by
the propositional attitudes. After all, language use is som ething that
is learned, by a brain already capable of vigorous cognitive activity;
language use is acquired as only one am ong a great variety of learned
m anipulative skills; and it is m astered by a brain that evolution has
shaped for a great m any functions, language use being only the very
latest and perhap s the least of them . A gainst the background of these
facts, language use appears as an extrem ely peripheral activity, as a
biologically idiosyncratic m ode of social interaction that is m astered
thanks to the versatility and pow er of a m ore basic m ode of activity.
W h y accept, then, a theory of cognitive activity that m odels its ele
m en ts on the elem ents of h um an language? A nd w hy assum e that the
fundam ental p aram eters of intellectual virtue are, or can be defined
over, the elem ents at this superficial level?
A serious advance in our appreciation of cognitive virtue w ould
thu s seem to require that w e go beyond FP, that w e transcend the
p overty of F P 's conception of rationality by transcending its proposi
tional kinem atics entirely, by developing a deeper and m ore general
kinem atics of cognitive activity, and b y distinguishing w ithin this
new fram ew ork w hich of the kinematically possible m odes of activity
are to be valued and encouraged (as m ore efficient, reliable, p ro d u c
tive, or w h atever). Eliminative m aterialism does not im ply the end of
our norm ative concerns. It implies only that they will h ave to be re
constituted at a m ore revealing level of u n d erstanding, the level that a
m atured neuroscience will provide. [A dded in 1989: The beginnings
of such a recon struction can be found in chapter 10, pp. 2 2 0 -2 2 3 .]
W h at a theoretically inform ed future m ight hold in store for us, w e
shall n ow turn to explore. N ot because w e can foresee m atters with
any special clarity, but b ecause it is im portant to try to break the grip
on our im agination held by the propositional kinem atics of FP. A s far
as the p resen t section is co n cern ed , w e m ay sum m arize our conclu
sion as follows. FP is nothing m ore and nothing less than a culturally
entrenched theory of h ow w e and the higher anim als w ork. It has no
special features that m ake it empirically invulnerable, no unique func
tions th at m ake it irreplaceable, no special status of any kind w hat
soever. W e shall turn a skeptical ear then, to any special pleading on
its behalf.
5 Beyond F olk P sychology
W hat m igh t the elim ination of FP actually involve: not just the com
paratively straightforw ard idiom s for sensation, but the entire ap p ar
atus of propositional attitudes? That d ep ends heavily on w hat
neuroscience m ight d iscover and on our determ ination to capitalize
on it. H ere follow three scenarios in w hich the operative conception
of cognitive activity is progressively divorced from the form s and
categories that characterize natural language. If the read er will in
dulge the lack of actual substance, I shall try to sketch som e plausible
form.
First, suppose that research into the structure and activity of the
brain, b oth fine-grained and global, finally does yield a new kinem a
tics and correlative dynam ics for w hat is now thought of as cognitive
activity. The theory is uniform for all terrestrial brains, n ot just hu
m an brains, and it m akes suitable conceptual contact w ith both evolu
tionary biology and nonequilibrium therm odynam ics. It ascribes to us
at any given time a set or configuration of com plex states that are
specified w ithin the theory as figurative "so lid s" w ithin a four- or
five-dim ensional p hase space. [Added in 1989: This guess has proved
to be v ery timid. The relevant cognitive statespaces typically have
hundreds, thou san ds, or even millions of distinct dim ensions, and
their partitioning into hypersolids is correspondingly com plex. See
chapter 9.] The law s of the theory govern the interaction, m otion, and
transform ation of these "so lid " states w ithin that space, and also
their relations to w h atever sensory and m otor transducers the system
possesses. A s w ith celestial m echanics, the exact specification of the
“ solids" involved and the exhaustive accounting of all dynam ically

relevant adjacent solids is not practically possible, for m an y rea
sons, but here too it turns out that the obvious approxim ations w e
fall back on yield excellent explanations/predictions of internal
change and external behavior, at least in the short term . A s for long
term activity, the theory provides pow erful and unified accou n ts of
the learning process, the nature of m ental illness, and variations in
character and intelligence across the anim al kingdom as well as across
individual hu m an s.
M oreover, it provides a straightforw ard account of "k n o w le d g e ,"
as traditionally conceived. A ccording to the new theory, any declara
tive sentence to w hich a speaker w ould give confident assent is m ere
ly a one-dim ensional projection — through the com pound lens of
W ernicke's and Broca's areas onto the idiosyncratic surface of the
speaker's language— a one dimensional projection of a four- or five
dim ensional solid that is an elem ent in his true kinematical state. (Re
call the shadow s on the wall of Plato's cav e.) Being projections of that
inner reality, such sentences do carry significant inform ation reg ard
ing it and are thus fit to function as elem ents in a com m unication
system . O n the other hand, being swbdimensional projections, they
reflect but a narrow part of the reality projected. They are therefore
unfit to rep resent the deeper reality in all its kinematically, dynam ical
ly, and even norm atively relevant asp ects. That is to say, a system of
propositional attitudes, such as FP, m u st inevitably fail to capture
w hat is going on here, though it m ay reflect just enough superficial
structure to sustain an alchemylike tradition am ong folk w ho lack a
better theory. From the perspective of the n ew er theory, h ow ever, it
is plain that there simply are no law -governed states of the kind FP
postulates. The real laws governing our internal activities are defined
over different and m uch m ore com plex states and configurations, as
are the norm ative criteria for developm ental integrity and intellectual
virtue.
A theoretical outcom e of the kind just described m ay fairly be
counted as a case of the elim ination of one theoretical ontology in
favor of an other, but the success here im agined for system atic
neuroscience need not have any sensible effect on com m on practice.
Old w ays die hard , and in the absence of som e practical necessity,
they m ay n ot die at all. Even so, it is not inconceivable that som e
segm en t of the population, or all of it, should becom e intim ately
familiar w ith the vocabulary required to characterize our kinematical
states, learn the law s governing their interactions and behavioral p ro
jections, acquire a facility in their first-person ascription, and displace
the u se of FP altogether, even in the m arketplace. The dem ise of FP's

ontology w ould then be com plete.
W e m ay now explore a second and rath er m ore radical possibility.
E veryone is familiar w ith C h om sk y's thesis that the h u m an mind or
brain contains innately and uniquely the abstract structures for learn
ing and using specifically h u m an natural languages. A com peting
hypothesis is that our brain does indeed contain innate structures,
but that those structures have as their original and still prim ary func
tion the organization of perceptual experience, w ith the adm inistra
tion of linguistic categories being an acquired and additional function
for w hich evolution has only incidentally suited them . (Richard G reg
ory h as defended such a view in his 1970b.) This hypothesis has the
advantage of not requiring the evolutionary saltation that C hom sky's
view w ould seem to require, and there are other ad van tages as well.
But these m atters need n ot concern us here. Suppose, for our p u r
poses, th at this com peting view is true, and consider the following
story.
R esearch into the neural structures that fund the organization and
processing of perceptual inform ation reveals that they are capable of
adm inistering a great variety of com plex tasks, som e of them show ing
a com plexity far in excess of that show n by natural language. N atural
languages, it turns out, exploit only a very elem entary portion of the
available m achinery, the bulk of w hich serves far m ore com plex
activities beyond the ken of the propositional conceptions of FP. The
detailed unraveling of w hat that m achinery is and of the capacities it
has m akes it plain that a form of language far m ore sophisticated
than "n a tu ra l" language, though decidedly alien in its syntactic and
sem antic structures, could also be learned and used by our innate
system s. Such a novel system of com m unication, it is quickly real
ized, could raise the efficiency of inform ation exch an ge betw een
brains by an order of m agnitude, and w ould en h an ce epistem ic
evaluation by a com parable am ount, since it w ould reflect the u n d er
lying stru ctu re of our cognitive activities in greater detail than does
natural language.
G uided by our new understanding of these internal stru ctu res, w e
m anage to co n stru ct a new system of verbal com m unication entirely
distinct from natural language, w ith a new and m ore pow erful com
binatorial g ram m ar over novel elem ents form ing novel com binations
with exotic properties. The com pounded strings of this alternative
system — call them “ U bersatze" — are n ot evaluated as true or false,
nor are the relations betw een them rem otely analogous to the rela
tions of entailm ent, etc., that hold betw een sentences. They display a
different organization and m anifest different virtues.
O nce co n stru cted , this "lan g u ag e" proves to be learnable, it has the
p ow er projected, and in tw o generations it has sw ept the planet.
E veryone u ses the new system . The syntactic form s and sem antic
categories of so-called "n a tu ra l" language disappear entirely. A nd
w ith them disappear the propositional attitudes of FP, displaced by a
m ore revealing schem e in w hich (of course) " U bersatzenal attitu d es"
play the leading role. FP again suffers elim ination.
This second story, note, illustrates a them e w ith endless variations.
There are possible as m any different "folk psychologies" as there are
possible differently structured com m unication system s to serve as
m odels for them .
A third and even stranger possibility can be outlined as follows. W e
know that there is considerable lateralization of function betw een the
tw o cerebral hem ispheres, and that the tw o hem ispheres m ake use of
the inform ation they get from each other by w ay of the great cerebral
com m issure, the corpus callosum , a giant cable of neurons co n n ect
ing them . Patients w hose com m issure has been surgically severed
display a variety of behavioral deficits that indicate a loss of access by
one hem isphere to inform ation it used to get from the other. H ow
ever, in people with callosal agenesis (a congenital defect in which the
corresponding cable is simply absent), there is little or no behavioral
deficit, w hich suggests that the tw o hem ispheres have learned to ex
ploit the inform ation carried in other, less direct pathw ays connecting
them through the subcortical regions. This suggests that even in the
norm al case a developing hem isphere learns to m ake use of the in
form ation the cerebral com m issure deposits at its doorstep. W h at w e
have, then , in the case of a norm al hu m an , is tw o physically distinct
cognitive system s (both capable of ind ep en d en t function) responding
in a system atic and learned fashion to exchanged inform ation. A nd
w hat is especially interesting about this case is the sheer am ou nt of
inform ation exch an ged . The cable of the com m issure consists of
roughly 200 million neurons (G azzaniga and LeD oux 1975), and even
if w e assum e that each of these fibers is capable of one of only tw o
possible states each second (a m ost conservative estim ate), w e are
looking at a channel w hose inform ation capacity is g reater than
2 x 108 binary bits per second. C om p are this to the less than 500
bits/second capacity of spoken English.
N ow , if tw o distinct hem ispheres can learn to com m unicate on so
im pressive a scale, w hy shouldn't tw o distinct brains learn to do it
also? This w ould require an artificial "co m m issu re" of som e kind, but
let us suppose that w e can fashion a workable transducer for im
plantation at som e site in the brain that research reveals to be suit
able, a tran sd u cer to convert a sym phony of neural activity into (say)
m icrow aves radiated from an aerial in the forehead, and to perform

the reverse function of converting received m icrow aves back into
neural activation. C onnecting it up need not be an insuperable prob
lem. W e simply trick the norm al processes of dendritic arborization
into grow ing their ow n m yriad connections with the active m icro
surface of the transducer.
O nce the channel is opened betw een tw o or m ore p eople, they can
learn (learn) to exchange inform ation and coordinate their behavior
with the sam e intim acy and virtuosity displayed by y ou r ow n cerebral
hem isp heres. Think w hat this m ight do for hockey team s, and ballet
com panies, and research team s! If the entire population w ere thus
fitted out, spoken language of any kind m ight well disappear com
pletely, a victim of the "W h y crawl w hen you can fly?" principle.
Libraries becom e filled not w ith books, but w ith long recordings of
exem plary bouts of neural activity. These constitute a grow ing cultu
ral heritage, an evolving "T hird W o rld ," to use Karl P o p p er's term .
But th ey do not consist of sentences or argum ents.
H ow will such people understand and conceive of oth er indi
viduals? To this question I can only answ er, "In roughly the sam e
fashion that your right hem isphere 'un d erstan d s' and 'conceives of'
your left h em isphere: intim ately and efficiently, but n ot proposi-
tionally!"
These speculations, I hop e, will evoke the required sen se of u n
tapped possibilities, and I shall in any case bring th em to a close
here. Their function is to m ake som e inroads into the aura of incon
ceivability that com m only surrounds the idea that w e m ight reject FP.
The felt conceptual strain even finds expression in an argu m en t to the
effect that the thesis of eliminative m aterialism is incoh eren t since it
denies the very conditions p resupposed by the assu m p tion that it is
m eaningful. I shall close w ith a brief discussion of this v ery popular
m ove.
A s I have received it, the reductio proceed s by pointing out that the
statem en t of eliminative m aterialism is just a m eaningless string of
m arks or noises, unless that string is the expression of a certain belief,
and a certain intention to com m unicate, and a know ledge of the gram
m ar of the language, and so forth. But if the statem ent of eliminative
m aterialism is true, then there are no such states to exp ress. The
statem en t at issue w ould then be a m eaningless string of m arks or
noises. It w ould therefore not be true. Therefore, it is not tru e. Q .E .D .
The difficulty w ith any nonform al reductio is that the conclusion
against the initial assum ption is alw ays no better than the m aterial
assum ptions invoked to reach the incoherent conclusion. In this case
the additional assum ptions involve a certain theory of m eaning, one
that p resu p p oses the integrity of FP. But formally speaking, one can
as well infer, from the incoherent result, that this (Gricean) theory of
m eaning is w hat m ust be rejected. Given the independent critique of
FP leveled earlier, this w ould even seem the preferred option. But in
any case, one cannot simply assum e this particular theory of m eaning
w ithout begging the question at issue, nam ely, the integrity of FP.
The question-begging nature of this m ove is m ost graphically illus
trated by the following analog, w hich I ow e to Patricia S. C hurchland
(1981). The issue here, placed in the seven teen th century, is w heth er
there exists such a substance as vital spirit. A t the tim e, this substance
w as held, w ithout significant aw aren ess of real alternatives, to be
w hat distinguished the anim ate from the inanim ate. Given the
m onopoly enjoyed by this conception, given the degree to w hich it
w as integrated w ith m any of our other conceptions, and given the
m agnitude of the revisions any serious alternative conception w ould
require, the following refutation of any antivitalist claim w ould be
found instantly plausible.
The antivitalist says that there is no such thing as vital spirit. But
this claim is self-refuting. The speaker can expect to be taken
seriously only if his claim can n ot. For if the claim is true, then the
speaker does not have vital spirit and m u st be dead. But if he is
dead, then his statem ent is a m eaningless string of noises, de
void of reason and truth.
The question-begging nature of this argu m en t does not, I assum e,

require elaboration. To those m oved by the earlier argu m en t, I com
m end the parallel for exam ination.
The thesis of this paper m ay be sum m arized as follows. The pro-
positional attitudes of folk psychology do not constitute an unbreach-
able barrier to the advancing tide of neuroscience. On the contrary,
the principled displacem ent of folk psychology is not only richly
possible; it rep resents one of the m ost intriguing theoretical displace
m en ts w e can currently im agine.
Chapter 2
Functionalism, Qualia, and Intentionality
Functionalism — con strued broadly as the thesis that the essence of

our psychological states resides in the abstract causal roles they play
in a com plex econ om y of internal states m ediating environm ental in
puts and behavioral ou tp u ts— seem s to us to be free from any fatal or
essential shortcom ings. Functionalism -on-the-hoof is an oth er m atter.
In various thinkers this core thesis is generally embellished w ith cer
tain riders, interpretations, and m ethodological lessons d raw n there
from . W ith som e of the m ore prom inent of these articulations w e are
in som e d isagreem ent, and w e shall turn to discuss them in the final
section of this p ap er. O ur prim ary con cern , h ow ever, is to defend
functionalism from a b attery of better-know n objections widely be
lieved to pose serious or insurm ountable problem s even for the core
thesis outlined above. In sections 1 and 2 w e shall try to outline w hat
form functionalism should take in ord er to escap e those objections.
1 F ou r P roblem s concerning Q ualia
'Q ualia' is a p hilosopher's term of art denoting those intrinsic or

m onadic properties of our sensations discrim inated in introspection.
The quale of a sensation is typically contrasted w ith its causal,
relational, or functional features, and herein lies a problem for func
tionalism . The quale of a given sensation— pain, say— is at best con
tingently connected w ith the causal or functional properties of that
state, and yet com m on intuitions insist that this quale is an essential
elem ent of pain, on som e view s, the essential elem ent. Functionalism ,
it is con clu d ed, provides an inadequate accou n t of our m ental states.
Before addressing the issues in g reater detail, let us be clear about
w hat th e functionalist need not deny. H e need not and should not
deny th at our sensations have intrinsic properties, and he should
This paper was coauthored w ith Patricia S. Churchland. It first appeared in

Philosophical Topics (1981), no. 1.
agree that those properties are the principal m eans of our introsp ec
tive discrim ination of one kind of sensation from another. W h at he is
com m itted to denying is that any particular quale is essential to the
identity of any particular type of m ental state. Initially they m ay seem
to be essential, but reflection will reveal that they do not h ave, and
should not be conceded, that status. In w hat follows w e address four
distinct but not unrelated problem s. Each problem is m anageable on
its ow n, but if they are perm itted to band togeth er for collective
assault, the result is rather confusing and form idable, in the fashion
of the fabled M usicians of Brem en. W ith the problem s sep arated , our
strategy will be to explain and exploit the insight that intrinsic p ro p
erties per se are no anathem a to a functionalist theory of m ental
states.
A . The problem o f in verted/gerrym an dered qualia

This problem is just the m ost straightforw ard illustration of the g en er
al w orry that functionalism leaves out som ething essential. The recipe
for concocting the appropriate intuitions runs thus. Suppose that the
sensations having the quale typical of pain in you play the functional
role of pleasure sensations in som eone else, and the quale typical of
p leasure sensations in you are had instead by the sensations that
h ave the functional role of pain in him. Functionally, w e are to sup
p ose, the tw o of you are indistinguishable, but his pleasure/pain qual
ia are simply inverted relative to their distribution am ong y ou r ow n
sensations, functionally identified. A variation on the recipe asks us
to im agine som eone with an inverted distribution of the color qualia
that characterize y ou r ow n visual sensations (functionally identified).
H e thu s has (w hat you w ould introspectively identify as) a sensation
of red in all and only those circu m stan ces w here you have a sensation
of green , and so forth.
These cases are indeed im aginable, and the connection betw een
quale and functional syndrom e is indeed a contingent one. W h eth er
it is the quale or the functional syn drom e that determ ines type iden
tity qua psychological state, w e m u st now address. The intuitions
evoked above seem to confound functionalist pretensions. The objec
tion to functionalism is that w hen the inversion victim has that sen sa
tion w hose functional properties indicate pleasure, he is in fa c t fee lin g
pain, functional properties notw ithstanding; and that w hen the victim
of a sp ectrum inversion says, "I have a sensation of g re e n " in the
p resen ce of a green object, he is in fa c t having a sensation o f red, func
tional properties notw ithstanding. So far as type identity of p sy ch o
logical states is concerned, the objection concludes, sam eness of
qualitative ch aracter dom inates over sam eness of functional role.
Functionalism, Qualia, and Intentionality 25
N ow there is no point in trying to deny the possibilities just out

lined. R ather, w hat the functionalist m u st argue is that they are better
described as follows. "Y o u r pains have a qualitative ch aracter rath er
different from that of his pains, and your sensations-of-green have a
qualitative ch aracter rath er different from that of his sensations-of-
green. Such internal differences am ong the sam e psychological states
are neither inconceivable n or even perhap s very u n u su a l." That is to
say, th e functionalist should concede the juggled qualia, while con
tinuing to reckon type identity in accordan ce w ith functional syn
drom e. This line has a certain intuitive appeal of its o w n , though
rather less than the opposing story, at least initially. H ow shall w e
decide betw een these com peting intuitions? By isolating the consid
erations that give rise to them and exam ining their integrity.
The "p ro -q u alia" intuitions, w e su ggest, derive from tw o main
sources. To begin w ith, all of us have a strong and entirely u n d er
standable tendency to think of each type of psychological state as
constituting a natural kind. After all, these states do play a vigorous
explanatory and predictive role in everyday com m erce, and the com
m on sen se conceptual fram ew ork that com p reh en d s them has all the
features and functions of a sophisticated empirical theory (see Sellars
1956; P. M. C hurchland 1979). To think of pains, for exam ple, as con
stituting a natural kind is to think of them as sharing an intrinsic
nature that is com m on and essential to every instance of pain. It is
u nderstandable then, that the qualitative character of a sensation,
the only nonrelational feature to w hich w e have access, should p re
sent itself as being that essential elem ent.
O u r inclination to such a view is further encouraged by the fact that
one's introspective discrim ination of a sensation's qualitative ch arac
ter is far and aw ay the m ost im m ediate, m ost autom atic, m ost deeply
en trench ed , and (in isolation) m ost authoritative m easu re of w hat
sensations one has. In on e's ow n case, at least, the functional features
of o n e's sensations play a m inor role in on e's recognition of them . It is
as if one had special access to the intrinsic nature of any given type of
sensation, an access that is independent of the purely contingent and
causal features that constitute its functional role.
Taken conjointly, these considerations will fund v ery strong in
tuitions in favor of qualia as the determ inants of type identity for
psychological states. But though natural enough, the rationale is
exceptionally feeble on both points.
Take the first. H ow ever accustom ed or inclined w e are to think of
our psychological states as constituting natural kinds, it is vital to see
that it is not a sem antic or a conceptual m atter, b ut an objective em pir
ical m atter, w heth er or not they do. Either there is an objective intrin
sic nature com m on to all cases of, e .g ., pain, as it occurs in hu m an s,

chim panzees, dogs, snakes, and o ctop u ses, or there is not. A nd the
fact is, the functionalist can point to som e rath er persuasive consid
erations in support of the view that there is not. Given the p hysio
logical and chem ical variety w e find in the nervous system s of the
m any anim als that feel pain, it appears very unlikely that their pain
states have a com m on physical nature underlying their com m on
functional nature (see Pu tn am 1971). It rem ains possible that they all
have som e intrinsic nonphysical nature in com m on. But dualism is
profoundly implausible on sheer evolutionary grounds. (Briefly, the
evolutionary p rocess just is the diachronic articulation of physical
m atter and energy. If w e accept an evolutionary origin for ourselves,
then our special capacities m ust be con stru ed as the capacities of
one particular articulation of m atter and energy. This conclusion is
confirm ed by our increasing understan din g of the nervous system ,
of both its p ast evolution and its cu rren t regulation of b eh avior.) In
sum , the em pirical presum ption against natural-kind status for
psychological states is substantial. W e should not place m u ch trust,
therefore, in intuitions born of an uncritical prejudice to the contrary.
Such intuitions m ay reflect ordinary language m ore or less faithfully,
but they beg the question against functionalism .
The facts of introspection provide no better grounds for thinking
that sensations constitute natural kinds, or for reckoning qualia as
their constituting essences. That the qualitative character Q of a
psychological state S should serve as the standard ground of S 's in
trospective discrim ination is entirely consistent w ith Q 's being a non-
essential feature of S. The black and yellow stripes of a tiger serve as
the standard ground on w hich tigers are visually discrim inated from
oth er big cats, b ut the stripes are hardly an essential elem ent of tiger-
hood: there are albino tigers as well as the very pale H im alayan
tigers. The telling question here is this: w hy should the qualia of our
familiar psychological states be though t any different? W e learn to
pick out those qualia in the first place, from the teem ing chaos of our
inner lives, only because the states thu s discrim inated are also the
nexus of various generalizations connecting them to oth er inner
states, to environm ental circum stances, and to overt behaviors of in
terest and im portance to us. H ad our cu rrent taxon om y of introspecti-
ble qualia b een unsuccessful in this regard , w e w ould m ost certainly
have throw n it over, centered our attention on different aspects of the
teem ing chaos w ithin, and recarved it into a different set of similarity
classes, a set that did display its objective integrity by its m any nom ic
connections, both internal and external. In short, the internal w orld
com es precarved into observational kinds no m ore than does the ex
ternal w orld, and it is evident that the introspective taxon om ies into
w hich w e eventually settle are no less shaped by considerations of
explanatory and causal coherence than are the taxon om ies of external
observation.
It is therefore a great irony, it seem s to us, that an yon e should
subsequently point to w h atever qualia our introspective m echanism s
have m an aged tenuously to fix upon as m ore or less usable indicators
of nom ologically interesting states, and claim them as constituting the
essence of such states. It is, of course, distantly possible that our
m echanism s of introspective discrim ination have lucked onto the
constituting essences of our psychological states (assum ing, contrary
to our earlier discussion, that each type has a uniform natural es
sence), b ut a priori that seem s about as likely as that the visual sys
tem lucked onto the constituting essence of tigerhood w h en it m ade
black-on-yellow stripes salient for distinguishing tigers.
It seem s very doubtful, therefore, that the type identity of any
psychological state derives from its sharing in any uniform natural
essen ce. M oreover, even if it does so share, it seem s entirely unlikely
that introspection provides any special access to that essen ce. C on se
quently, this beggars the intuition that sustains the inverted-qualia
objections.
The p receding investigation into the w eight and significance of fac
tors determ ining type identity of psychological states d oes m ore than
that, how ev er. It also enriches the com peting intuition, nam ely, that
the typ e identity of psychological states is determ ined by functional
characteristics. To rep eat the point m ade earlier, since the taxon om y
of observational qualia constructed by the questing child fo llo w s the
discovered taxon om y of states as determ ined by interesting causal
roles, it is evident that sam eness of functional role dom inates over
differences in qualitative character, so far as the type identity of
psychological states is concerned. That a single category, united by
functional considerations, can em brace diverse and disparate qualita
tive ch aracters has a ready illustration, ironically en ou gh , in the case
of pain.
C onsider the wide variety of qualia willfully lum ped togeth er in
com m on practice un d er the heading of pain. C om p are the qualitative
character of a severe electric shock w ith that of a sh arp blow to the
kneecap; com pare the character of hands dully aching from m aking
too m an y snowballs w ith the piercing sensation of a jet engine heard
at very close range; com pare the ch aracter of a frontal h ead ach e w ith
the sensation of a scalding pot grasped firmly. It is evident that w hat
unites sensations of such diverse characters is the similarity in their
functional roles. The sudden onset of any of them p rom p ts an in
voluntary w ithdraw al of som e sort. O ur reaction to all of them is im

m ediate dislike, and the violence of the dislike increases w ith the
intensity and duration of the sensation. All of them are indicators of
physical traum a of som e kind, actual or im m inent. All of them tend to
p rodu ce shock, im patience, distraction, and vocal reactions of famil
iar kinds. Plainly, it is these collected causal features that unite the
class of painful sensations, n ot som e uniform quale, invariant across
cases. (For a general account of the intentionality of sensations, in
w hich qualia also retreat into the background, see P . M . C hurchland
1979, ch ap ter 2.)
The converse illustration is provided by states having a uniform or
indistinguishable qualitative ch aracter, states that are nevertheless
distinguished by us according to differences in their functional roles.
F or exam ple, our em otions have a certain qualitative ch aracter, but it
is often insufficient to distinguish w hich of several em otions should
be ascribed. O n a particular occasion, the felt knot in o n e's soul m ight
be mild sorrow , severe disappointm ent, or gathering despair, and
w hich of these it is— really is— w ould depend on the circum stances
of its production, the rest of o n e's psychological state, and the co n
sequences to w hich it tends to give rise. Its type identity need n ot be a
m ystery to its p ossessor— he has introspective access to som e of the
context that em beds it— but the identification cannot be m ade on
qualitative grounds alone. Similarly, a therapist m ay be n eeded, or a
thoughtful friend, to help you distinguish you r decided unease about
som e person as you r hatred for him , en vy of him , or simple fear of
him . The fe lt quality of your unease m ay be the sam e for each of these
cases, but its causes and effects w ould be significantly different for
each. H ere again, functional role is the dom inant factor in the type
identity of psychological states.
The reason that functional role dom inates introspectible qualitative
differences and similarities is not that the collected law s descriptive of
a state's functional relations are analytically true, or that they exh au st
the essence of the state in question (though, withal, they m ay). The
reason is that the com m onsen se conceptual fram ew ork in w hich our
psychological term s are sem antically em bedded is an em pirical theory.
A s w ith theoretical term s generally, their changeable position in
sem antic space is fixed by the set of theoretical law s in w hich they
figure. In the case of folk psychology, those laws express the causal
relations that connect psychological states w ith one another, w ith en
vironm ental circum stances, and with behavior. Such laws n eed not
be seen, at any given stage in our grow ing understanding, as exhaust
ing the essence of the states at issue, but at any given stage they co n
stitute the best-founded and m ost authoritative criterion available for

identifying those states.
W e conclude against the view that qualia constitute an essential
elem ent in the type identity of psychological states. Variations w ithin
a single type are both conceivable and actual. The im agined cases of
qualia inversion are of interest only b ecause they place directly at
odds intuitions that norm ally coincide: the noninferential im pulse of
observational habit (i.e., qualia) against the ponderous background of
theoretical understanding (i.e ., functional role). H ow ever, the qual
itative character of a sensation is a relevant m ark of its type identity
only w h en and only insofar as that ch aracter is the uniform con com
itant of a certain repeatable causal syndrom e. In the qualia-inversion
thou gh t experim ents, that uniform ity is broken. A nd consequently,
so is the relevance of those qualia for judging type identity, at least
insofar as they can claim a uniform relevance across people and across
tim es.
B. The problem o f absent qualia

The p reced in g argum ents m ay settle the inverted-qualia problem , but
the position w e have defended is thought to raise in tu rn an even
m ore serious problem for functionalism (see Block and Fod or 1972;
Block 1978). If the particular quale a sensation has contributes nothing
to its type identity, then w hat are w e to say of a psychological system
that is functionally isom orphic to us, but w hose functional states have
no qualia w hatever? Surely such system s are possible (nom ically as
well as logically), runs the objection. Surely functionalism entails that
such a system feels pain, w arm th, and so on. But since its functional
states have no qualitative character w hatever, surely su ch a system
fee ls nothing at a lll Functionalism , accordingly, m u st be false.
This argu m en t is m uch too glib in the con trast it assu m es betw een
functional features (w hich supposedly m atter to functionalism ) and
qualitative character (w hich supposedly does not). A s the functional
ist should be the first to adm it, our various sensations are introspec-
tively discrim inated by us on the basis of their qualitative ch aracter,
and an y adequate psychological theory m ust take this fact into
accoun t. H ow m ight functionalism do this? Straightforw ardly. It
m u st require of any state that is functionally equivalent to the
sen sation-of-w arm th, say, that it have som e intrinsic p roperty or
oth er w hose p resen ce is detectable by (that is, is causally sufficient for
affecting) our m echanism s of introspective discrim ination in such a
w ay as to cause, in a conceptually com petent creatu re, belief states
such as the belief that it has a sensation-of-w arm th. If these sorts of
causal relations are not part of a given state's functional identity, then
it fails to be a sensation-of-w arm th on purely functional grounds.

(Sydney Shoem aker m akes m u ch the sam e point in his 1975. W e do
not know if he will agree with the points that follow.)
So functionalism does require that sensations have an intrinsic p ro p
erty that plays a certain causal role. But it is adm ittedly indifferent as
to precisely w hat that intrinsic p roperty m ight h appen to be for any
given type of sensation in any given p erson . So far as functionalism is
co ncern ed, that intrinsic p roperty m ight be the spiking frequency of
the signal in som e neural p ath w ay, the voltage across a polarized
m em b rane, the tem porary deficit of som e neurochem ical, or the
binary configuration of a set of direct-current pulses. So long as the
p roperty is one to w hich the m echanism s of introspective discrim
ination are keyed, the p roperty fills the bill.
"B u t these are not qualia!" chorus the outraged objectors. A re they
not indeed. Recall the characterization of qualia given on the first
page of this chapter: "th o se intrinsic or m onadic properties of our
sensations discrim inated in in tro sp ection ." O ur sensations are any
w ay token-identical w ith the physical states that realize them , so
there is no problem in construing a spiking frequency of 60 hertz as
an intrinsic property of a given sensation. A nd w hy should such a
p roperty, or any of the others listed, not be at the objective focus of
introspective discrim ination? To be sure, they w ould be opaquely dis
crim inated, at least by creatures w ith a prim itive self-conception like
our ow n. That is to say, the spiking frequency of the im pulses in a
certain neural p ath w ay need not p rom p t the noninferential belief,
"M y pain has a spiking frequency of 60 h e rtz "; it m ay p rom p t only
the belief, "M y pain has a searing qu ality." But w ithal, the property
you opaquely distinguish as searingness m ay be precisely the p ro p er
ty of having 60 hertz as a spiking frequency.
There are m any preceden ts for this sort of thing in the case of the
intrinsic properties of m aterial objects standardly discrim inable in
observation. The redness of an object turns out to be a specific reflect
ance triplet for three critical w avelengths in the electrom agnetic sp ec
trum . The pitch of a singer's note turns out to be its frequency of
oscillation in air pressu re. The w arm th of a coffee cup turns out to
be the vibrational energy of its m olecules. The tartness of one's
lem onade turns out to be its high relative concentration of H + ions.
A n d so forth.
These chem ical, electrom agnetic, and m icrom echanical properties
have been briskly discrim inated by us for m any millennia, but only
opaquely. The reason is that w e have not possessed the con cep ts
n ecessary to m ake m ore penetrating jud gm en ts, and our m echanism s
of sensory discrim ination are of insufficient resolution to reveal on
their ow n the intricacies that w ere eventually uncovered by other

m ean s. U nam biguous perception of m olecular kinetic en ergy, for ex
am ple, w ould require a sensory apparatus capable of resolving dow n
to about 1 0 -10 m eters, and capable of tracking particles having the
velocity of rifle bullets, millions of them , sim ultaneously. O ur sen
sory ap p aratu s for detecting and m easuring m olecular kinetic energy
is rath er m ore hum ble, but even so, it con n ects us reliably w ith the
p aram eter at issue. M ean m olecular kinetic energy m ay n ot seem like
an observable p roperty of m aterial objects, but m ost assuredly it is.
(For a w orking-out of these them es in detail, see P. M. C hurchland
1979.)
Similarly, spiking frequency m ay not seem like an introspectible
p roperty of sensations, but there is no reason w hy it should not be,
and there is no reason w hy the epistem ological story for the faculty of
inner sense should be significantly different from the story told for
outer sen se. Qualia, therefore, are not an ineffable m ystery, any m ore
than colors or tem peratu res are. They are physical features of our
psychological states, and w e m ay expect qualia of som e sort or other
in any physical system that is sufficiently com plex to be functionally
isom orphic w ith our ow n psychology. The qualia of su ch a robot's
states are n ot "a b s e n t." They are m erely unrecognized by us under
their physical/electronic descriptions, or as discrim inated by the
m odalities of outer sense rather than inner sense.
W e m ay sum m arize all of this by saying that the functionalist need
not, and perh ap s should not, attem pt to deny the existence of qualia.
R ather, he should be a realist about qualia— in particular, he should
be a scientific realist.
It is im portan t to appreciate that one can be reductionistic about
qualia, as outlined above, w ithout being the least bit reductionistic
about the taxon om y of states appropriate to psychological theory.
Once qualia have been denied a role in the type identity of psycholog
ical states, the path described is open. If this line on qualia is correct,
then it vindicates N ed Block's prophecy (1978, p. 309) th at the explica
tion of the nature of qualia does not reside in the dom ain of psychol
ogy. O n the view argued here, the nature of specific qualia will be
revealed by neurophysiology, n eurochem istry, and neurophysics.
C. The problem o f distinguishing states w ith qualia fro m states w ithout

O ne could distinguish m any differences b etw een the sensations and
the propositional attitudes, but one particular difference is of special
interest here. A sensation-of-w arm th, for exam ple, h as a distinct
qualitative ch aracter, w hereas the belief-that-Tom -is-tall does not.
Can functionalism explain the difference?
Y es it can. The picture to be avoided here depicts sensations as

dabbed w ith m etaphysical paint, while beliefs rem ain undabbed and
colorless. The real difference, w e su ggest, lies less in the objective
nature of sensations and beliefs them selves than in the nature of the
introspective m echanism s used to discrim inate and identify the states
of each class. This hypothesis requires explanation.
H ow m any different types of sensation are there? O ne hu n dred?
O ne thou sand ? Ten thousand? It is difficult to m ake an estim ate,
since m ost sensations are arrayed on a qualitative continuum of som e
sort, and it is to som e extent arbitrary w here and how finely the lines
betw een different kinds are draw n. It is plain, h ow ever, that the
num ber of distinct continua that w e recognize, and the num ber of
significant distinctions w e draw w ithin each , is sufficiently small that
the brain can use the following strategy for m aking noninferential
identifications of sensations.
C onsider the various physical p roperties that, in you, are ch aracter
istic of the repeatable brain state that realizes a given sensation. Sim
ply exploit w hichever of those physical properties is accessible to your
innate discrim inatory m echanism s, and contrive a standard habit of
conceptual respon se ("L o , a sensation of w a rm th ") to the property-
evoked activation of those m echanism s. W hile this strategy will w ork
nicely for the relatively small class of sensations, it will not w ork at all
well for the class of beliefs, or for any of the other propositional atti
tudes. The reason is not that the brain state that realizes a certain
belief lacks intrinsic properties characteristic of it alone. R ather, the
reason is that there are far too m any beliefs, actual and possible, for
us to have any hope of being able to discrim inate and identify all of
them on such a one-by-one basis. The num ber of possible beliefs is
at least a denum erable infinity, and the num ber of possible beliefs
expressible in an English sentence of ten w ords or less is still
stu p en dou s. If w e assum e a vocabulary of 105 w ords for English, the
num ber of distinct strings of ten w ords or less is 1050. If w e con serva
tively assum e that only one string in every trillion trillion is gram
matically and sem antically well form ed, this still leaves us over
1025 distinct sen ten ces. Even if there w ere a distinct and accessible
m onadic p roperty for each distinct belief state, therefore, the capacity
of m em ory is insufficient to file all of them . Evidently the brain
m u st u se som e m ore system atic strategy for discrim inating and
identifying beliefs, a strategy that exploits in som e w ay the unique
com binatorial structure of any belief.
But this is a very com plex and sophisticated m atter requiring the
resou rces of our higher cognitive capacities, capacities tuned to the
com plex relational, structural, and com binatorial features of the
dom ain in w hich the discrim inations are m ade. Unlike the sensation
case, no narrow range of stim ulus-response connections will begin to
characterize the m echanism s at w ork here.
Sensations and beliefs, accordingly, m u st be introspectively dis
crim inated by entirely distinct cognitive m echanism s, m echanism s
facing quite different problem s and using quite different strategies
for their solutions. Sensations are identified by w ay of their intrin
sic properties; beliefs are identified by w ay of their highly abstract
structural features. It should not be w ond ered at, then, th at there is a
subjective contrast in the nature of our aw areness of each.
[A dded in 1989: I m ust now express a loss of confidence in this
argu m en t. The problem is that sensations now ap p ear to be decidedly
m ore various than I had originally estim ated and to h ave a m uch
m ore intricate com binatorial structure than I had earlier supposed
(see ch ap ter 5, sec. 7). A ccordingly, the con trasts on w hich the p re
ceding argu m en t places so m u ch w eight now ap p ear spurious: w hat
seem ed a large difference in kind now seem s a m ere difference in
d egree.]
D. The differentiation problem

This problem arises because w e are occasionally able to discrim inate
b etw een qualitatively distinct sensations w here w e are ignorant of
any corresp on d in g functional differences betw een them , and even
w here w e are w holly ignorant of the causal properties of both of
them , as w h en they are new to us, for exam ple. These cases are
thought to constitute a problem in that functional considerations
should bid us count the states as type identical, w hereas by h yp oth
esis they are type distinct (see Block 1978, p. 300).
The objection has tw o defects. First, sheer ignorance of functional
differences need not bind us to counting the sensations as functional
ly identical. The functionalist can and should be a realist about func
tional properties. Functional identities are not determ ined by w hat
w e do or do not know , but by w hat is actually out there in the w orld
(or in there in the brain). Second, the objection begs the question
against functionalism by assum ing that a discrim inable qualitative
difference betw een tw o sensations entails that they are type distinct
qua psychological states. W e have already seen that this inference is
w ron g: pains display a variety of qualitative characters, but because
of their functional similarities, they still count as pains.
In sh ort, w e can and do m ake discrim inations am on g our sen sa
tions in ad van ce of functional understanding. But w h eth er the dis
crim inations thus m ade m ark a difference of any im portance for the
taxon om y of psychological theory is an oth er question. In som e cases
they will; in other cases they will not. W h at decides the m atter is
w heth er those qualitative differences m ark any causal or functional
differences relevant to the explanation of psychological activity and
overt behavior.
So long as introspectible qualia w ere thought to be ineffable, or
epiphenom enal, or dualistic, or essential for type identity, one can
un derstan d the functionalist's reluctance to have anything to do with
them . But once w e have seen how the functionalist can acknow ledge
them and their epistem ic role, w ithin a naturalistic fram ew ork, the
reluctance should disappear. For the taxon om y of states appropriate
for psychological theory rem ains dictated entirely by causal and ex
p lanatory factors. Qualia are just accidental hooks of opportunity for
the introspective discrim ination of dynam ically significant states.
2 The P roblem o f N onstandard Realizations
Som e of the issues arising here have already been broached in the
subsection on absent qualia. H ow ever, novel problem s arise as well,
and organization is best served by a sep arate section. All of the prob
lem s here begin with the functionalist's central contention that the
functional organization n ecessary and sufficient for person h ood is an
abstract one, an organization realizable in principle in an indefinite
variety of physical system s. Such liberalism seem s innocent enough
w h en w e contem plate the p rosp ect of hum anoid aliens, biom echani
cal androids, and electrom echanical robots w hose physical con stitu
tions are at least rough parallels of our ow n. W ho could deny that
C 3PO and R2D 2— of Star W ars fam e— are p ersons? But our liberal
intuitions are quickly flum m oxed w hen w e consider bizarre physical
system s th at m ight nevertheless realize the abstract causal organiza
tion at issue, and such cases m ove one to reconsider o n e's generosity
in the m ore familiar cases as well.
The following discussion will explore but tw o of these nonstan d ard
"p e rso n s” : N ed Block's "C h in ese n atio n " (1978) and John Searle's
"C hinese-speak in g ro o m " w ith the m onolingual anglophone locked
inside it (Searle 1980). Block is concerned w ith the absence of qualia
from states posing as sensations, and Searle is concerned w ith the
absence of intentionality from states posing as propositional attitudes.
A. Q ualia in the C hinese nation

Block's exam ple will be exam ined first. H e has us im agine a certain
Turing m achine T m, w hich is realized in the population of China, as
follows. Each citizen has a tw o-w ay radio link to a certain robotic
device w ith sensory transducers and m otor effectors. This robot is the
b ody of the sim ulated p erson , and it interacts w ith its collective brain
as follows. It sends a sensory input m essage L to every single citizen
and subsequently receives a m otor output m essage O, from exactly
one citizen. W hich citizen sends w hat outp u t is determ ined as
follows.
O verh ead from a satellite som e state letter Sk is displayed in lights
for all to see. For each possible state letter Sk there is p erm anently
assigned a distinct subset of the population. In the rare ev en t w hen Sk
is displayed and input J; is received, one p erson in the Sk group, a
p erson to w hom Ij has been assigned, perform s the following p re
assigned task. She sends to the robot the unique output m essage O,
antecedently assigned to her for just this occasion, radios the satellite
to display the state letter Sf, antecedently assigned to h er for just this
occasion, and then subsides, w aiting for the next opportunity to do
exactly the sam e thing in exactly the sam e circum stances.
A s organized above, each citizen realizes exactly one square of the
machine table that specifies T m. (A machine table is a matrix or checker
board w ith state letters heading the colum ns and input letters h ead
ing the row s. A n y square is the intersection of som e Sk and Ip and it
specifies an output O, and a shift to som e state S„, w here possibly
p = k. See figure 2. 1. ) Block asks us to assum e that Tm adequately
sim ulates you r ow n functional organization. O ne is likely to grant
him this, since any inp u t-outpu t function can in principle be realized
in a suitable Turing m achine. In pondering an apparently fussy
detail, Block w on d ers, "H o w m any hom unculi are req u ired?" and
an sw ers, "P erh ap s a billion are enough; after all, there are only about
a billion n euron s in the brain" (p. 278). H ence his choice of China as
the potential artificial brain. (Block underestim ates here. The num ber
of n eu ron s in the brain exceeds 100 billion.)
Finally, Block finds it starkly implausible to sup p ose that this
nationw ide realization of Tm has states w ith a qualitative character
like pains, tastes, and so on. It is difficult n ot to agree w ith him . His
hom unculi do not even interact w ith one an other, save indirectly
through the satellite state letter and even less directly throu gh the
ad ventures of the robot body itself. The shim m ering intricacies of
on e's inner life are not to be found here.
The w ay to avoid this criticism of functionalism is just to insist that
any subject of beliefs and sensations m u st not only be Turing equiva
lent to us (that is, produ ce identical outpu ts given identical inputs); it
m ust be com putationally equivalent to us as well. That is, it m u st have
a system of inner states w hose causal interconnections m irror those in
our ow n case. This is not an arbitrary restriction. Folk psychology is,
and scientific psychology should be, realistic about our m ental states,
Figure 2.1
A Turing machine table. The machine is always in some one or other internal state S„
represented by one of the vertical columns. When it receives an input l m, it executes the
instructions in the specific square where the S„ column and the I„, row intersect, and
then it waits for the next input.
and m ere parity of gross behavior does not guarantee parity of causal
organization am ong the states that produ ce it. The com putational
organization displayed in the C hinese-nation Turing m achine is not
even distantly analogous to our ow n. If it w ere analogous to our ow n,
w orries about absent qualia could be handled as outlined in the sub
section about absent qualia above. That is, w e could insist that the
qualia are there all right, but are unrecognized by us un d er their
physical descriptions.
There is a further reason w hy it is not arbitrary to insist on a com
putational organization m ore along the lines of our ow n, and w e m ay
illustrate it by exam ining a further defect in Block's exam ple. It is
dem onstrable that no T,„ realized as described in the population of
China could possibly simulate you r input-output relations. There are
not nearly enough C hinese for this job, not even rem otely enou gh . In
fact, a spherical volum e of space centered on the Sun and ending at
Pluto's orbit packed solidly w ith cheek-to-cheek C hinese (roughly

1036 hom unculi) w ould still not be rem otely en ou gh , as I shall show .
Since it is realized on a one-m an /on e-sq uare basis, the C hinese
Tm can have at m ost 109 distinct possible outputs, and at m ost 109/S
distinct possible inputs, w here S is the total num ber of distinct state
letters. That is, Tm has rath er less than 109 possible inputs. H ow
m any distinct possible inputs characterize y ou r ow n functional
organization? Since the p resen t argum ent requires only a low er limit,
let us consider just one of you r retinas. The surface of y o u r retina
contains roughly 10® light-sensitive cells, w hich w e shall con serva
tively assu m e to be capable of only tw o states: stim ulated and u n
stim ulated. Good eyesight has a resolution limit of about one foot at a
distance of a mile, or slightly less than one arc-m inute, and this angle
projected from the lens of the eye subtends about six m icrons at the
retina. This is roughly the distance betw een the individual cells to be
found there, so it is evident that individual cells, and n ot ju st groups,
can serve as discrim inative atom s, functionally speaking.
If w e take distinct stim ulation patterns in the set of retinal cells as
distinct inputs to the brain, it is evident that w e are here dealing w ith
2 to the (108)th pow er distinct possible inputs. This is an appallingly
large num ber. Since 2332 = Ю100 (a googol), 2 108 = lO30'000-000 distinct
possible inputs from a single retina! Since a one-m an /on e-sq uare Tur
ing m achine m ust have at least as m any hom unculi as possible in
puts, an y such realization adequate to the inputs from a single retina
w ould require no less than lO30-000-000 distinct hom unculi. H ow ever,
there are only about 1080 distinct atom s in the accessible universe.
Small w on d er the C hinese nation m akes an unconvincing simulation
of our inner lives. Plainly, w e should never have acquiesced in the
prem ise that a Turing m achine thus realized could even begin to
sim ulate you r overall functional organization. The C hinese robot
body can have at m ost a m ere 30 binary input sensors, since 2 30 ~ 109,
and the n um ber of distinct inputs cannot exceed 109.
This argu m en t does not depend on inflated estim ates concerning
the retina or its input to the brain. (It m ight be objected, for exam ple,
that retinal cell stim ulation is not ind ependent of the state of its
im m ediate neighbor cells, or that the optic nerve has only 800,000
axo n s.) If y ou r retina contained only 332 discrim inatory units, instead
of 108, the num ber of distinct inputs w ould still be 2332, or roughly
10100: ninety-one orders of m agnitude beyond the capacity of the
C hinese nation, and tw enty orders of m agnitude beyond the atom s
in the universe. N or h ave w e even begun to consider the other
dim ension of the required m achine table: the range of states, S, of
the brain that receives these inputs, a brain w hich has at least 1011
distinct cells in its ow n right, each with about 103 connections with
oth er cells. O ur estim ate of the num ber of distinct states of the brain
m u st be substantially in excess of ю 30-000-000, our num ber for the
retina.
O ur conclusion is that no brute-force, one-device/one-square
realization of a Turing m achine constructible in this universe could
even begin to sim ulate you r inp u t-ou tp u t organization. E ven the
humblest of creatures are beyond such simulation. A n unprepossess
ing gastropod like the sea slug A plysia californica has well in excess of
332 distinct sensory cells, and thus is clearly beyond the reach of the
crude m eth ods at issue. This does not m ean that the h um an input-
o u tp u t relations cannot be rep resented by an abstract Turing m achine
Tm. W h at it does m ean is that any physical m achine adequate to such
sim ulation m ust have its com putational architecture and executive
h ard w are organized along lines vastly different from , and m uch m ore
unified and efficient than, those displayed in Block's exam ple. That
exam ple, therefore, is not even rem otely close to being a fair test of
our intuitions. Quite aside from the question of qualia, the Chinese
Turing m achine couldn't sim ulate an earth w orm .
This w eakness in the exam ple is not adequately m ade up by allow
ing, as Block does at one point (p. 284), that each hom unculus m ight
be responsible for a wide range of inputs, each w ith corresponding
o utp uts. O n this m odification, each hom unculus w ould thus realize,
dispositionally, m an y m achine-table squares sim ultaneously. Sup
pose, then, that w e m ake each C hinese citizen responsible for one
billion squares peculiar to him (this w as the size of the original Tm).
This raises the num ber of distinct inputs processible by the system
to 109 citizens X 109 squares = 1018 possible inputs, still well short
of the lO30-000'000 w e are striving for.
W ell, how m any squares m uch each citizen realize if the nation as a
w hole is to instantiate som e Turing m achine adequate to handle the
required input? The answ er is, of course, 10<30'000'000 —9) squares each.
But how will each hom unculus-citizen handle this aw esom e load? N ot
by being a simple one-device/one-square Turing m achine in turn,
as w e have already seen. N o physical sim ulation adequate to you r
input-output relations, therefore, can avoid having the m ore unified
and efficient m odes of com putational organization alluded to in the
p receding p aragrap hs, even if they show up only as m odes of orga
nization of its various subunits. That is, any successful sim ulation of
you m u st som ew here display a com putational-executive organization
that is a m u ch m ore plausible hom e for qualitative states than Block's
exam ple w ould suggest. [Added in 1989: A good exam ple of a
m ore plausible hom e is the vector coding and parallel distributed

processing arrangm en t discussed in chapter 5.]
But can a num ber of distinct p ersons or near person s collectively
constitute a further p erson? A pparently so, since the system consist
ing of you r right hem isphere and you r left hem isphere (and you r
cerebellum and thalam us and limbic system , etc.) seem s to do p re
cisely th at. F urther attem pts to con struct hom unculi-headed cou n ter
exam ples to functionalism should p erhaps bear this fact in m ind.
The arg u m en t of the preceding pages does not, of cou rse, show
that the specific details of our com putational organization are essen
tial to achieving the inform ational capacity required. A nd this raises a
question w e m ight have asked anyw ay: if w e do require of any sub
ject of sensations, beliefs, and so forth, that it be functionally equiva
lent to u s in the strong sense of "com pu tationally eq u iv alen t," do we
then n ot run the opposite danger of allowing too fe w things to count
as sites of genuine m entality? (See again Block 1978, p. 310ff, on spe
cies chauvinism.) If we restrict the application of the term "m entality"
to creatu res having sensations, beliefs, intentions, e tc., w e shall in
deed h ave becom e too restrictive. Yet the functionalist n eed not p re
tend that our internal functional organization exhausts the possible
kinds of m entality. H e need only claim that our kind of internal func
tional organization is w hat constitutes a psychology of beliefs, desires,
sensations, and so forth. H e is free to suggest that an alien functional
organization, com parable only in sophistication to our o w n , could
constitute an alien p sychology of quite different internal states. W e
could then speak of M artian m entality, for exam ple, as well as of
hum an m entality.
Still, it m ight be w on d ered, w hat is the shared essence that m akes
both of us instances of the now m ore general term "m e n ta lity "? There
need be non e, beyond the general idea of a sophisticated control cen
ter for com plex behavior. O ne of the functionalist's principal theses,
after all, is that there are no natural essen ces to be found in this
dom ain. If he is right, it is folly to seek them . A n d in any case, it is
question begging to dem and that he find them .
O n the oth er h and, there m ay yet p rove to be som e interesting
natural kind of w hich both w e and the M artian are variant instances:
som e highly abstract therm odynam ic kind, perh ap s. In that case,
orth od ox functionalism w ould be m istaken in one of its p urely n ega
tive theses. O n this m atter, see the final section of this essay.
B. Intentionality in the Chinese room

Let us now turn to John Searle's w orries about m eaning and inten
tionality. The states at issue here are beliefs, thou gh ts, desires, and
the rest of the propositional attitudes. O n the functionalist's view , the

type identity of any of these states is determ ined by the netw ork of
relations it b ears to the oth er states and to external circum stances and
behavior. In the case of the propositional attitudes, those relations
characteristically reflect a variety of logical and com putational rela
tions am ong the propositions that the attitudes "c o n ta in ." W e can
thus at least im agine a com puter of sufficient capacity p rogram m ed so
as to display an econ om y of internal states w hose interconnections
m irror those in our ow n case. The sim ulation w ould create the re
quired relational order by exploiting the logical and com putational
relations defined over the form al/structural/com binatorial features of
the individual propositional states.
Searle has no doubt that such a sim ulation could, in principle, be
con stru cted . His objection to functionalism is that the states of such a
system w ould nevertheless lack real m eaning and intentionality: "N o
purely formal m odel will ever be sufficient by itself for intentionality
b ecause the form al properties are not by them selves constitutive of
intentionality" (Searle 1980, p. 422). H is reason s for holding this
position are illustrated in the following thou gh t experim ent.
Im agine a m onolingual anglophone locked in a room w ith a sub
stantial store of sequences of C hinese sym bols, and a set of com plex
transform ation rules, w ritten in English, for perform ing operations
on sets and sequences of C hinese sym bols. The occu p an t periodically
receives a new sequence of C hinese sym bols through a postal slot. He
applies his transform ation rules dutifully to the ordered pair <new
sequence, old store of sequences >, and they tell him to w rite out
a further sequence of C hinese sym bols, w hich he sends back out
throu gh the postal slot.
N ow , unknow n to the occup ant, the large store of sequences em
bodies a rich store of inform ation on som e one or m ore topics,
all w ritten in Chinese. The new sequences sent through the door
are questions and com m ents on those topics. The transform ation
rules are a cunningly devised p rogram designed to sim ulate the
thou gh t processes and conversational behavior of a native speaker
of C hinese. The symbol sequences the occu p an t sends out are
"re s p o n s e s" to the queries and com m en ts received. W e are to
sup pose that the transform ation rules are well devised, and that
the sim ulation is as convincing as you please, considered from the
outside.
H ow ever convincing it is, says Searle, it rem ains plain that the
ro om 's occu pan t does not u n d erstan d C hinese: he applies tran s
form ation rules, and he u n derstan ds those rules, but the sequences of
C hinese sym bols are m eaningless to him . Equally clear, claims Searle,
is that the entire system of the room plus its contents does n ot u n d er
stand C hinese either. N othing here u n derstan ds C hinese, save those
sending and receiving the m essages, and those w ho w rote the p ro
gram . N o com putational state or output of that system h as any m ean
ing or intentionality save as it is interpretively im posed from
w ithout by those w ho interact with it.
H ow ever, concludes Searle, this system already contains every
thing relevan t to be found in the physical realization of any purely
formal p rogram . If m eaning and intentionality are m issing here, they
will be m issing in any such attem pt to sim ulate h u m an m ental
activity. Instantiating a p rogram cannot be a sufficient condition of
u n d erstanding,
b ecau se the formal symbol m anipulations by them selves d on 't

h ave any intentionality; they are quite m eaningless; they aren 't
even sym bol m anipulations, since the sym bols d o n 't symbolize
anything. In the linguistic jargon, they have only a syn tax but no
sem antics. Such intentionality as com p u ters app ear to h ave is
solely in the m inds of those w ho p rogram them , those w ho send
in the input and those w ho interpret the output, (p. 422)
The set of com m entaries published in the sam e issue provid es m any
useful and interesting criticisms of Searle's argum ent and of his con
clusions as well. The critical consensus is roughly as follows. If the
system of the room plus contents w ere u p graded so that its con versa
tional skills extend beyond a handful of topics to include the entire
range of topics a norm al hum an could be expected to know ; and if the
system w ere supplied w ith the sam e inductive capacities w e enjoy;
and if the "belief sto re" w ere integrated in the norm al fashion with
som e appropriately com plex goal stru cture; and if the room w ere
causally con nected to a body in such a fashion that its inputs reflected
appropriate sensory discrim inations and its outputs p rodu ced ap p ro
priate b ehavior; then the system of the room plus con ten ts jolly well
w ould u n derstand C hinese, and its various com putational states—
beliefs that p, desires that q— w ould indeed have m eaning and in
tentionality in the sam e w ay as with a norm al C hinese speaker.
Searle is quite willing to consider upgradings of the kind
described— he attem pts to anticipate them in his p ap er— but he is
convinced they change nothing relevant to his case. A s it em erges
clearly in his A u th or's R esponse (pp. 4 5 0 -4 5 6 ), of central im portance
to his argu m en t is the distinction betw een
cases of w hat I will call intrinsic intentionality, w hich are cases of

actual m ental states, and w hat I will call observer-relative ascrip
tions of intentionality, w hich are w ays that people have of speak

ing about entities figuring in our activities but lacking intrinsic
intentionality. [The latter] are alw ays d ep en den t on the intrinsic
intentionality of the observers, (pp. 4 5 1 -4 5 2 )
Exam ples of the latter w ould be the w ord s and sentences of on e's
native tongue. These have m eaning and intentionality, allows Searle,
but only insofar as they bear certain relations to our beliefs, thoughts,
and intentions— states with intrinsic intentionality. A sim ulation of
h u m an m entality grounded in a formal p rogram m ay yield states h av
ing this derivative observer-relative brand of intentionality, concedes
Searle, but they cannot have intrinsic intentionality. A nd since they
lack a feature essential to genuine m ental states, they cannot be
genuine m ental states, and to that extent the simulation m ust be a
failure.
A s w e see it, this criticism of functionalism is profoundly in error. It
is a m istake to try to m eet it, how ever, by continuing w ith the
strategy of trying to upgrade the im agined simulation in h op es of
finally w inning Searle's concession that at last its states have achieved
intrinsic intentionality. The correct strategy is to argue that our ow n
m ental states are just as innocent of "intrinsic intentionality" as are
the states of any m achine sim ulation. O n our view , all ascriptions
of m eaning or propositional content are relative (in senses to be
explained). The notion of "intrinsic intentionality" m akes no m ore
em pirical sense than does the notion of position in absolute space.
W e shall try to explain these claims.
There are basically just tw o w ays in w hich one can assign p roposi
tional content to the representational states of an oth er organism . An
exam ple of the first is the translation of a foreign language. A n exam
ple of the second is the calibration of an instrum ent of m easurem en t
or detection.
In the case of translation, one assigns specific propositional co n
tents to the alien representations because one has found a general
m apping betw een the alien rep resentation s and our ow n such that
the netw ork of formal and m aterial inferences holding am ong the
alien representations closely m irrors the sam e netw ork holding
am ong our ow n representations. Briefly, their collected rep resen ta
tions display an intensional structure that m atch es the intensional
stru cture displayed by our ow n.
The story is essentially the sam e w hen w e are assigning proposi
tional con tent to an alien's thou gh ts, beliefs, etc. It m atters n aught
w h eth er the alien's representation is overt, as with a sentence, or
covert, as w ith a belief. W e assign a specific content, p, to one of the
alien's rep resentation s on the strength of w hatever assu ran ces w e

have that his rep resentation plays the sam e abstract inferential role in
his intellectual (com putational) econom y that the belief-that-p plays
in ours. A nd w hat goes for aliens goes also for on e's b roth ers and
sisters.
This is not to say that the representational states of oth er hum ans
have con ten t only insofar as others interpret them in som e w ay. After
all, the set of abstract relations holding am ong the rep resentations in
som eon e's intellectual econom y is an objective feature intrinsic to that
person. But it does m ean that the content, call it the translational con
tent, of any specific representation of his is a m atter o f the in feren tial/
com putational relations it bears to all the rest o f his representations. There
can be no question of an isolated state or token possessing an intrinsic
translational con ten t; it will have a specific translational con ten t only
if, and only insofar as, it enjoys a specific set of relations to the other
elem ents in a system of representations.
C on trast translational content with w hat is naturally term ed calibra-
tional content. The repeatable states of certain physical system s are
m ore or less reliable indicators of certain features of their environ
m ent, and w e m ay assign con ten t (e .g ., "T h e tem p eratu re is 0° C ") to
such states (e .g ., a certain height in a colum n of red alcohol) on the
strength of such em pirical connections. This goes for the h u m an sys
tem as w ell. The various states w e call "p ercep tu al beliefs" can be
assigned co n ten ts in this sam e m an n er: as a function of w h at environ
m ental circu m stan ce standardly triggers their occu rrence. In fact, if a
system h as any system atic responses to its environm ent at all, then
calibration can take place even w here translation can n ot, either be
cause the system simply lacks the internal econom y n ecessary for
translational con ten t or because the intensional structure of that eco
nom y is incom m ensurable with our ow n. Furtherm ore, calibrational
content m ay regularly diverge from translational content, even w here
translational con tent is possible. C onsider a tribal u tterance that cali
brates as "T h ere is th u n d e r," but w hich translates as "G o d is shout
in g "; or one that calibrates as "T his m an has a bacterial in fectio n ,"
but w hich translates as "T h is m an is possessed by a pink d e m o n ."
A ccordingly, Searle is right to resist the suggestion th at m erely
hooking up the room system , via som e sensors, to the outside w orld
w ould supply a unique m eaning or content to the room 's rep resen ta
tional states. G enuine m ental states do indeed have a co n ten t or in
tentionality that is independent of, and possibly quite different from ,
their straightforw ard calibrational content. That ind ep en d en t inten
tionality is their translational content.
But this content falls well short of being the "intrinsic intentional
ity " Searle im agines our states to have. Translational con ten t is not
environm entally determ ined, nor is it observer-relative, but it is m ost

certainly a relational m atter, a m atter of the state's inferential/com
putational relations within a system of other such states. Accordingly,
it is entirely possible for translational con ten t to be possessed by the
states of a m achin e— as the realization of a purely formal program .
W h at m ore than this Searle im agines as fixing the content of our
m ental states, w e are unable to surm ise. H e floats the distinction be
tw een intrinsic intentionality and other kinds of intentionality by
m ean s of illustrative exam ples only (p. 451); he hazard s no palpable
accou nt of w h at intrinsic intentionality consists in; and the intuitions
to w hich he appeals can be explained in less m ysterious w ays, as
outlined above. To conclude, there simply is no such thing as intrinsic
intentionality, at least as Searle conceives it. Functionalists n eed not
be concern ed , then, that com pu ter sim ulations of h um an m entality
fail to display it.
W e com plete this section by und erscorin g a contrast. In the first
half of this section w e conceded to the critic of functionalism that our
m ental states have qualia, but w e argued th at the states of a m achine
sim ulation could have them as well. In this second half w e have con
ceded to the critic of functionalism that the states of certain m achine
sim ulations m u st lack intrinsic intentionality. But w e insist that our
ow n states are devoid of it as well.
3 Functionalism and M ethodology
D espite the defenses offered above, w e do wish to direct certain criti

cisms against functionalism. The criticisms are mainly methodological
rath er than substantive, h ow ever, and w e shall here provide only a
brief su m m ary, since they have been explained at length elsew here.
A . Conceptual conservatism
N o functionalist will suppose that the functional organization reco g
nized in the collected lore of folk psychology exh au sts the functional
intricacies that m ake up our internal econ om y. All will agree that folk
psychology rep resents only a partial, and in som e respects even a
superficial, grasp of the m ore com plex organization that em pirical
psychology will eventually unravel. E ven so, there is a decided
tend en cy on all sides to suppose that, so far as it goes, folk psych ol
ogy is essentially correct in the picture that it paints, at least in basic
outlines. Em pirical psychology will add to it, and explain its princi
ples, m any exp ect, but alm ost no one exp ects it to be o verth row n or
transm ogrified by such research.
This sanguine outlook is not unique to functionalists, b ut they are

especially vulnerable to it. Since the type identity of m ental states is
held to reside not in any shared physical or oth er natural essences,
but in the structure of their causal relations, there is a tend en cy to
construe the generalizations connecting them as collectively stipu lat
ing w hat it is to be a belief, a desire, a pain, and so forth. Folk psych ol
ogy is thu s rem oved from the arena of em pirical criticism . This is
unfortu n ate, since the "d en atu rin g " of folk psychology does not
change its epistem ic status as a speculative accou n t of o u r internal
w orkings. Like any other theory, it m ay be radically false, and like
any oth er deeply entrenched theory, its falsity is unlikely to be re
vealed w ith ou t a vigorous exploration of that possibility.
A functionalist can, of course, accept these points w ith ou t danger
to w hat is basic in his position. N evertheless, they are w orth m aking
for tw o reason s. First, eliminative m aterialism is not a v e ry w ide
spread opinion am ong adherents of functionalism , despite being en
tirely consistent w ith their view . A nd secon d , there are v ery good
reasons for doubting the integrity of folk psychology in its central
stru ctures as well as in its peripheral details (See P. M. C hurchland
1979, ch ap ter 5, and 1981a; P. S. C hurchland 1980a, 1980b; and Stich
1983).
B. Top dow n versus bottom up

Given that the essence of our psychological states resides in the set of
causal relations they bear to one anoth er, e tc., and given that this
abstract functional organization can be realized in a nom ically h eter
ogen eou s variety of substrates, it is fair en ou gh that the functionalist
should be m ore interested in that abstract organization than in the
m achin ery that realizes it. W ith the science of psychology, it is u n d er
standing the "p ro g ra m " that counts; an understan din g of su ch hard
w ares as m ay execu te it is secondary and inessential.
This m u ch is fair en ou gh , but so long as w e are so profoundly
ignorant of our functional organization as w e are at p resen t, and
ignorant of w here to draw the line betw een "h a rd w a re " an d "p ro
g ra m " in organism s, w e cannot afford to be so indifferent to the
n eu rosciences as the preceding rationale m ight suggest. If w e wish to
unravel the functional intricacies of our internal econ om y, one ob
vious w ay to go about this is to unravel the intricacies of the physical
system th at executes it. This bottom -up approach is n ot the only
ap proach w e m ight follow, b ut it does boast a num ber of ad van tages:
it is very strongly em pirical; it is not at all constrained by the p reco n
ceptions of folk psychology; it has the capacity to force surprises on
us; it perm its a nonbehavioral com parison of cognitive differences
across species; it enjoys direct connections w ith evolutionary etholo

gy; and at least in principle it can reveal the functional organization
w e are looking for.
N euroscience is an aw kw ard and difficult pursuit, h ow ever, and
there is an overw helm ing preference am ong philosophers, p sych o
logists, and artificial-intelligence research ers for a m ore top -dow n
ap p roach : hypothesize functional system s (program s) and test them
against our m olar behavior, as conceived w ithin com m on sense. This
is entirely legitim ate, but if the functionalist is m oved by the argu
m en t from abstraction to ignore or devalue the bottom -up approach,
his m ethodology is dangerously conservative and one-sided. (W e
have discussed these shortcom ings at length in P. S. C hurchland
1980a, 1980c; and in P. M. Churchland 1981a, 1982, 1980.)
C. Reductionism
Thanks to the argum ent from abstraction, functionalists tend to be
strongly antireductionist. They deny that there can be any general
characterization of w hat m akes som ething a thinker that is expressible
in the language of any of the physical sciences. Given the variety of
possible substrates (biological, chem ical, electrom echanical) that
could realize a thinking system , it is difficult n ot to agree w ith them .
But it does not follow, from multiple instantiability per se, that no
such general characterization is possible. It follows only that the re
quired characterization cannot be exp ressed in the theoretical v oca
bulary peculiar to any one of the available substrates. It rem ains en
tirely possible that there is a level of physical description sufficiently
abstract to en com pass all of them , and y et sufficiently pow erful to
fund the characterization required.
A s it h ap p ens, there is indeed a physical theory of sufficient g en er
ality to encom p ass the activity of all of these substrates, and any
o th ers one m ight think of. The theory is therm od yn am ics— the
general theory of energy and entropy. It has already supplied us with
a profoundly illuminating characterization of w h at the nineteenth
cen tu ry called "vital activity ," that is, of the ph en om en on of life. A nd
it is far from unthinkable that it m ight do the sam e for w hat this cen
tury calls "m en tal activity" (for a brief exploration of these ideas, see
P. M . C hurchland 1982). The theoretical articulation of such a ch arac
terization w ould be a very great achievem ent. It w ould be un fortu n
ate if the search for it w ere im peded by the general conviction that it is
im possible, a conviction born of the antireductionist urgings of a
false o rth od oxy am ong functionalists.
Chapter 3
Reduction, Qualia, and the Direct Introspection
of Brain States
Do the phenom enological or qualitative features of our sensations

constitute a p erm an en t barrier to the reductive aspirations of any
m aterialistic neuroscience? I here argue that they do not. Specifically,
I w ish to ad dress the recent antireductionist argum ents p osed by
Thom as N agel (1974), Frank Jackson (1982), and H ow ard Robinson
(1982). A n d I w ish to explore the possibility of hum an subjective co n
sciousness w ithin a conceptual environm ent constituted by a m a
tured and successful neuroscience.
If w e are to deal sensibly with the issues here at stake, w e m ust
ap p roach them w ith a general theory of scientific reduction already in
h and, a theory m otivated by and adequate to the m any instances and
varieties of interconceptual reduction displayed elsew here in our scien
tific history. W ith an independently grounded account of the nature
and grou n ds of intertheoretic reduction, w e can approach the specific
case of subjective qualia free from the m yopia that results from trying
to divine the p roper conditions on reduction by simply staring long
and hard at the problem atic case at issue.
1 Intertheoretic Reduction
W e m ay begin by rem arking that the classical accou n t of in terth eore

tic reduction (N agel 1961) now appears to be im portantly m istaken,
though the repairs n ecessary are quickly and cleanly m ad e. Sup pres
sing nicities, w e m ay state the original accou nt as follows. A new and
m ore com p reh en sive theory reduces an older theory just in case the
new th eory, w hen conjoined w ith appropriate corresp on d en ce rules,
logically entails the principles of the older theory. (The point of the
corresp on d en ce rules or "bridge law s" is to con n ect the disparate
ontologies of the tw o theories; often these are exp ressed as identity
statem en ts, such as Tem perature = m v2!3k.) Schem atically,
T h is e ssa y first ap p eared in th e Jou rn al o f P hilosophy 82, n o . 1 (Ja n u a ry 1985).

TN & (corresp on den ce rules)
logically entails
T o-
Difficulties w ith this view begin w ith the observation that m ost re
duced theories turn out to be, strictly speaking and in a variety of
resp ects, fa lse. (Real gases d on 't really obey P V = jxRT, as in classical
therm od ynam ics; the planets d on 't really m ove in ellipses, as in
Keplerian astron om y; the acceleration of falling bodies isn't really
uniform , as in Galilean dynam ics; etc.) If reduction is deduction, m od
us tollens w ould thus require that the prem ises of the new reducing
theories (statistical therm odynam ics in the first case, N ew tonian
dynam ics in the second and third) be som eh ow false as well, in co n
tradiction to their assum ed truth.
This com plaint can be tem porarily deflected by pointing out that
the prem ises of a reduction m ust often include, not just the new re
ducing theory, but also som e limiting assum ptions or counterfactual
b ou n d ary conditions (such as that the m olecules of a gas have only
m echanical energy, or that the m ass of the planets is negligible co m
p ared to the su n 's, or that the distance any body falls is negligibly
different from zero). Falsity in the reducing prem ises can thus be co n
ced ed , since it is safely confined to those limiting or counterfactual
assum ptions.
This defense will not deal w ith all cases of falsity, h ow ever, since in
som e cases the red u ced theory is so radically false that som e or all of
its ontology m u st be rejected entirely, and the "corresp on d en ce
ru les" connecting that ontology to the n ew er ontology therefore dis
play a problem atic status. N ew ly conceived features cannot be iden
tical w ith, nor even nom ically connected w ith, old features, if the old
features are illusory and uninstantiated. For exam ple, relativistic
m ass is not identical w ith N ew tonian m ass, n or even coextensive
w ith it, even at low velocities. N evertheless, the reduction of N ew to
nian by Einsteinian m echanics is a paradigm of a successful red u c
tion. For a second exam ple, neither is caloric-fluid-pressure identical
w ith, nor even coextensive w ith, m ean m olecular kinetic energy. But
an overtly flu id therm odynam ics (i.e., one com m itted to the existence
of "calo ric") still finds a m oderately im pressive reduction w ithin
statistical therm odyn am ics. In sum , even theories w ith a nonexistent
ontology can enjoy reduction, and this fact is problem atic on the
traditional account at issue.
C ases like these invite us to give up the idea that w hat gets d e
d uced in a reduction is the theory to be red u ced . A m ore accu rate,
The Direct Introspection of Brain States 49
general, and illuminating schem a for interth eoretic red u ction is as

follows:
r N & limiting assum p. & b oundary cond.
logically entails
IN (a set of theorem s of (restricted) TN),

e .g ., (x)(A x =з Bx),
{x)((Bx & Cx) =3 Dx),
w hich is relevantly isom orphic with
T0 (the older theory),

e .g ., (x )(Jx =>Kx),
(x)((K x & Lx) з M i).
That is to say, a reduction consists in the deduction, w ithin TN, not of

T0 itself, b u t rather of a roughly equipotent im age of T0 , an im age still
exp ressed in the vocabulary p roper to TN. The corresp on d en ce rules
play no part w h atever in the deduction. They show up only later, and
not necessarily as m aterial-m ode statem en ts, b ut as m ere ordered
pairs: <A x, Jx > , <Bx, K x> , <Cx, Lx>, <D x, M x> . Their function is to
indicate w hich term substitutions in the im age IN will yield the princi
ples of T0 - The older theory, accordingly, is never d ed u ced ; it is just
the target of a relevantly adequate m im icry. C on stru ed in this w ay, a
corresp on d en ce rule is entirely consistent w ith the assu m p tion that
the older predicate it en com p asses has no extension w h atev er. This
allows th at a true theory m ight reduce even a substantially false one.
The point of a reduction, according to this view , is to show that the
new or m ore com prehensive theory contains explanatory and predic
tive resou rces w hich parallel, to a relevant degree of exactn ess, the
exp lan atory and predictive resou rces of the reduced theory. The
intra-theoretic deduction (of fN w ithin TN) and the intertheoretic m ap
ping (of T0 into IN) jointly constitute a fell-sw oop dem onstration that
the older theory can be displaced w holesale by the new w ith ou t sig
nificant explanatory or predictive loss. (This sketch of intertheoretic
reduction is draw n from P. M. C hurchland 1979, section 11. For a
m ore detailed accoun t, see H ooker 1981.)
M aterial-m ode statem ents of identity can occasionally be m ade, of
course. W e do w ish to assert that visible light = electrom agnetic
w aves b etw een 0.35 /j.m and 0.75 д т , that sound = atm ospheric com
pression w aves, that tem peratu re = m ean m olecular kinetic energy,
and th at electric current = net m otion of charged particles. But a cor
respon d en ce rule does not itself m ake such a claim. A t b est, it records
the fact that the new predicate applies in all those cases w here its
50 The N ature of Mind
r 0 -dop p elgänger predicate w as norm ally thought to apply. O n this

view , full-fledged identity statem ents are licensed by the com parative
sm oothness of the relevant reduction (i.e ., the limiting assum ptions or
bou n dary conditions on TN are n ot wildly counterfactual, all or
m ost of T0 's principles find close analogues in IN, e tc.). This sm ooth
n ess perm its the com fortable assim ilation of the old ontology w ithin
the n ew , and thus allows the old theory to retain all or m ost of its
ontological integrity. It is smooth intertheoretic reductions that m otivate
and sustain statem ents o f cross-theoretic identity, not the other w ay around.
The preceding fram ew ork allows us to fram e a useful conception of
reduction for specific properties, as opp osed to entire theories, and it
allows us to fram e a useful conception of the contrary notion of
"e m e rg e n t" properties. A p roperty F, postulated by an older theory
or conceptual fram ew ork T0 , is reduced to a p roperty G in som e new
theory TN just in case
(1) TN reduces T0 ,
(2) 'F' and 'G' are corresp on d en ce-ru le paired in the red u c
tion, and
(3) the reduction is sufficiently sm ooth to sustain the
ontology of T0 and thus to sustain the identity claim
'F-n ess = G -ness'.
Intuitively, and in the m aterial m ode, this m ean s that F -n ess reduces
to G-ness just in case the causal pow ers of F-n ess (as outlined in the
law s of T0 ) are a subset of the causal p ow ers of G-ness (as outlined in
the law s of TN).
Finally, a p roperty F will be said to be an em ergent p roperty (relative
to TN) just in case
(1) F is definitely real and instantiated,

(2) F is cooccu rrent with som e feature or com plex circum stance
recognized in TN, but
(3) F cann ot be reduced to any p roperty postulated by or defin
able w ithin TN.
Intuitively, this will happen w hen TN does n ot have the resou rces
adequate to define a property w ith all of the causal pow ers possessed
by F-n ess. Claims about the em ergen ce of certain properties are th ere
fore claims about the relative p overty in the resou rces of certain aspir
ant theories. H aving outlined these notions, w e shall turn to address
substantive questions of em ergence and irreducibility in a few m o
m ents.
(A w ord of caution is p erhaps in order h ere, since the expression

'em ergen t property' is often used in tw o diam etrically opposed
senses. In scientific con texts, one frequently h ears it used to apply to
w hat m igh t be called a “ netw ork p ro p e rty ," a property that appears
exactly w hen the elem ents of som e substrate are suitably organized, a
p rop erty that consists in the elem ents of that substrate standing in
certain relations to one another, a set of relations that collectively sus
tain the set of causal pow ers ascribed to the "e m e rg e n t" property. In
this innocent sense of 'em ergen t', there are a great m an y em ergent
properties, and quite probably the qualia of our sensations should be
num bered am ong them . But in philosophical contexts one m ore often
en cou nters a different sense of 'em ergen t', one that implies that an
em ergen t p roperty does not consist in any collective or organizational
feature of its substrate. The first sense positively implies reducibility;
the secon d implies frreducibility. It is em ergen ce in the second sense
that is at issue in this p ap er.)
Before w e continue, several points about reduction n eed to be
em p hasized. The first is that in arguing for the em ergen ce of a given
p ro p erty F relative to som e theory TN, it is not sufficient to point out
that the existence or appearance of F -ness cannot be d educed from
TN. It is occasionally claim ed, for exam ple, that the objective features
of w arm th or blueness m u st be irreducibly em ergen t properties, since
h ow ever m u ch one bends and squeezes the m olecular theory co n
cerning H 20 , one cannot deduce from it that w ater will be blue, but
only th at w ater will scatter electrom agnetic radiation at such and such
w avelen gth s. A nd how ever m uch one w rings from the m echanics of
m olecular m otion, one cannot deduce from it that a roaring hearth
will be w arm , but only that its m olecules will have such and such a
m ean kinetic energy and will collectively em it electrom agnetic radia
tion at longish w avelengths.
These prem ises about nondeducibility are entirely tru e, but the
conclusion against reducibility does not follow. It is a serious m istake
to m ake even indirect deducibility (i.e ., deducibility w ith the help of
corresp on d en ce rules) a requirem ent on successful reduction, as w e
saw at the beginning of this section. A nd there are additional reasons
w hy it w ould be even m ore foolish to insist on the m u ch stronger
condition of direct deducibility. For exam ple, formal considerations
alone g uarantee that, for any predicate 'F ' not already in the p ro
prietary lexicon of the aspirant reducing theory TN, no statem ents
w h atever involving 'F ' (beyond tautologies and other trivial excep
tions) will be deducible from TN. The deducibility requirem ent w ould
thus trivialize the notion of reduction by m aking it im possible for
any con ceptual fram ew ork to reduce any oth er, distinct conceptual
fram ew ork. Even tem p eratu re, that paradigm of a successfully
red u ced property, w ould be rend ered irreducible, since the term
'tem p eratu re' does not app ear in the lexicon of statistical m echanics.
There is a further reason w hy the dem an d for direct deducibility is
too strong. The fact is, it is an historical accident that w e hum ans
currently use precisely the conceptual fram ew ork w e do use. W e
m ight have used any one of an infinite num ber of oth er conceptual
fram ew orks to describe the observable w orld, each one of w hich
could have been roughly adequate to com m on experience, and m any
of w hich w ould be roughly isom orphic (each in its different w ay) with
som e p art of the correct accou n t that a utopian theory will eventually
provide. A ccordingly, w e can legitim ately ask of a putatively correct
theory of a given objective dom ain that it accou n t for the p h enom ena
in (that is, function successfully in) that dom ain. But w e cannot insist
that it also be able to predict how this, th at, or the other conceptually
idiosyncratic h um an culture is going to conceive of that dom ain. That
w ould be to insist that the new theory do predictive cultural an thropolo
gy for u s, as well as m echanics, or electrom agnetic theory, or w hat
have you . The dem and that m olecular theory directly entail our ther
m al or color con cep ts is evidently this sam e unreasonable dem and.
All w e can properly ask of a reducing theory is that it have the
resou rces to conjure up a set of properties w hose nom ological
pow ers/roles/features are system atic analogues of the pow ers/roles/
features of the set of properties postulated by the old theory. Since
both theories presu m e to describe the sam e em pirical dom ain, these
system atic nom ological parallels constitute the best grounds there can
be for concluding that both theories have m anaged to latch onto the
sam e set of objective properties. The h yp othesized identity of the
properties at issue explains w hy and T0 are taxonom ically and
nom ically parallel: they are both at least partially correct accou n ts of
the very sam e objective properties. ZN m erely fram es that account
w ithin a m u ch m ore penetrating conceptual system — that of TN.
M oreover, it is to be expected that existing conceptual fram ew orks
will eventually be reduced or displaced by new and better ones, and
those in turn by fram ew orks better still, for w ho will be so brash as to
assert that the feeble conceptual achievem ents of our adolescent sp e
cies com prise an exhaustive accou n t of anything at all? If w e p ut aside
this conceit, then the only alternatives to intertheoretic reduction are
epistem ic stagnation or the outright elim ination of old fram ew orks as
w holly false and illusory.
2 T heoretical Change and P erceptual Change
Esoteric properties and arcane theoretical fram ew orks are not the
only things that occasionally enjoy intertheoretic reduction. O bserv
able properties and com m onsense conceptual fram ew orks can also
enjoy sm ooth reduction. Thus, being a m iddle-A sound is identical
w ith being an oscillation in air pressu re at 440 hertz; being red is
identical w ith having a certain triplet of electrom agnetic reflectance
efficiencies; being w arm is identical with having a certain m ean level
of m icroscopically em bodied energies, and so forth.
M oreover, the relevant reducing theory is capable of replacing the
old fram ew ork n ot just in contexts of calculation and inference. It
should be appreciated that the reducing theory can displace the old fram ew ork
in all o f its observational contexts as w ell. Given the reality of the p rop
erty identities just listed, it is quite open to us to begin fram ing our
sp on tan eou s perceptual reports in the language of the m ore sophisti
cated redu cin g theory. It is even desirable that w e begin doing this,
since th e new vocabulary observes distinctions w hich are in fact w ith
in the discrim inatory reach of our native perceptual system s, though
those objective distinctions go unm arked and unnoticed from w ithin
the old fram ew ork. W e can thus m ake m ore penetrating use of our
native perceptual equipm ent. Such displacem ent is also desirable for
a secon d reason : the greater inferential or com putational p ow er of the
n ew conceptual fram ew ork. W e can thus m ake better inferential use
of our new perceptual judgm ents than w e m ade of our old ones.
It is difficult to convey in w ords the enorm ity of such perceptual
transform ations and the naturalness of the new con cep tu al regim e
once established. A nonscientific exam ple m ay help to get the initial
point across.
C on sid er the en orm ou s increase in discrim inatory skill that spans
the gap betw een an untrained child's auditory ap prehension of a
sym p hon y, and the sam e p erson's apprehension of the sam e sym
p h on y forty years later, w hen hearing it in his capacity as con d u ctor
of the o rch estra perform ing it. W hat w as before a seam less voice is
now a m osaic of distinguishable elem ents. W hat w as before a dimly
ap p reh en d ed tune is now a rationally stru ctu red sequence of distin
guishable and identifiable chords supporting an appropriately related
m elody line. The m atured m usician h ears an entire w orld of stru c
tured detail, concerning w hich the child is both dum b and deaf.
O th er m odalities provide com parable exam ples. C on sid er the p rac
ticed and chem ically sophisticated wine taster, for w hom the category
"re d w in e " u sed by m ost of us divides into a netw ork of fifteen or
tw enty distinguishable elem ents: ethanol, glycol, fru ctose, sucrose,
tannin, acid, carbon dioxide, and so forth, w hose relative con cen tra
tions h e can estim ate w ith accuracy. O r consider the astron om er, for
w h om the speckled black dom e of h er you th has b ecom e a visible
abyss, scattering nearby planets, yellow dw arf stars, blue and red
giants, distant globular clusters, and even a rem ote galaxy or tw o, all
discriminable as such and locatable in three-dim ensional space w ith
h er unaided (repeat: unaided) eye.
In each of these cases, w hat is finally m astered is a conceptual
fram ew ork— w hether m usical, chem ical, or astronom ical— a fram e
w ork that em bodies far m ore w isdom about the relevant sensory
dom ain than is im m ediately app aren t to untu tored discrim ination.
Such fram ew orks are characteristically a cultural heritage, pieced
togeth er over m any generations, and their m astery supplies a rich
n ess and penetration to our sen sory lives that w ould be impossible
in their absence. (The role of theory in perception and the sys
tem atic en h an cem en t of perception through theoretical p rogress are
exam ined at length in P. M. C hurchland 1979, sections 1 throu gh 6.)
O ur introspective lives are already the extensive beneficiaries of this
p h en om en on. The introspective discrim inations w e m ake are for the
m ost part learned; they are acquired w ith practice and experience,
often quite slowly. A nd the specific discrim inations w e learn to m ake
are those it is useful for us to m ake. G enerally, those are the dis
crim inations that others are already m aking, the discrim inations em
bodied in the psychological vocabulary of the language w e learn. The
conceptual fram ew ork for psychological states that is em bedded in
ordinary language is a m odestly sophisticated theoretical achieve
m en t in its ow n right, and it shapes our m atu red introspection p ro
foundly. If it em bodied substantially less w isdom in its categories and
connecting generalizations, our introspective apprehension of our in
ternal states and activities w ould be m u ch dim inished, though our
native discrim inatory m echanism s rem ain the sam e. Correlatively, if
folk psychology em bodied substantially m ore w isdom about our inner
nature than it actually does, our introspective discrim ination and
recognition could be very m u ch greater than it is, though our native
discrim inatory m echanism s rem ain u nch an ged .
This brings m e to the central positive suggestion of this p ap er. C on
sider now the possibility of learning to describe, conceive, and intro-
spectively apprehend the teem ing intricacies of our inner lives w ithin
the conceptual fram ew ork of a m atured n euroscience, a neuroscience
th at successfully reduces, either sm oothly or roughly, our com m on
sense folk p sychology. Suppose w e trained our native m echanism s to
m ake a new and m ore detailed set of discrim inations, a set that co r
respon d ed n ot to the primitive psychological taxon om y of ordinary
language, but to som e m ore penetrating taxon om y of states draw n
from a com pleted neuroscience. A n d sup p ose w e trained ourselves to
respon d to that reconfigured discrim inative activity w ith jud gm en ts
that w ere fram ed, as a m atter of course, in the appropriate con cep ts
from n euroscien ce. (I believe it w as Paul K. Feyerabend and Richard

Rorty w ho first identified and explored this suggestion. See
Feyerabend 1963a and Rorty 1965. This occurred in a theoretical en
vironm ent prep ared largely by Sellars 1956. The idea h as b een ex
plored m ore recently in P. M. C hurchland 1979 and in ch apter 1
above.)
If the exam ples of the sym phony con d uctor (w ho can h ear the A m 7
ch ord s), the enologist (w ho can see and taste the glycol), and the
astron om er (w ho can see the tem perature of a blue giant star) provide
a fair parallel, then the en h an cem ent in our introspective vision could
approxim ate a revelation. D opam ine levels in the limbic system , the
spiking frequencies in specific neural p ath w ays, reson an ces in the nth
layer of the occipital cortex, inhibitory feedback to the lateral genicu
late n ucleus, and countless other neurophysical nicities could be
m oved into the objective focus of our introspective discrim ination,
just as G m 7 chords and A dim chords are m oved into the objective
focus of a trained m usician's auditory discrim ination. W e will of
cou rse have to learn the conceptual fram ew ork of a m atu red n eu ro
science in ord er to pull this off. A nd w e will have to practice its non-
inf erential application. But that seem s a small price to p ay for the
q uan tu m leap in self-apprehension.
All of this suggests that there is no problem at all in conceiving the
eventu al reduction of m ental states and properties to neurophysio-
logical states and properties. A matured and successful neuroscience
need only include, or prove able to define, a taxon om y of kinds w ith a
set of em bedding law s that faithfully mim ics the taxon om y and causal
generalizations of fo lk psychology. W h eth er future neuroscientific
theories will prove able to do this is a w holly empirical question, not
to be settled a priori. The evidence for a positive answ er is substantial
and familiar, and it cen ters on the grow ing explanatory su ccess of the
several n eurosciences.
But there is negative evidence as well; I have even u rged som e of it
m yself (1981a). M y negative argum ents there center on the explana
tory and predictive p overty of folk psychology, and they question
w h eth er it h as the categorial integrity to m erit the reductive p reserva
tion of its familiar ontology. That line suggests substantial revision or
outright elim ination as the eventual fate of our m entalistic ontology.
The qualia-based argu m en ts of N agel, Jackson, an d Robinson,
h ow ever, take a quite different line. They find no fault w ith folk
psych ology. Their con cern is w ith the explanatory and descriptive
pov erty of any possible neuroscience, and their line su ggests that
em ergen ce is the correct story for our m entalistic ontology. Let us
now exam ine their argum ents.
3 Thom as N agel's A rgum ents
F or Thom as N agel, it is the phenom enological features of our experi

en ces, the properties or qualia displayed b y our sensations, that con
stitute a problem for the reductive aspirations of any m aterialistic
n euroscience. In his classic position p aper (1974), I find three distinct
argu m ents in support of the view that such properties will never find
any plausible or adequate reduction w ithin the fram ew ork of a
m atu red neuroscience. All three argum ents are beguiling, b u t all
three, I shall argue, are unsound.
The fir s t argum ent

W hat m akes the proposed reduction of m ental p h en om en a different
from reductions elsew here in science, says N agel, is that
it is impossible to exclude the phenom enological features of ex

perience from a reduction, in the sam e w ay that one excludes the
p hen om en al features of an ordinary substance from a physical or
chem ical reduction of it— nam ely, by explaining them as effects
on the m inds of hum an observers. (1974, p. 437)
The reason it is impossible to exclude them , continues N agel, is that

the phenom enological features are essential to experience, and to the
subjective point of view . But this is not w h at interests m e about this
argu m en t. W hat interests m e is the claim that reductions of various
substances elsew here in science exclude the phen om enal fea tu res o f the
substance.
This is simply false, and the point is extrem ely im portant. The
phen om en al features at issue are those such as the objective red n ess
of an apple, the w arm th of a coffee cup, and the pitch of a sound.
These properties are not excluded from o u r reductions. R edness, an
objective p henom enal property of apples, is identical w ith a cer
tain w avelength triplet of electrom agnetic reflectance efficiencies.
W arm th , an objective p henom enal p roperty of objects, is identical
w ith the m ean level of the objects' m icroscopically em bodied en er
gies. Pitch, an objective phenom enal property of a sound, is identical
w ith its oscillatory frequency. These electrom agnetic and m icro
m echanical properties, out there in the objective w orld, are genuine
phen om en al properties. D espite w idespread ignorance of their d y
nam ical and m icrophysical details, it is these objective physical
p roperties to w hich everyon e's perceptual m echanism s are keyed.
The reductions w hose existence N agel denies are in fact so co m
plete that one can already displace entirely large chunks of our com
m on sen se vocabulary for observable properties, and learn to fram e
on e's p ercep tu al judgem ents directly in term s of the reducing theory.

The m ean kinetic energy (KE) of the m olecules in this room , for exam
ple, is currently about 6 .2 x 10” 21 joules. The oscillatory frequency of
this sound (I here w histle C one octave above m iddle C) is about 524
hertz. A n d the three critical electrom agnetic reflectance efficiencies
(at 0 .4 5 , 0.53, and 0.63 /xm) of this w hite piece of p aper are all above
80 p ercen t. These m icrophysical and electrom agnetic p roperties can
be felt, h eard , and seen , respectively. O ur native sen sory m ech an
ism s can easily discrim inate such properties, one from an oth er, and
their p resen ce from their absence. They have been doing so for mil
lennia. The "reso lu tio n " of these m echanism s is inadequate, of
cou rse, to reveal the m icrophysical details and the extend ed causal
roles of the properties thus discrim inated. But they are abundantly
adequate to perm it the reliable discrim ination of the properties at
issue. (See m y 1979, sections 2 through 6. See also Paul and Patricia
C h urchland 1981b, pp. 1 2 8 -1 3 0 [this volum e, chapter 2, p p . 3 0 -3 1 ].)
On this view, the standard perceptual properties are not "second
a ry " properties at all, in the standard sense that implies that they
have no real existence save inside the m inds of h um an observers. On
the co n trary, they are as objective as you please, w ith a w ide variety
of objective causal properties. M oreover, it w ould be a m istake even
to try to "kick the phenom enal properties in w a rd s," since that only
p ostp o n es the problem of reckoning their place in n atu re. W e shall
only con fron t them again later as w e address the place in nature of
m ental p hen om en a. A n d as N agel correctly points out, the relocation
dodge is no longer open to us once the problem atic p roperties are
already located within the m ind.
N agel concludes from this that subjective qualia are unique in being
im m une from the sort of reductions found elsew here in science. I
draw a very different conclusion. The objective qualia (redness,
w arm th , etc.) should n ever have been "kicked inw ards to the m inds
of o b serv ers" in the first place. They should be confronted squarely,
and th ey should be red u ced w here they stand: outside the hum an
observer. A s w e saw , this can and has in fact b een don e. If objective
p h en om en al properties are so treated, then subjective qualia can be
confronted w ith parallel forthrightness, and can be red u ced w here
they stan d : mside the h um an observer. So far then, the external and
the internal cases are not different: they are parallel after all.
The secon d argum ent

A secon d argu m en t urges the point that the intrinsic character of
experiences, the qualia of sensations, are essentially accessible from
only a single point of view , the subjective point of view of the ex
periencing subject. The properties of physical brain states, by co n
trast, are accessible from a variety of entirely objective points of view .
W e can n ot hope adequately to account for the form er, therefore, in
term s of properties appropriate to the latter dom ain, (see N agel 1974,
p p . 4 4 2 -4 4 4 .)
This som ew h at diffuse argum ent app ears to be an instance of the
following argum ent.
(1) The qualia of m y sensations are directly know n by m e, by

introspection, as elem ents of m y conscious self.
(2) The properties of m y brain states are not directly know n by
m e, by introspection, as elem ents of m y conscious self.
. ' . (3) The qualia of m y sensations Ф the properties of m y brain
states.
A n d perhap s there is a second argum en t here as well, a com plem ent

to the first:
(1) The properties of m y brain states are know n-by-the-

variou s-external-sen ses, as having such and such physical
properties.
(2) The qualia of m y sensations are n ot know n-by-the-
variou s-external-sen ses, as having such and such physical
properties.
. ' . (3) The qualia of m y sensations Ф the properties of m y brain
states.
The argu m ent form here is apparently
(1) Fa
(2) ~Fb
(3) а Ф b.
G iven Leibniz's Law and the extensional nature of the p roperty F,

this is a valid argum ent form . But in the exam ples at issue, F is
obviously not an extensional property. The fallacy com m itted in both
cases is am ply illustrated in the following parallel argum ents.
(1) H itler is w idely recognized as a m ass m urderer.

(2) Adolf Schicklgruber is not widely recognized as a m ass
m urderer.
(3) H itler Ф Adolf Schicklgruber.
(1) A spirin is know n by John to be a pain reliever.

(2) Acetylsalicylic acid is not know n by John to be a pain
reliever.
. ' . (3) A spirin Ф acetylsalicylic acid,
or, to cite an exam ple v ery close to the case at issue,
(1) Tem peratu re is know n by m e, by tactile sensing, as a feature

of m aterial objects.
(2) M ean m olecular kinetic en ergy is not know n by m e, by
tactile sensing, as a feature of m aterial objects.
. ' . (3) T em perature Ф m ean m olecular kinetic energy.
The problem w ith all of these argum ents is that the "p ro p e rty "
ascribed in prem ise (1) and witheld in prem ise (2) consists only in the
subject item 's being recognized, perceived, or know n as som eth in g, under
som e specific description or other. Such apprehension is n ot a genuine
feature of the item itself, fit for divining identities, since one and the
sam e subject m ay be successfully recognized und er one description
(e .g ., 'qualia of m y m ental state'), and yet fail to be recognized under
an oth er, equally accurate, coreferential description (e .g ., 'prop erty
of m y brain state'). In logician's term s, the propositional function, ‘x
is know n (perceived, recognized) by m e, as an F' is one of a large
num ber of intensional contexts w hose distinguishing feature is that
they do not alw ays retain the sam e truth value through substitution
of a coreferential or coextensive term for w hatever holds the place of
'x'. A ccordingly, that such a context (i.e ., the one at issue) should
show a difference in truth value for tw o term s 'a! an d 'V (i.e.,
'qualia of m y sensations' and 'property of m y b rainstates') is there
fore hard ly grou n d s for concluding that 'a' and 'V can n ot be core
ferential or coextensive term s! (I believe it w as Richard Brandt and
Jaegw on Kim (1967) w ho first identified this fallacy specifically in con
nection w ith the identity theory.)
This objection is decisive, I think, but it does not apply to a differ
ent version of the argum en t, w hich w e m u st also consider. It m ay be
u rged that one's brain states are m ore than m erely not (yet) know n by
introsp ection : they are not know able by introspection u n d er any cir
cu m stan ces. In corresp on dence, Thom as N agel has advised m e that
w hat he w ishes to defend is the following m odalized version of the
argu m en t.
(1) M y m ental states are knowable by m e by introspection.

(2) M y brain states are not know able by m e by introspection.

. ' . (3) M y m ental states + m y brain states.
H ere N agel will insist that being know able by m e by introspection is a

genuine relational p roperty of a thing, and that this version of the
argu m en t is free of the intensional fallacy discussed above.
A n d so it is. But now the reductionist is in a position to insist that
the argum en t contains a false prem ise: prem ise (2). A t the very least,
he can insist that (2) begs the question. For if m ental states are indeed
identical w ith brain states, then it is really brain states that w e have
b een introspecting all along, though w ithout appreciating their fine
grained nature. A nd if w e can learn to think of and recognize those
states u n d er their familiar m entalistic descriptions— as all o f us have—
then w e can certainly learn to think of and recognize them un d er
their m ore penetrating neurophysiological descriptions. Brain states,
that is, are indeed k n o w a b le by introspection, and N agel's argum ent
com m its the sam e error instanced below .
(1) T em perature is know able by tactile sensing.

(2) M ean m olecular kinetic energy is not know able by tactile
sensing.
. ‘ . (3) T em perature Ф m ean m olecular kinetic energy.
H ere the conclusion is know n to be false. Tem perature is indeed

m ean m olecular kinetic energy. Since the argum ent is valid, it m ust
therefore have a false prem ise. Prem ise (2) is clearly the stinker. Just
as one can learn to feel that the su m m er air is about 70°F, or 21°C , so
one can learn to feel that the m ean KE of its m olecules is about
6 .2 X 1 0 -21 joules, for w hether w e realize it or not, that is the p roperty
o u r native discrim inatory m echanism s are keyed to. A n d if one can
com e to know , by feeling, the m ean KE of atm ospheric m olecules,
w h y is it unthinkable that one m ight com e to know , by introspection,
the states of o n e's brain? (W hat w ould th at feel like? It w ould feel
exactly the sam e as introspecting the states of o n e's m ind, since they
are one and the sam e states. O ne w ould sim ply em ploy a different
and m ore penetratin g conceptual fram ew ork in their description.)
O ne m u st be careful, in evaluating the plausibility of N agel's
second prem ise, to distinguish it from the second prem ise of the very
first version of the argum ent, the version that com m its the intension
al fallacy. My guess is that N agel has profited som ew hat from the
am biguity h ere. For in the first version, b oth prem ises are true. A nd
in the second version, the argu m ent is valid. N either version, h ow
ever, m eets both conditions.
The m atter of introspecting on e's brain states will arise once m ore
in the final section of this paper. For now , let us m ove on.
The third argum ent

The last argu m en t here is the one m ost w idely associated w ith
N agel's pap er. The leading exam ple is the (m ooted) ch aracter of the
experiences enjoyed by an alien creature such as a bat. The claim is
that, no m atter how m uch one knew about the b at's neurophysiology
and its interaction w ith the physical w orld, one could still n ot know ,
nor p erh ap s even im agine, w hat it is like to be a bat. Even total know l
edge of the physical details still leaves som ething out. The lesson
d raw n is that the reductive aspirations of neurophysiology are
doom ed to dash them selves, unrealized, against the im penetrable
keep of subjective qualia. (see N agel 1974, pp. 438ff.)
This argu m en t is alm ost identical to an argu m en t p ut forw ard in a
recent p ap er by Frank Jackson (1982). Since Jackson's version deals
directly w ith h um an s, I shall confront the problem as h e form ulates
it.
4 Jackson's K n ow ledge A rgum ent
Im agine a brilliant neuroscientist nam ed M ary w ho has lived her en

tire life in a room that is rigorously controlled to display only various
shades of black, w hite, and grey. She learns about the outside w orld
by m ean s of a black/w hite television m onitor, and being brilliant, she
m anages to transcend these obstacles. She becom es the w orld's
greatest neuroscientist, all from w ithin this room . In particular, she
com es to know everything there is to know about the physical stru c
ture and activity of the brain and its visual system , of its actual and
possible states.
But there w ould still be som ething she did not know , and could not
even im agine, about the actual experiences of all the o th er people
w ho live outside her black/w hite room , and about her possible experi
ences w ere she finally to leave her room : the nature of the experience
of seeing a ripe tom ato, w hat it is like to see red or have a sensation-
of-red. Therefore, com plete know ledge of the physical facts of visual
percep tion and its related brain activity still leaves som ethin g out.
H ence, m aterialism cannot give an adequate reductionist accou n t of
all m ental ph en om en a.
To give a conveniently tightened version of this argu m en t,
(1) M ary know s everything there is to know about brain states

and their properties.
(2) It is not the case that M ary know s everything there is to

know about sensations and their properties.
Therefore, by Leibniz's Law ,
(3) Sensations and their properties 4= brain states and their

properties
It is tem pting to insist that w e here confront just an oth er instance of

the intensional fallacy discussed earlier, but Jackson 's defenders
(e .g ., Cam pbell 1983) insist that 'know s about' is a perfectly tran sp a
ren t, entirely extensional context. Let us sup p ose that it is. W e can, I
think, find at least tw o other shortcom ings in this sort of argum ent.
The fir s t shortcom ing

This defect is simplicity itself. 'K now s about' m ay be tran sp aren t in
both prem ises, but it is not univocal in both prem ises. (David Lewis
[1983] and Laurence N em irow [1980] have both raised this sam e
objection, though their analysis of the am biguity at issue differs from
m ine.) Jackson 's argu m en t is valid only if 'know s about' is univocal
in both prem ises. But the kind of know ledge ad d ressed in prem ise (1)
seem s pretty clearly to be different from the kind of know ledge
ad d ressed in (2). Know ledge in (1) seem s to be a m atter of having
m astered a set of sentences or propositions, the kind one finds w rit
ten in neuroscience texts; w hereas know ledge in (2) seem s to be a
m atter of having a representation of red n ess in som e prelinguistic or
sublinguistic m edium of rep resentation for sen sory variables, or to be
a m atter of being able to m ake certain sen sory discrim inations, or
som ething along these lines.
Lew is and N em irow plum p for the "ab ility" analysis of the relevant
sen se of 'know s about', but they need n ot be so narrow ly com m it
ted, and the com plaint of equivocation need not be so narrow ly
based . A s m y alternative gloss illustrates, oth er analyses of 'kn ow
ledge by acquaintance' are possible, and the charge of equivocation
will be sustained so long as the type of know ledge invoked in prem ise
(1) is distinct from the type invoked in prem ise (2). Im portantly, they
do seem very different, even in advance of a settled analysis of the
latter.
In short, the difference betw een a p erson w ho know s all about the
visual cortex but has never enjoyed a sensation of red, and a person
w ho know s no neuroscience but know s well the sensation of red,
m ay reside not in w hat is respectively know n by each (brain states by
the form er, qualia by the latter), but rath er in the different type of
know ledge each has of exactly the sam e thing. The difference is in the
m an n er of the know ing, n ot in the nature of the thing(s) know n. If

one rep laces the am biguous occurrences of 'know s about' in Jack
son 's arg u m en t with the tw o different expansions su ggested above,
the resulting argu m en t is a clear non sequitur.
(a) M ary has m astered the com plete set of true propositions
about p eople's brain states.
(b) M ary does not have a representation of red n ess in her pre-
linguistic m edium of representation for sen sory variables.
Therefore, by Leibniz's Law ,
(c) The red n ess sensation Ф any brain state.
Prem ises (a) and (b ) are com possible, even on a m aterialist view . But
they do n ot entail (c).
In su m , there are p retty clearly m ore w ays of "h av in g k n ow led ge"
than h avin g m astered a set of sentences. A nd nothing in m aterialism
p recludes this. The m aterialist can freely adm it that one h as "k n o w l
ed g e" of o n e's sensations in a w ay that is ind ep en d en t of the scien
tific theories one h as learned. This does n ot m ean that sensations are
beyond the reach of physical science. It ju s t m eans that the brain uses
more m odes and m edia o f representation than the sim ple storage o f sentences.
A nd this proposition is pretty obviously true: alm ost certainly the
brain u ses a considerable variety of m odes and m edia of rep resen ta
tion, p erh ap s h u n d red s of them . Jackson 's argu m en t, and N agel's,
exploit this variety illegitimately: both argu m en ts equivocate on
'know s ab ou t'.
This criticism is supported by the observation that, if Jackson's
form of arg u m en t w ere sound, it w ould prove far too m u ch . Suppose
that Jackson w ere arguing not against m aterialism , but against dual
ism : against the view that there exists a nonm aterial su b stan ce— call
it 'ecto p lasm '— w h ose hidden constitution and nom ic intricacies
grou n d all m ental p h en om en a. Let our cloistered M ary be an "e cto -
plasm ologist" this tim e, and let her kn ow ! (by description) every
thing th ere is to know about the ectoplasm ic processes underlying
vision. There w ould still be som ething she did not k now 2 (by ac
quaintance): w h at it is like to see red. D ualism is therefore inadequate
to accou n t for all m ental p henom ena.
This arg u m en t is as plausible as Jackson 's, and for the sam e reason :
it exploits the sam e equivocation. But the truth is, such argum ents
show nothin g, one w ay or the other, about how m ental p h en om en a
m ight be accou n ted for.
The second shortcom ing

There is a further shortcom ing w ith Jackson 's argum ent, one of p ro
found im portance for understan din g one of the m ost exciting co n
sequences to be expected from a successful neuroscientific accou n t of
m ind. I draw y ou r attention to the assum ption that even a utopian
know ledge of neuroscience m ust leave M ary hopelessly in the dark
about the subjective qualitative nature of sensations not yet enjoyed.
It is true, of course, that no sentence of the form 'x is a sensation-of-
red ' will be deducible from prem ises restricted to the language of
n euroscience. But this is no point against the reducibility of p h e
nom enological properties. A s w e saw in section 1, direct deducibility
is an intolerably strong dem and on reduction, and if this is all the
objection com es to, then there is no objection w orth addressing.
W h at the defender of em ergen t qualia m u st have in mind h ere, I
think, is the claim that M ary could not even im agine w hat the relevant
experience w ould be like, despite h er exhaustive neuroscientific
know ledge, and h en ce, that she m u st still be m issing certain crucial
inform ation.
This claim , how ever, is simply false. Given the truth of prem ise (1),
prem ise (2) seem s plausible to Jackson, N agel, and Robinson only
b ecau se none of these philosophers has adequately considered how
m u ch one m ight know if, as prem ise (1) asserts, one knew everything
there is to know about the physical brain and nervous system . In
particular, none of these philosophers has even begun to consider the
ch an ges in our introspective apprehension of our internal states that
could follow u p on a w holesale revision in our conceptual fram ew ork
for our internal states.
The fact is, w e can indeed im agine how neuroscientific inform ation
w ould give M ary detailed inform ation about the qualia of various
sensations. Recall our earlier discussion of the transform ation of
p ercep tion throu gh the system atic reconceptualization of the relevant
p ercep tu al dom ain. In particular, supp ose that M ary h as learned to
conceptualize her inner life, even in introspection, in term s of the
com pleted neuroscience w e are to im agine. So she does not identify
h er visual sensations crudely as 'a sensation-of-black', 'a sensation-
o f-grey', or 'a sensation-of-w hite'; but rath er identifies them m ore
revealingly as various spiking frequencies in the nth layer of the occi
pital cortex (or w hatever). If M ary has the relevant neuroscientific
concep ts for the sensational states at issue (nam ely, sensations-of-
red), but has never yet been in those states, she m ay well be able to
im agine being in the relevant cortical state, and im agine it w ith sub
stantial su ccess, even in advance of receiving external stimuli that
w ould actually produce it.
O ne test of h er ability in this regard w ould be to give h er a stim ulus

that w ould (finally) produce in her the relevant state (nam ely, a spik
ing frequency of 90 hertz in the gam m a netw ork: a "sen satio n -o f-red "
to us), and see if she can identify it correctly on introspective grounds
alone, as 'a spiking frequency of 90 h ertz, the kind a tom ato w ould
cau se'. It does not seem to m e to be im possible that she should
su cceed in this, and do so regularly on similar tests for oth er states,
conceptualized clearly by her, but n ot previously enjoyed.
This m ay seem to som e an outlandish suggestion, but the following
will show that it is not. M usical chords are auditory p h en om en a that
the y ou n g and unpracticed ear hears as undivided w holes, discrim in-
able on e from an other, but w ithout elem ents or internal stru ctu re. A
m usical education changes this, and one com es to h ear chords as
grou p s of discriminable n otes. If one is sufficiently practised to have
absolute pitch, one can even nam e the notes of an app reh en ded
chord. A n d the reverse is also true: if a set of notes is specified verbal
ly, a trained pianist or guitarist can identify the chord and recall its
sound in auditory im agination. M oreover, a really skilled individual
can co n stru ct, in auditory im agination, the sound of a ch ord he m ay
never h ave h eard before and certainly does not rem em b er. Specify for
him a relatively unusual o n e— an F # 9thaddl3th for exam p le— and let
him brood for a bit. Then play for him three or four ch ord s, one of
w hich is the target, and see if he can pick it out as the sound that
m eets the description. Skilled m usicians can do this. W h y is a similar
skill beyon d all possibility for M ary?
A h, it is tem pting to reply, m usicians can do this only because
chords are audibly stru ctu red sets of elem ents. Sensations-of-color
are not.
But n eith er did chords seem , initially, to be stru ctu red sets of ele
m ents. They also seem ed to be undifferentiated w holes. W h y should
it be unthinkable that sensations-of-color p ossess a com parable inter
nal stru ctu re, unnoticed so far, but aw aiting our determ ined and in
form ed inspection? Jack son 's argu m en t, to be successful, m ust rule
this possibility out, and it is difficult to see h ow he can do this a priori,
especially since there has recently em erged excellent em pirical evi
dence to su ggest that ou r sensations-of-color are indeed structured sets o f
elem ents.
The R etinex theory of color vision recently p roposed by Edw in
Land (1977) rep resents any color apprehendable by the h u m an visual
system as being uniquely specified by its joint position along three
vertices— its reflectance efficiencies at three critical w avelengths,
those w avelen gth s to w hich the retina's triune cone system is selec
tively respon sive. Since colors are ap p reh en ded by u s, it is a good
hyp oth esis that those three param eters are rep resented in our visual
system s, and that our sensations-of-color are in som e direct w ay d e
term ined by them . Sensations-of-color m ay turn out literally to be
three-elem en t chords in som e neural m edium ! In the face of all this, I
do n ot see w hy it is even briefly plausible to insist that it is utterly
im possible for a conceptually sophisticated M ary accurately to im
agine, and then reliably pick out, color sensations she has not
previously enjoyed. W e can already foresee how it m ight actually
be done.
The preceding argum ent does not collapse the distinction (betw een
know ledge by description and know ledge by acquaintance) urged
earlier in the discussion of equivocation. But it does show that the
"taxo n o m ie s" that reside in our prelinguistic m edia of representation
can be profoundly shaped by the taxonom ies that reside in the lin
guistic m edium , especially if one has had long practice at the obser
vational discrim ination of item s that an sw er to those linguistically
em bodied categories. This is just a further illustration of the plasticity
of h um an perception.
I do not m ean to suggest, of course, that there will be no limits to
w hat M ary can im agine. H er brain is finite, and its specific an atom y
will have specific limitations. For exam ple, if a b at's brain includes
com putational m achinery that the h um an brain sim ply lacks (w hich
seem s likely), then the subjective ch aracter of som e of the b at's inter
nal states m ay well be beyond h um an im agination. Clearly, h ow ever,
the elusiveness of the b at's inner life here stem s not from the m eta
physical "e m e rg e n ce " of its internal qualia, but only from the finite
capacities of our idiosyncratically hum an brains. W ithin those sheerly
structural lim itations, our im aginations m ay soar far beyond w hat
Jackson, N agel, and Robinson suspect, if w e p ossess a neuroscientific
conceptual fram ew ork that is at last adequate to the intricate p h e
n om en a at issue. (See especially chapter 5, section 7, and ch apter 9,
section 4.)
I suggest, then, that those of us w ho prize the flux and con ten t
of our subjective phenom enological experience need not view the
ad van ce of materialistic neuroscience w ith fear and foreboding.
Quite the contrary. The genuine arrival of a m aterialist kinem atics
and dynam ics for psychological states and cognitive p rocesses will
constitute not a gloom in w hich our inner life is supp ressed or
eclipsed, but rather a daw ning, in w hich its m arvellous intricacies
are finally revealed— m ost notably, if w e apply ourselves, in direct
self-conscious introspection.
Chapter 4
Knowing Qualia: A Reply to Jackson
In a recen t p ap er concerning the direct introspection of brain states

(1985b) I leveled three criticisms against Frank Jackson's "k n ow led ge
a rg u m e n t." A t stake w as his bold claim that no m aterialist accou n t of
m ind can possibly accou n t for all m ental p h en om en a. Jackson has
replied to those criticisms in his 1986. It is to those replies, and to the
issues that prom p ted them , that the p resent ch apter is directed.
1 The P ersisten t Equivocation
Jackson con cedes the criticism I leveled at m y ow n statem en t of his

arg u m en t— specifically, that it involves an equivocation on 'know s
ab ou t'— but he insists that m y reconstruction does not rep resen t the
argu m en t he w ishes to defend. I accept his instruction, and turn m y
attention to the su m m ary of the argum ent he provides at the bottom
of p age 293. M ary, you will recall, has been raised in innocence of any
color experience, but has an exhaustive com m and of n eu roscience.
(1) M ary (before her release) know s everything physical there is

to know about oth er people.
(2) M ary (before h er release) does not know everything there is
to know about oth er people (because she learns som ething
about them on h er release).
. ‘ . (3) There are truths about other people (and herself) w hich
escape the physicalist story.
Regim enting further, for clarity's sake, yields the following.
(1) (x){(H x& cPx)z>K m x]

(2) ( 3 x)[H x & ~Kmx\ (viz., "w h a t it is like to see re d ")
. •. (3) ( 3 x)[H x & ~ P x ]
H ere m = M ary; Кух = y know s about x; H x = x is about p ersons;

Px = x is about som ething physical in ch aracter; and x ran ges over
"k n o w ab les," generously construed so as n ot to beg any questions

about w h eth er they are propositional or otherw ise in nature.
Thus exp ressed , the argum ent is formally valid: the salient m ove is
a m odus tollens that applies the second conjunct of prem ise (2),
'~ K m x ' , to the w aiting consequent of prem ise (1), 'Km x'. The ques
tions now are w heth er the prem ises are jointly tru e, and w h eth er the
crucial notion 'K m x' is univocal in both of its appearan ces. H ere I am
surprised that Jackson sees any p rogress at all w ith the above for
m ulation, since I continue to see the sam e equivocation found in m y
earlier casting of his argum ent.
Specifically, prem ise (1) is plausibly true, w ithin Jackson's story
about M ary's color-free upbringing, only on the interpretation of
'know s about' that casts the object of know ledge as som ething pro-
positional, as som ething adequately expressible in an English sen
tence. M ary, to p ut it briefly, gets 100 p ercen t on every w ritten and
oral exam ; she can pronoun ce on the truth of any given sentence
about the physical characteristics of p erson s, especially the states of
their brains. H er "k now ledge by description " of physical facts about
p ersons is w ithout lacunae.
Prem ise (2), h ow ever, is plausibly true only on the interpretation of
'know s about' that casts the object of know ledge as som ething non-
propositional, as som ething inarticulable, as som ething that is n on
truth-valuable. W hat M ary is m issing is som e form of "k n ow led ge by
acq u ain tan ce," acquaintance w ith a sen sory ch aracter, p rototyp e, or
universal, perhaps.
G iven this prim a facie difference in the sense of 'know s ab ou t', or
the kind of know ledge appearing in each prem ise, w e are still looking
at a prim a facie case of an argu m en t invalid by reason of equivocation
on a critical term . Replace either of the 'K 's above with a distinct
letter, as acknow ledgm ent of the am biguity d em an d s, and the infer
en ce to (3) evaporates. The burd en of articulating som e specific and
u nitary sense of 'know s about', and of arguing that both prem ises
are true und er that interpretation of the epistem ic operator, is an u n
d ischarged burden that still belongs to Jackson.
It is also a heavy burden, since the resou rces of m odern cognitive
neurobiology already provide us w ith a plausible accou n t of w hat the
difference in the tw o kinds of know ledge am ou nts to, and of h ow it is
possible to have the one kind w ithout the other. L et m e illustrate with
a case distinct from that at issue, so as not to beg any questions.
A n y com p etent golfer has a detailed representation (perhaps in his
cerebellum , p erhaps in his m otor cortex) of a golf sw ing. It is a m otor
rep resentation , and it constitutes his "k n ow in g h o w " to execute a
p rop er sw ing. The sam e golfer will also have a discursive rep resen ta
Qualia: A Reply to Jackson 69
tion of a golf sw ing (perhaps in his language cortex, or in the neigh

boring tem poral and parietal regions), w hich allows him to describe a
golf sw ing or p erhaps draw it on paper. The m otor and the discursive
rep resentation s are quite distinct. Localized brain trau m a, or surgery,
could rem o ve either one while sparing the other. Short of that, an
inarticulate golf cham pion m ight have a superb rep resentation of the
form er kind, but a feeble representation of the latter kind. A n d a
p hysicist or sports physiologist m ight have a detailed and penetrating
rep resentation of the m echanics of a good sw ing, and yet be unable to
duff the ball m ore than ten feet because he lacks an adequate m otor
rep resentation , of the desired behavioral sequence, in the brain areas
that control his limbs. Indeed, if our physicist is chronically disabled
in his m o to r capacities, he m ay have no m otor rep resentation of a golf
sw ing w h atsoev er. In one m edium of representation, his rep resen ta
tional achievem ents on the topic m ay be com plete; while in another
m edium of representation , he has nothing.
A con trast betw een "k now in g h o w " and "k n ow in g th a t" is one
already acknow ledged in com m on sense, and thus it is n ot surprising
that som e of the earliest replies to Jackson 's argum ent (N em irow
1980; Lew is 1983) tried to portray its equivocation in these familiar
term s, and tried to explicate M ary's m issing know ledge solely in
term s of h er m issing som e one or m ore abilities (to recognize red, to
im agine red, etc.). W hile the approach is well m otivated, this binary
distinction in types of know ledge barely begins to suggest the range
and variety of different sites and types of internal rep resentation to be
found in a norm al brain. There is no reason w h y w e m u st be bound
by the cru d e divisions of our prescientific idiom s w hen w e attem p t to
give a precise and positive explication of the equivocation displayed
in Jackson's argu m en t. A n d there are substantial grounds for telling a
som ew h at different story concerning the sort of nondiscursive know l
edge at issue. Putting caution and qualification m om entarily aside, I
shall tell such a story.
In creatu res w ith trichrom atic vision (i.e ., w ith three types of reti
nal cone), color information is coded as a pattern of spiking frequencies
across the axonal fibers of the parvocellular subsystem of the optic
nerve. That m assive cable of axons leads to a second population of
cells in a central body called the lateral geniculate nucleus (LG N ),
w hose axonal projections lead in turn to the several areas of the visual
cortex at the rear of the brain's cerebral hem isperes, to V I, V2, and
ultim ately to V4, w hich area appears to be especially devoted to the
p rocessing and rep resentation of color inform ation (Zeki 1980; Van
E ssen and M aunsell 1983; Hubei and Livingstone 1987). H um an
cognition divides a sm ooth continuum of color inputs into a finite
n u m ber of prototypical categories. The lam inar structure at V4 is

p erh ap s the earliest place in the processing hierarchy to w hich w e
m ight ascribe that familiar taxonom y. A creatu re com petent to m ake
reliable color discrim inations has there developed a representation of
the range of familiar colors, a rep resentation that appears to consist in
a specific configuration of w eighted synaptic connections m eeting the
millions of neu ron s that m ake up area V4.
That configuration of synaptic w eights partitions the "activation
sp a ce " of the neurons in area V4: it partitions that abstract space into
a stru ctured set of subspaces, one for each prototypical color. Inputs
from the eye will each occasion a specific p attern of activity across
these cortical n eurons, a pattern or v ector that falls w ithin one of
those subspaces. In such a pigeon holing, it now appears, does visual
recognition of a color consist (see ch apters 5 and 9 for the general
theory of inform ation processing here appealed to). This recognition
dep en ds upon the creature possessing a prior rep resentation — a
learned configuration of synapses m eeting the relevant population of
cells— that antecedently partitions the creatu re's visual taxonom y
so it can respon d selectively and appropriately to the flux of visual
stim ulation arriving from the retina and LG N .
This distributed representation is not rem otely propositional or dis
cursive, but it is entirely real. All trichrom atic anim als have one, even
those w ithout any linguistic capacity. It apparently m akes possible
the m any abilities w e expect from color-com petent creatures: discrim
ination, recognition, im agination, and so o n . Such a representation
is presum ably w hat a person w ith M ary's upbringing w ould lack, or
p ossess only in stunted or incom plete form . H er representational
space w ithin the relevant area of n euron s w ould contain only the sub
space for black, w hite, and the intervening shades of gray, for the
visual exam ples that have shaped her synaptic configuration w ere
limited to these. There is thus m ore than just a clutch of abilities
m issing in M ary: there is a com plex rep resentation — a processing
fram ew ork that d eserves to be called "c o g n itiv e "— that she either
lacks or h as in reduced form . There is indeed som ething she "d o e s
n ot k n o w ." Jackson's prem ise (2), w e m ay assum e, is thus true on
these w holly m aterialist assum ptions.
These sam e assum ptions are entirely consistent w ith the further
assu m p tion that elsew here in M ary's brain— in the language areas,
for exam ple— she has stored a detailed and even exhaustive set of
discursive, propositional, truth-valuable rep resentation s of w hat goes
on in peop le's brains during the experience of color, a set she has
b rought into being by the exhaustive reading of authoritative texts in
a com pleted cognitive neuroscience. She m ay even be able to explain
her o w n representational deficit, as sketched above, in com plete

neurophysical detail. Jackson's prem ise (1), w e m ay thus assum e, is
also true on these wholly m aterialist assum ptions.
The view sketched above is a live candidate for the correct story of
sensory coding and sensory recognition. But w heth er or n ot it is true,
it is at least a logical possibility. A ccordingly, w hat w e have sketched
here is a consistent but entirely physical m odel (i.e ., a m odel in w hich
Jackson 's conclusion is false) in w hich both of Jackson's prem ises are
true u n d er the appropriate interpretation. They can hard ly entail a
conclusion, then , that is inconsistent w ith physicalism . Their com -
possibility, on purely physicalist assum ptions, resides in the different
ch aracter and the num erically different m edium of rep resentation at
issue in each of the tw o prem ises. Jackson's argu m en t, to refile the
ch arge, equivocates on 'know s about'.
2 O ther Invalid Instances
A n argu m en t form w ith one invalid instance can be exp ected to have
others. This w as the point of a subsidiary objection in m y 1985b
pap er: if valid, Jackson's argum ent, or one formally parallel, w ould
also serve to refute the possibility of substance dualism . I did n ot there
exp ress m y point w ith notable clarity, how ever, and I accep t resp on
sibility for Jackson 's quite missing m y intention. Let m e try again.
The basic point is that the canonical presentation of the know ledge
argu m en t, as outlined on p. 67 above, w ould be just as valid if the
predicate term 'P' w ere everyw h ere replaced by ' £ '. A n d the result
ing prem ises w ould be just as plausibly true if
(1) ' £ ' stood for 'is about som ething ectoplasm ic in ch aracter'
(w here 'ectoplasm ' is an arbitrary nam e for the dualist's
nonphysical substance), and
(2) the story is altered so that M ary b ecom es an exhaustive
expert on a com pleted ectoplasm ic science of h u m an nature.
The plausibility w ould be com parable, I subm it, b ecause a long dis
cursive lecture on the objective, statable, law -governed p roperties of
ectoplasm , w hatever they m ight be, w ould be exactly as useful, or
u se less , in helping M ary to know -by-acquaintance "w h a t it is like to see
re d ," as w ould a long discursive lecture on the objective, statable,
law -governed properties of the physical m atter of the brain. E ven if
substance dualism w ere true, therefore, and ectoplasm w ere its
heroic principal, an exactly parallel "k n ow ledge arg u m en t" w ould
"s h o w " that there are som e aspects of consciousness that m u st for
ever escape the ectoplasm ic story. Given Jackson 's antiphysicalist in
tentions, it is at least an irony that the sam e form of argum ent should
incidentally serve to blow substance dualism out of the w ater.
Though I am hardly a substance dualist (and neither is Jackson), I
do regard substance dualism as a theoretical possibility, one that
m ight conceivably succeed in explicating the psychological ontology
of com m on sense in term s of the underlying properties and law-
governed behavior of the nonm aterial substance it postulates. A n d I
m u st p rotest that the parallel know ledge argu m en t against substance
dualism w ould be wildly unfair, and for the v ery sam e reason that its
analogue against physicalism is unfair: it w ould equivocate on
'know s about'. It w ould be no m ore effective against dualism than it
is against m aterialism .
The parallel un der exam ination contains a further lesson. If it
w orks at all, Jackson's argum ent w orks against physicalism not b e
cau se of som e defect that is unique to physicalism ; it works because no
am ount o f discursive know ledge, on any topic, w ill constitute the nondis-
cursive fo rm o f know ledge that M ary lacks. Jackson 's argum ent is one
instance of an indiscrim inately antireductionist form of argum ent. If it
w orks at all, an analog will w ork against any proposed reductive,
discursive, objective account of the natu re of our subjective experi
en ce, no m atter w hat the reducing theory m ight happen to be. I see
this as a further sym ptom of the logical pathology described earlier.
Since the argum ent "w o rk s" for reasons that have nothing essential
to do w ith physicalism , it should "w o rk " against the explanatory
aspirations of other ontologies as well. A n d so it " d o e s ." The price of
em bracing Jackson's argum ent is thus dram atically higher than first
ap p ears. For it m akes any scientific accou n t of our sensory experience
entirely im possible, n o m atter w hat the ontology em ployed.
3 A G enuinely N onequivocal K now ledge A rgum ent
W e can appreciate the equivocation m ore deeply if w e explore a v er

sion of Jackson 's argum ent that does not equivocate on 'know s
ab ou t'. The equivocation can quickly be closed, if w e are d eter
m ined to do so, and the results are revealing. Given that the problem
is a variety in the possible form s of know ing, let us simply rew rite the
argu m en t w ith suitable quantification over the relevant form s of
know ing. The first prem ise m u st assert that, for any know able x, and
for any f o r m /o f know ledge, if x is about h u m an s and x is physical in
character, then M ary k n o w s(/) about x. The second prem ise is m o d
ified in the sam e m od est fashion, and the conclusion is identical.
C anonically,
(1') (x )(f)[(H x & P x)= > K (f)m x]

(2') (3 x )(3 f)[H x & ~ K ( f) m x ]
. ■. (3 ') ( 3 x)[H x & ~ P x ]
This arg u m en t is also form ally valid, and its prem ises explicitly en
com p ass w h atever variety there m ay be in form s of know ing. W hat
can w e say about its soundness?
A ssu m e that M ary has had the upbringing described in Jackson's
story, an d thus lacks any know ledge-by-acquaintance w ith "w h a t it is
like to see re d ." Prem ise (2') will then be true, as and for the reasons
that Jack son 's story requires. W h at will be the truth value of prem ise
(1') on these assum ptions?
Prem ise (1') is now a very strong claim indeed, m u ch stron ger than
the old prem ise (1), and a m aterialist will be sure to insist that it is
false. The reason offered will be that, because of her deprived u p
bringing, M ary quite clearly lacks one form of know ledge of a certain
physical asp ect of people. Specifically, she lacks a p ro p er configura
tion of synaptic connections m eeting the neurons in the appropriate
area of h er visual cortex. She thus lacks an appropriately partitioned
activation vector space across those n eurons, and therefore has no
rep resentation , at that site, of the full range of sensory coding vectors
that m ight som eday com e from the retina and the LG N . In other
w ord s, there is som ething physical about person s (their color sen sa
tions, or identically, their coding v ectors in their visual p ath w ays),
and there is som e form of know ledge (an antecendently partitioned
prelinguistic taxon om y), such that M ary lacks that form of know ledge
of that aspect of p ersons. A ccordingly, prem ise (1 ') is false and the
conclusion (3') is not sustained.
F rom a m aterialist's point of view , it is obvious that (1') will be false
on the assum ptions of Jackson's story. For that story denies her the
upbringing that norm ally provokes and shapes the d evelopm ent of
the relevan t representation across the appropriate population of cor
tical n eu ron s. A nd so, of course, there is a form of know ledge, of a
physical asp ect of p erson s, that M ary does n ot have. A s just illus
trated, the m aterialist can even specify that form of know ledge, and
its objects, in neural term s. But this m eans that prem ise (1 '), as p ro p
erly quantified at last, is false. M ary does not have know ledge of
everyth in g physical about persons, in every w ay that is possible for
her. (That is w hy prem ise (2') is true.)
T here is, of course, no guarantee that the m aterialist's accou n t of
sensations and sensory recognition is correct (although the ex
perim ental and theoretical evidence for a view of this general kind
continues to accum ulate). But neither is Jackson in a position to insist
74 The N ature of Mind
th at it m u st be m istaken. That w ould beg the v ery question at issue:

w heth er sensory qualia form a m etaphysically distinct class of p h e
n om en a beyond the scope of physical science.
To sum m arize, if w e w rite a deliberately non-equivocal form of
Jackson 's argu m ent, one that quantifies appropriately over all of the
relevant form s of know ledge, then the first prem ise m ust alm ost cer
tainly be false un d er the conditions of his ow n story. So, at any rate,
is the m aterialist in a strong position to argue. Jackson's exp ressed
h ope for "h igh ly plausible p rem ises" is not realized in (1 '). The
original prem ise (1) w as of course m u ch m ore plausible. But it
failed to sustain a valid argum ent, and it w as plausible only b ecause
it failed to address all the relevant form s of know ledge.
4 C onverting a Third-Person A ccount into a First-P erson A ccount
M y final objection to Jackson w as aimed m ore at breaking the grip

of the ideology behind his argum ent than at the argum ent itself. That
ideology includes a dom ain of properties— the qualia of subjective
exp erience— that are held to be m etaphysically distinct from the
objective physical properties ad dressed by orth od ox science. It is not
a surprise, then , on this view , that one m ight know all physical facts,
an d y et be ignorant of som e dom ain of these nonphysical qualia. The
contrast betw een w hat is know n and w h at is not know n sim ply re
flects an an tecen d ent m etaphysical division in the furniture of the
w orld.
But there is an other w ay to look at the situation, one that finds no
such division. O ur capacity for recognizing a range of (currently) in-
articulable features in our subjective experience is easily explained on
m aterialist principles; the relevant sketch app ears earlier in this essay
and elsew here in this volum e (ch ap ter 5, section 7). O ur discursive
inarticulation of those features is no surprise either, and signifies
nothing about their metaphysical status (chapter 10, section 5). Indeed,
that veil of inarticulation m ay itself be sw ept aside by suitable learn
ing. W h at w e are now able spontaneously to report about our internal
states and cognitive activities need not define the limit on w h at w e
m ight be able to report, spontaneously and accurately, if w e w ere
taught a m ore appropriate conceptual sch em e in w hich to exp ress our
discrim inations. In closing, let m e again urge on Jackson this exciting
possibility.
The intricacies of brain function m ay be subjectively opaque to us
n ow , but they need not rem ain that w ay forever. N euroscience m ay
ap p ear to be defective in providing a purely "th ird -p erso n acco u n t"
of m ind, but only familiarity of idiom and spontaneity of conceptual
respon se are required to m ake it a "first-p erson a cco u n t" as well.

W hat m akes an account a “'first-person accou n t" is not the conten t o f that
account, but the fa c t that one has learned to use it as the vehicle o f spon
taneous conceptualization in introspection and self-description. W e all of us,
as children, learned to use the fram ew ork of current folk p sychology
in this role. But it is entirely possible for a person or culture to learn
and use som e other fram ew ork in that role, the fram ew ork of cogni
tive neuroscien ce, perh ap s. Given a deep and practiced familiarity
w ith the developing idioms of cognitive neurobiology, w e m ight
learn to discrim inate by introspection the coding vectors in our inter
nal axonal p ath w ays, the activation p attern s across salient neural
populations, and m yriad other things besides.
Should that ever happ en, it w ould then be obvious to everyone
w ho h ad m ade the conceptual shift that a com pleted cognitive
n euroscien ce w ould constitute not a pinched and exclusionary pic
ture of h u m an consciousness, one blind to the subjective dim ension
of self, as Jackson's argum ent suggests. R ather, it w ould be the vehi
cle of a grand recon stru ction and expansion of our subjective con
sciousn ess, since it w ould provide us w ith a conceptual fram ew ork
that, unlike folk p sychology, is at last equal to the kinem atical and
dynam ical intricacies of the w orld w ithin. (See also ch ap ter 1 of this
volum e, and C hurchland 1979, section 16.)
Real preced en ts for such a reform ation can be draw n from our ow n
history. W e did not lose con tact w ith a m etaphysically distinct dim en
sion of reality w h en w e stopped seeing an im m utable, sparkle-strew n
quintessential crystal sphere each time w e looked to the h eaven s, and
began to see instead an infinite space of gas and du st and giant stars
stru ctured by gravitational attractions and violent nuclear p rocesses.
O n the con trary, w e now see far m ore than w e used to, even w ith the
unaided eye. The diverse "c o lo rs" of the stars allow us to see directly
their absolute tem peratu res. Stellar tem p eratu re is a function of stel
lar m ass, so w e are just as reliably seeing stellar m asses. The intrinsic
lum inosity or brightness of a star is tightly tied to these sam e features,
and thus is also visually available, no m atter how bright or faint the
star m ay ap p ear from Earth. A p paren t brightness is visually obvious
also, of cou rse, and the contrast betw een the app aren t and the intrin
sic brigh tnesses gives you the star's rou gh distance from E arth. In this
w ay is the ch aracter and three-dim ensional distribution of com plex
stellar objects in a volum e of interstellar space hu n dreds of light years
on a side m ad e visually available to you r unaided eyes from you r ow n
back y ard , given only the right conceptual fram ew ork for grasping it,
and observational practice in using that fram ew ork. F rom within the
new fram ew ork, one finds a system atic significance in experiential
details that hitherto w ent largely or entirely unnoticed (com pare

Feyerabend 1963b).
The case of inner space is potentially the sam e. W e will not lose
contact w ith a m etaphysically distinct dim ension of self w hen w e stop
introspecting inarticulable qualia, and start introspecting "in ste a d "
sensory coding vectors and sundry activation pattern s w ithin the v ec
tor spaces of our accessible cortical areas. A s w ith the revolution in
astron om y, the p rosp ect is one w e should w elcom e as m etaphysically
liberating, rath er than deride as m etaphysically irrelevant or m eta
physically impossible.
Chapter 5
Some Reductive Strategies in
Cognitive Neurobiology
A pow erful conception o f representation and com putation— draw n fro m

recent w ork in the neurosciences— is here outlined. Its virtues are explained
and explored in three im portant areas: sensory representation, sensorim otor
coordination, and m icrophysical im plem entation. It constitutes a highly
gen eral conception o f cognitive activity that has significant reductive
potential.
1 Introduction
The aim of this p aper is to m ake available to philosophers an in

triguing theoretical ap proach to representation and com putation
currently under exploration in the neurosciences. The ap p roach is
intriguing for at least three reasons. First, it provides a highly
general an sw er to the question of how the brain m ight represent the
m any asp ects of the w orld in w hich it lives. Later in the p ap er I shall
explore that an sw er as it applies to a case familiar to philosophers: the
case of the various subjective sensory qualia displayed in on e's m an
ifold of sensory intuition. There w e shall find the outlines of a
genuine neurobiological reduction of the familiar sensory qualia. This
application is but one am ong m any, h ow ever, as I shall also try to
show . O ne im portant result is that diverse cases of representation ,
cases that appear to com m on sense as being entirely distinct in ch ar
acter, em erge as being fundam entally the sam e. The ap p roach thus
finds unity in diversity.
The secon d intriguing aspect concerns com putation. The style of rep
resentation to be outlined lends itself uniquely well to a pow erful
form of com putation, a form well suited to the solution of a wide
variety of problem s. O ne of them is a problem relatively unfam iliar to
philosophers, the problem of sensorim otor coordination. H ow ever
unfam iliar it m ay be, this problem is of fundam ental im p ortan ce to
T h is p a p e r first ap p eared in M in d 95, n o. 379 (Ju ly 1986).

cognitive theory, since the adm inistration of appropriate behavior in

the light of current experience is w here intelligence has its raw begin
nings. Since sensorim otor coordination is the m ost fundam ental
problem that any animal m ust solve, a m ean s of solution coupled to a
general accoun t of representation m u st surely arouse our curiosity.
Third, the approach is intriguing in the w ay in w hich it em braces
the m ystery of the m icrophysical organization of the brain and the ques
tion of how its specific organization im plem ents the representational
and com putational activities that the brain as a w hole displays. H ere
too the approach finds empirical encou ragem ent, for there are at least
tw o m ajor w ays of physically im plem enting the abstract ap proach to
rep resentation and com putation p rop osed , and each of them bears
a suggestive resem blance to real neural structures displayed p rom
inently throu ghou t the empirical brain, nam ely, the lam inar organ i
zation of the cerebral cortex and the dense orthogonal m atrix of the
cerebellar cortex.
O verall, the approach constitutes an unabashedly reductive
strategy for the neuroscientific explanation of a variety of familiar
cognitive p h en om en a. The propriety of such strategies is still, of
course, a keenly debated issue w ithin the philosophy of m ind. H is
torically, this debate has been im poverished by the lack of any very
im pressive general theories from neurobiology, theories that prom ise
actually to effect the neurobiological reduction of som e familiar class
of cognitive ph en om en a. A t least, if such theories did exist in the
outlying literature, they did not m anage to m ake it into the philo
sophical debates. Given this absence of relevant theory, antireduc
tionist argu m en ts could and often did p roceed simply by holding up
som e asp ect of our cognitive life and asking the rhetorical question,
"H o w could this ever be accounted for, or even addressed, by any
possible story about the nuts and bolts of n eu ro n s?"
Such rhetorical questions unfairly exploit the feebleness of our im
aginations, since a reply even rem otely adequate to the ph en om en a
is not som ething w e can reasonably be expected to think up on
dem an d. A s it h ap p ens, h ow ever, potentially adequate replies have
indeed em erged from recent w ork in cognitive p sychology and cogni
tive neurobiology. Their existence, I believe, m u st soon shift our
attention from the abstract issue of w h eth er any such reduction is
possible to the concrete issue of w hich of various alternative n eu ro
biological theories truly provides the right reduction, and to its long
term consequences for our overall self-conception.
The basic idea, to be explained as w e proceed , is that the brain
rep resents various aspects of reality by a position in a suitable state
space, and the brain perform s com putations on such representations
Reductive Strategies in Cognitive Neurobiology 79
by m ean s of general coordinate transform ations from one state space to

an other. These notions m ay seem arcane and forbidding, but the
graphics below will dem ystify them very swiftly. The theory is entire
ly accessible— indeed, in its sim plest form it is visually accessible—
even to the nonm athem atical reader.
I w as initially introduced to this theoretical approach by reading the
provocative papers of the neuroscientists A ndras Pellionisz and
Rodolfo Llinas (1979, 1982, 1984, 1985). Their presentations are m uch
m ore general and m ore penetrating than the sketch to be provided
here. But for didactic reason s, discussion of their ground-breaking
w ork will be postponed until the later sections of the p ap er. A t the
outset I w ish to keep things as simple as possible.
W e o p en our discussion, then, by confronting three apparently dis
tinct puzzles:
• The m ystery of how the brain represents the w orld, and how it
p erform s com putations on those representations
• The m ystery of sensorim otor coordination
• The m ystery of the brain's m icrophysical organization.
It is especially encouraging that these problem s appear to adm it of a

sim ultaneous solution. Let us begin by addressing the third m ystery,
the m icrostructural m ystery.
2 L am inar Cortex, Vertical Connections, and Topographic M aps
The ou ter surface of the brain's great cerebral h em ispheres consists

of a thin layer, the classical ''g rey m a tte r," in w hich m ost of their
neuronal cell bodies are located (see figure 5.1a). The rem aining
"w h ite m atter" consists primarily of long axons projecting from the
cells in this layer to other parts of the brain. If one exam ines the inter
nal structu re of this w rinkled layer, one finds that it subdivides into
further layers (see figure 5.1b). H um an cortex has six of these layers.
O ther creatu res display a different num ber, but the lam inar p attern is
standard.
These further layers are distinguished by the type and con cen tra
tion of cells within each sublayer, and by the m assive intralayer or
"h o rizo n tal" projections within each sublayer. M oreover, these dis
tinct layers are further distinguished by their p roprietary inputs or
outputs. The top several layers tend to have only inputs of one kind
or an oth er, from the sensory periphery, from other parts of the cor
tex, or from other parts of the brain. A nd the bottom layer seem s
invariably to be an output layer.
M olecular
layer
E xte rnal
g ran u lar
layer
E xternal
p yram id al
layer
Internal
g ranu lar
layer
In te rn al
p yram id al
layer
M u ltifo rm
layer
G olgi N issl Weigert
(b) Method of staining
Figure 5.1
(a) A cross-section of the cerebral hemispheres showing the outer gray layer, the cere
bral cortex. (b) The internal laminar structure of the cortical layer, as revealed by three
different stains, (c) Brodmann's areas
Finally, these distinct layers are system atically con n ected , in the
fashion of nails struck through plyw ood, by large num bers of vertical
ly oriented cells that perm it com m unication betw een the several
layers. These vertical cells conduct neuronal activity d ow n w ard s,
from the superficial input layers above to the output layer below.
If w e now leave our m icroscopic edgew ise perspective and look at
the cortical sheet from the outside, w e find that the cortical surface
divides into a patchw ork of sm aller regions (see figure 5.1c). These
areas are distinguished to som e degree by differences in their lam inar
cytoarch itectu res. An initial taxonom y, into w hat are called "B rod -
m an n 's are a s" after their discoverer, is based on this criterion. These
areas or subareas of these areas are of further interest b ecause several
of th em plainly constitute topographic m aps of som e aspect of the sen
sory or m otor periphery, or of som e other area of the brain. For exam
ple, the neighborhood relations holding betw een the cells in a given
layer of the visual cortex at the rear of the brain corresp on d to the
neighborhood relations holding betw een the cells in the retina from
w hich they receive inputs. The bundle of axonal projections from the
retinal cells to the cortical cells preserves the topographic organiza
tion of the retinal cells. The surface of the prim ary visual cortex thus
constitutes a topographic m ap of the retinal surface.
It is term ed a 'topograp hic m ap' rath er than simply a 'm ap ' be
cause the distance relations am ong retinal cells are generally not p re
served. Typically, such m aps are m etrically deform ed, as if they w ere
m ade of rubber and then stretched in som e fashion.
M any such m aps have been identified. The so-called "visu al cor
te x " (areas 17, 18) has already b een m entioned. The upper layer of the
som atosensory cortex (area 3) is a topographic m ap of the b od y's tac
tile surface. The low er layer of the m otor cortex (area 4) is a top o
graphic m ap of the bod y's m uscle system . The auditory cortex (areas
41, 42) contains a topographic m ap of frequency space. A nd there are
m any oth er cortical areas, less well u n d erstood as to exactly w hat
they m ap, but w hose topographical re-presen tation of distant stru c
ture is plain.
This general pattern of neural organization is not confined to the
surface of the great cerebral hem ispheres. A s well, various nuclei of
"g re y m atter" in the m ore central regions of the brain— for exam ple,
the superior colliculus, the hippocam pus, and the lateral geniculate
nu cleu s— display this sam e m ultilayered, topographically organized,
vertically con n ected structure. N ot everything does (the cerebellum ,
for exam ple, is rather different, of w hich m ore later), but the p attern
described is one of the m ajor organizational patterns to be found in
the brain.
W h y this p attern ? W hat is its functional or cognitive significance?
W hat do these structures do, and h ow do they do it? W e can
approach a possible answ er to these questions by addressing the
second m ystery, the problem of sensorim otor coordination.
3 Sensorim otor Coordination
L et m e begin by suggesting that vertically connected lam inar stru c

tures are one of evolution's sim plest solutions to a crucial type of
problem , one that any sensorim otor system beyond the m ost
rud im entary m ust som ehow solve. In ord er to appreciate this type
of problem , let us consider a schem atic creature of a deliberately
contrived simplicity.
Figure 5 .2 b is a plan view of a crablike schem atic creature (5.2a)
w ith tw o rotatable eyes and an extendable arm . If this equipm ent is to
be useful to the crab, the crab m u st em body som e functional rela
tionship, betw een its eye-angle pairs w h en an edible object is triangu
lated, and its subsequent shoulder and elbow angles, so that the arm
can assu m e a position that m akes con tact w ith the edible target.
C rudely, it m u st be able to grasp w hat it sees, w herever the seen
object lies.
W e can characterize the required arm /eye relationship as follows.
First of all, let us represent the input (the pair of eye angles) by a point
Figure 5.2
in a tw o-dim ensional sensory-system coordinate space or state space

(figure 5 .3 a). The output (the pair of arm angles) can be also be rep re
sented by an appropriate point in a separate tw o-dim ensional motor
state space (figure 5.3b).
W e n ow need a function to take us from any point in the sensory
state space to a suitable point in the m otor state space, a function that
will coordinate arm position w ith eye position in the m an n er d e
scribed. (I here sketch the deduction of the relevant function so that
its origin will not be a m ystery, but the read er m ay leap p ast the
algebra w ithout any loss of com prehension. The only point to rem em
ber is th at w e are deducing a suitable function to take us from eye
configurations to arm configurations.)
The tw o eye angles < a , /3 > determ ine tw o lines that intersect at the
seen object. The coordinates (a, b) of that point (in real space) are
given by
a = —4(tan a + tan /3) / (tan a — tan /3)

b = —8(tan a •tan /3) / (tan a — tan /3)
The tip of the arm m ust m ake contact w ith this point. If w e assum e
that b oth the forearm and the upper arm have a fixed length of 7
units, the elbow will therefore have to lie at the intersection of two
Figure 5.3
circles of radius 7 units: one centered at (a, b), and the oth er centered
at (0, 0), w here the upper arm projects from the crab's b od y. If w e
solve for the relevant intersection, the real-space elbow coordinates
(xe, i/e) are given by
Xe = ((49 - ((fl2 + b2)2/ 4b2) ■(1 - (( a2/b2)/{(a 2/b2) + 1 )) ) ) ^
+ ({{alb) ■((a2 + b2)/2 b ))/((a2/b2) + 1 ) Щ / ( ( а 2/Ь2) + 1)1/2
y e = (49 - xe2)1/2
The three points in real space, (a, b), (xe, y e), (0, 0), determ ine the
position of the arm , w hose upper arm and forearm angles <6, <p> are
finally given by
0 = tan ~1(y e /x e)
ip = 180 — (0 — ta n _1((t> — y e) l( a ~ xe)))
These are the desired coordinates of the arm in m otor state space. The
reader will note that the assem bled functions that yield them are
rath er tangled ones.
Tangled or not, if the crab is draw n on a com p u ter screen such that
its final arm position (draw n by the com p uter as output) is the speci
fied function of its eye positions (entered by us as inp u t), then it
constitutes a very effective and w ell-behaved sensorim otor system ,
especially if w e w rite the controlling p rogram as follows.
L et the p rogram hold the crab's arm folded against its chest (at
0 = 0°, q> = 180°), until som e suitable stim ulus registers on the fovea of
both ey es. The arm is then m oved from its initial state-sp ace position
(0°, 180°) along a straight line in m otor-state space, to its com puted
target position in m otor-state space. This is the state-space position at
w hich, in real sp ace, the tip of the arm contacts the triangulation point
of the eyes. This arran gem ent produces a m odestly realistic system
that reach es unerringly for w hatever it sees anyw here w ithin reach of
its arm (figure 5 .4 a - d ) .
The algebraic rep resentation of the crab's sensorim otor tran sform a
tion, as rep resented in the six equations listed earlier, supplies no
intuitive conception of its overall nature. A geom etrical presen tation is
m u ch m o re revealing. Let us therefore consider the projection of the
active p ortion of the crab's sensory state space (figure 5.5a) onto the
o rth ogon al grid of its m otor state space (figure 5 .5 b), as im p osed by
the function un d er discussion. That is to say, for every point in the
displayed sen sory grid, w e have plotted the corresponding arm -
position w ithin the m otor grid.
H ere w e can see at a glance the distortion of the vertical and hori
zontal lines of sensory space, as projected onto m otor space. The
86
The Nature of Mind
Figure 5.4
The crab's arm in action
Figure 5.5
Reductive Strategies in Cognitive Neurobiology
A transformation of coordinates
87
topological features of the sensory space are p reserved , but its m etric
al properties are not. W hat w e see is a system atic transform ation o f
coordinates. (The heavy scored triangle and rectangle are draw n in
solely to help the reader locate corresponding positions in the d e
form ed and undeform ed grids. N ote also that the left border or Д axis
of figure 5.5a shrinks to the left radial point in figure 5.5b, and that the
top b order of figure 5.5a shrinks to the right radial point in figure
5.5b.)
4 Coordinate Transform ation: Its Physical Im plem entation
The transform ation described above sustains effective and realistic

sensorim otor behavior. But how could a real nervous system p o s
sibly com pute such a com plex coordinate transform ation? It is not
realistic to expect it to com pute this com plex trigonom etric function
step by step as our com pu ter sim ulation does. N evertheless, given
their sophisticated sensorim otor coordination, biological system s
som eh ow m ust be com puting transform ations like these, and others
m ore com plex still. H ow m ight they do it?
Figure 5 .5 suggests a surprisingly simple m ean s. If w e suppose that
the crab contains internal representations of both its sensory state
space and its m otor state space, then the following arran gem en t will
effect the desired transform ation. L et the crab's sensory state space be
rep resen ted by a physical grid of signal carrying fibers, a grid that is
m etrically deform ed in real space in just the w ay displayed in figure
5.5b. Let its m otor state space be represented by a second grid of
fibers, in an undeform ed orthogonal array. Position the first grid over
the second , and let them be connected by a large num ber of short
vertical fibers, extending from coordinate intersections in the sensory
grid d ow n to the nearest coordinate intersection in the underlying
m otor grid, as in figure 5.6.
S uppose that the fibers of the sensory grid receive input from the
eyes' proprioceptive system , such that the position of each eye stim u
lates a unique fiber in the upper (deform ed) grid. The left eye acti
v ates one fiber from the right radial point, and the right eye activates
one from the left. Joint eye position will thus be rep resented by a
sim ultaneous stim ulation at the appropriate coordinate intersection in
the upp er grid.
U n d ern eath that point in the upper m ap lies a unique intersection
in the m otor grid. Suppose that this intersecting pair of orthogonal
m otor fibres, w hen jointly activated, induces the arm to assum e the
position that is appropriate to the specific m otor coordinate intersec
tion w here this m otor signal originates.
EYES
Y
Figure 5.6
89
Effecting a coordinate transformation by means of contiguous, metrically deformed, vertically connected topographic maps
H appily, the relative metrical deform ations in the m aps have

placed in corresp on den ce the appropriate points in the upper and
low er m aps. W e need now suppose only that the vertical connections
b etw een the sensory grid and the m otor grid function as and gates or
threshold sw itches, so that a signal is sent d ow n the vertical con n ec
tion to the m otor grid just in case the relevant sensory intersection
point is sim ultaneously stim ulated by both of its intersecting sensory
fibres. Such a system will com pute the desired coordinate transform a
tions to a degree of accuracy limited only by the grain of the tw o grids
and by the density of their vertical connections. I call such a system a
“ state-sp ace san d w ich ."
Three points are w orth noting im m ediately about the functional
properties of such an arran gem en t. First, it will rem ain partially func
tional despite localized dam age. A small lesion in either grid will p ro
duce only a partial dyskinesia (tw o p erm an en t “ sh ad o w s" of fiber
inactivity d ow nstream from the lesion), for w hich a shift of bodily
position will usually com pensate (by bringing the target's state-space
position out of the shadow ).
Indeed, w e can do even better than this. L et the dimple on the back
of the crab's schem atic eye (figure 5.6) be replaced by a bar so as to
stim ulate not one proprioceptive cell, but rather a set of adjacent
p roprioceptive cells. The position of each eye will then be represented
in the up per m ap by the activation of a hand of fibers centered on the
"c o rre c t" fiber. Join t eye position will then be registered in the upper
layer by a distributed area of stim ulation, an area centered on the
"c o rre c t" point. This will cause a corresponding area of stim ulation in
the bottom grid, and thus a band of stim ulation will be sent to each
m uscle. If the m uscles are connected so as to assum e a position
appropriate to the "m e a n " fiber w ithin that distributed signal, even if
that specific fiber h appens to be inactive, then an appropriate m otor
respon se will be forthcom ing even if the sandw ich has suffered the
scattered loss of a great m any cells. Such a system will be functionally
persistent despite w idespread cell d am age. The quality of the sen
sorim otor coordination w ould be som ew h at degraded und er cell
d am age, but a roughly appropriate m otor response w ould still be
forthcom ing.
Second, this system will be very fast, even w ith fibers w ith bio
logical conduction velocities of betw een 10 and 100 m eters per
second. In a creature the size of a crab, in w hich the total conduction
p ath is less than 10 centim eters, this system will yield a m otor re
sp onse in well u n d er 10 m illiseconds. In the crab sim ulation de
scribed earlier, m y m icrocom pu ter (doing its trigonom etry within the
softw are) takes 20 tim es that interval to produ ce a m otor respon se on
screen, and its conduction velocities are on the order of the speed of
light. Evidently, the m assively parallel architecture of the state-space
sandw ich buys it a large advantage in speed, even w ith vastly slower
com p onen ts.
Third, the quality of the crab's coordination will not be uniform
over its field of m otor activity, since in the m axim ally deform ed areas
of the sen sory grid, small errors in sensory registration p rodu ce large
errors in the m otor response (see again figure 5.5b). A ccordingly, the
crab is least well coordinated in the area close betw een its eyes, and to
its extrem e right and left.
All three of these functional properties are biologically realistic.
A nd the sandw ich appears biologically realistic in one further respect:
it is relatively easy to imagine such a system being grow n . Distinct
layers can confine distinct chem ical gradients, and can thus guide
distinct m orp h ogenetic p rocesses. A ccordingly, distinct top ogra
phic m ap s can app ear in closely adjacent layers. But given that the
m aps are so closely contiguous, and given that they are ap p rop riate
ly deform ed, the problem of connecting them up so as to produ ce a
functional system becom es a trivial one: the solution is just to grow
conductive elem ents that are roughly orthogonal to the layers.
Different creatures will have different m ean s of locating objects,
and different m otor system s to effect contact w ith them , but all of
them will face the sam e problem of coordinating positions in their
sen sory state space with positions in their m otor state sp ace, and the
style of solution here outlined is evidently quite general in nature. In
fact, th e coordination of distinct biological subsystem s by coordinate
transform ation is a m atter that presum ably extends far beyond the
obvious case of basic sensorim otor coordination. The sam e strategy
m ay also be useful, or even essential, in the execution of higher cogni
tive activities, as w e shall see later on. The point to be em phasized
here is that a state-space sandw ich constitutes a simple and bio
logically realistic m eans for effecting any tw o-dim ensional to two-
dim ensional coordinate transform ation, w hatever its m athem atical
com plexity and w h atever features— external or internal, abstract or
co n crete— the coordinate axes m ay rep resent to the brain. If the
transform ation can be graphed at all, a sandw ich can com p u te it. The
sen sorim otor problem solved above is m erely a tran sp aren t exam ple
of the general technique at w ork.
B eyond its functional realism , the system of intercon n ected m aps
in figure 5 .6 is suggestively similar to the know n physical structure of
typical lam inar cortex, including the m any topographic m ap s distrib
u ted across the cerebral surface. In all of these areas, inputs address a
given layer of cells, w hich layer frequently em bodies a m etrically de
form ed topographic m ap of som ething or other. A nd outputs leave

the area from a different layer, w ith w hich the first layer enjoys m as
sive vertical connections.
I therefore propose the hypothesis that the scattered m aps w ithin
the cerebral cortex, and m any subcerebral lam inar structures as well,
are all engaged in the coordinate transform ation of points in one neu
ral state space into points in another, by the direct interaction of m et
rically deform ed, vertically connected topographic m aps. Their m ode
of representation is state-space position, their m ode of com putation is
coordinate transform ation, and both functions are sim ultaneously im
plem ented in a state-space sandw ich.
[A dded in 1989: The second and especially the third part of this
hypothesis now seem alm ost certainly m istaken, at least as an
accou n t of the cerebral cortex. The cell population of a given layer in a
given cortical area is indeed coding state-space positions, but m ore
likely by m ean s of the overall pattern of activation levels across the
entire population of cells, rather than by the narrow spatial location of
m axim al cell activation. A nd the axonal projections from that layer
into adjacent cell layers do indeed effect a transform ation from one
state-sp ace to another, but m ore probably this is done by m ean s of
the m atrix-m ultiplication style of transform ation explained in section
6 below , rath er than by the simple transfer of an activational hot-sp ot
betw een m utually deform ed m ap s. The topographic m appings so
characteristic of m any cortical areas now appear as the occasional arti
facts of a deeper coding strategy, the high-dim ensional vector coding
explained in section 6 below. The discussion of this section rem ains
instructive, h ow ever, as a possible accou n t of areas like the superior
colliculus, and as an illustration of the relevant com putational ideas.]
I can cite not a single cerebral area for w hich this functional h yp oth
esis is know n to be true. The decoding of cortical m aps is a business
that has only begun. Clear successes can be num bered on the fingers
of at m ost three or four h ands, and they are generally confined to the
superficial cortical layers. There is a m ajor subcortical area, h ow ever,
w hose upper-level and low er-level m aps h ave b een at least partially
decod ed . This subcortical area does display the general p attern d e
picted in figure 5.6.
The su perior colliculus is a phylogenetically very old lam inar stru c
ture (figure 5.7a) located on the dorsal m idbrain. A m ong oth er things,
it sustains the familiar reflex w hereby the eye m akes an involuntary
saccade so as to foveate (look directly at) any sudden change or m o ve
m en t that registers on the retina aw ay from the foveal center. W e
h ave all had the experience of being in a darkened m ovie theater
w hen som eone dow n in the front row left suddenly ignites a m atch or
Figure 5.7
(a) A projection drawing of a Nissl-stained cross-section of cat superior colliculus illustrating laminar organization. Dots correspond to
collicular neurons. From Kanaseki and Sprague 1974; reprinted with permission, (b) A retinotopic map, a metrically deformed topographic
93
map of the visual hemifield, in rectangular coordinates, on the superficial layer of the right colliculus of the cat. M = medial; L = lateral;
A = anterior; P = posterior. Adapted from Schiller 1984
lighter to light a cigarette. Every eye in the house m akes a ballistic

saccade to fixate this brief stim ulus before returning to the screen.
This is the colliculus at w ork. A ppropriately en ou gh , this is som e
tim es called the "visu al grasp reflex."
In h u m an s and the higher m am m als the superior colliculus is a
visual center secondary to the m ore im p ortan t striate cortex (areas 17
and 18 on the Brodm ann m ap) and located at the rear of the cerebral
hem isp heres, but in low er anim als such as the frog or snake, w hich
lack any significant cortex, the superior colliculus (or optic tectum , as
it is called in them ) is their principal visual cen ter. It is an im portant
cen ter even for m am m als, how ever, and it w orks roughly as follows.
The top -m ost layer of the superior colliculus (SC) receives p rojec
tions directly from the retina, and it constitutes a m etrically deform ed
top ograp h ic m ap of the retinal surface (figure 5 .7 b ) (Schiller 1984;
G oldberg and Robinson 1978; C yn ad er and Berm an 1972; G ordon
1973). Vertical elem ents con n ect this layer to the deepest layer of the
SC. These vertical connections app ear to consist of a chain of tw o or
three short interneurons descending stepw ise through tw o interven
ing layers (Schiller 1984, p. 460, 466), of w hich m ore later. A lso, the
d endrites of som e of the deep-layer neurons app ear to ascend directly
into the visual layer to m ake synaptic connections w ith visual cells
(M ooney et al. 1984, p. 185). The neurons of the deepest layer project
their outp u t axons via tw o distinct n ervou s path w ays, one of w hich
leads eventually to the pair of extraocular m uscles responsible for v er
tical eye m ovem en ts, and the oth er of w hich leads eventually to the
pair responsible for horizontal eye m ovem en ts (H uerta and H arting
1984, p. 287).
Intriguingly, this underlying m otor layer also em bodies a top o
g raphic m ap , a m ap of a state space that rep resen ts changes in the
contractile position of the ocular m uscles (Robinson 1972, p. 1800).
M icrostim ulation by an electrode at a given point in this deepest layer
cau ses the eyes to execute a saccade of a characteristic size and direc
tion, a saccade that m oves the fovea into the position form erly o ccu
pied by that retinal cell that projects to the im m ediately overlying visual
cell in the top-m ost layer o f the colliculus (Robinson 1972; Schiller and
Stryker 1972). In other w ords, the relative m etrical deform ations in
the tw o m aps have placed into corresp on d en ce the appropriate
points in the upp er and low er m ap s. (This m ean s that the "d efo rm a
tio n " seen in figure 5.7b should not of itself be taken as evidence for
the state-sp ace sandw ich h ypothesis. W hat is crucial is the d eform a
tion of the m aps relative to each oth er.)
Finally, any sufficiently stron g retinally p rodu ced stim ulation in
th at top-m ost visual m ap is con veyed d ow n w ard by the appropriate
vertical elem ents to the m otor m ap, w here it produ ces a saccade of
just the size and direction appropriate for foveating the external sti
m ulus th at provoked it. The superior colliculus thus ap p ears to be an
instance of both the structural and the functional pattern displayed in
figure 5 .6 .
O ne m ight exp ect a biological sandw ich to code the position of re
tinal stim ulations w ith an area of stim ulation in the upp er m ap , rather
than a single point, so as to be functionally persistent in the face of
small lesions and scattered cell death, as explained earlier. A ctivity in
the SC does display this pattern (M cllwain 1975; 1984, p . 268). The
schem atic m odel of figure 5 .6 also predicts that the size and direction
of the m o to r response induced by m icrostim ulation at various points
within the collicular sandw ich will be a function solely of w here in
either m ap the stim ulation occurs, and not of the m agn itu de of the
stim ulation nor of its vertical position betw een the tw o m ap s. Experi
m en tation has already yielded this result (Robinson 1972; Schiller and
Stryker 1972). The superior colliculus, it appears, is a real sensorim o
tor coordinate transform er of roughly the kind at issue. It foveates on
changing or m oving visual targets by essentially the sam e m eans
w hereby the schem atic cortex of the crab reaches out for triangulated
objects.
A w ord of caution is in order here, since the accou n t just offered
does not do justice to the full com plexity of the superior colliculus. In
m am m als, especially the higher m am m als, the SC is a tightly inte
grated part of a larger m odulating system that includes inputs from
the visual cortex and the frontal eye fields, and outputs to the neck
m uscles. The functional properties of the entire system are m ore
varied and m ore subtle than the preceding accou n t su ggests, and the
job of sorting them out is still un d erw ay (M ays and Sparks 1980;
Schiller and Sandell 1983). The preceding is subm itted as an account
of the central or m ore prim itive functions of the SC, at best.
W ith the exam ples of the crab's "c o rte x " and the superior colliculus
in m ind, it is appropriate to focus on the m any other topographically
organized, m ultilayered cortical areas scattered th rou g h o u t the brain,
and ask w hat coordinate transform ations they m ight be effecting.
H ere it is v ery im portant to appreciate that the top ograp h ic m aps w e
seek to d ecode need n ot, and generally will not, be m ap s of som e
thing anatom ically obvious, such as the surface of the retin a, or the
surface of the skin. M ore often they will be m aps of som e abstract state
space, w h ose dim ensional significance is likely to be opaque to the
casual observer, though of great functional im portance to the brain.
Two nice exam ples of such abstract m aps are the m ap of echo delays in
the b at's auditory cortex, and the m ap of binaural disparities in the

ow l's inferior colliculus (Konishi 1986).
In the case of the crab's schem atic cortex, the angular state of an
external system (the eyes) is directly m apped onto the angular state of
an other external system (the jointed arm ). But in a creature of any
com plexity, w e can expect a long chain or hierarchy of internal sys
tem s interacting with one an other, system s that are m aps of the out
p ut of other internal system s and w hose outp u t drives the activities of
further internal system s. To u nderstand such m aps will require that
w e un d erstan d the function of the other m ap s in the overall system .
All of this suggests that the brain m ay boast m any m ore top o
graphic m aps than have so far been identified or even su spected.
Certainly the brain has a teem ing abundance of topographically
organized areas, and recent w ork has exp an d ed the num ber of
k now n sensory-related m aps considerably (M erzenich and Kaas
1980; A llm an et al. 1981). All of this further suggests that w e will
m ake b etter progress in trying to un d erstan d the significance of the
m an y topographically organized cortical areas w hen w e ap proach
them as m aps of abstract but functionally relevan t state spaces.
There has b een a tendency am ong neuroscientists to restrict the
term "to p o g rap h ic m ap " to neural areas that m irror som e straight
forward aspect of the physical world or sensory system, such as the
retina, or the surface of the skin. This is unfortu n ate, since there is no
reason for the brain to show any such p reference in w hat it con stru cts
m aps of. A bstract state spaces are just as m appable as con crete
physical ones, and the brain surely has no ad van ce know ledge of
w hich is w hich. W e should expect it, rather, to evolve m aps of w h at is
functionally significant, and that will frequently be an abstract state
space.
5 Cortex w ith M ore than Two Layers
W hile w e are discussing the biological reality of the lam inar m ech a
nism p roposed , consider the objection that our m odel cortex has only
tw o layers, w hereas typical h um an cortex has six layers and, if w e
count fine subdivisions, p erhaps eight or nine in som e areas. W hat
are they for?
There is no difficulty in perceiving a function for such additional
layers. Let us retu rn again to the superior colliculus, w hich illustrates
one of m any possibilities here. Betw een the visual and m otor m ap s of
the superior colliculus there are, in som e creatu res, one or tw o in
term ediate layers (see again figure 5.7). These ap p ear to constitute an
auditory m ap an d /or a som atosen sory m ap (a facial or w hisker m ap),
w h ose function is again to orient the ey e's fovea, this tim e tow ard
the sou rce of sud den auditory an d /or som atosensory stim ulation (Gold
berg an d Robinson 1978). N ot surprisingly, these intervening m aps
are each m etrically deform ed in such a fashion as to be in rou gh co
ordinate register w ith the m otor m ap and hence w ith each other.
A ltogeth er, this elegant three- or four-layer top ograp h ic sandw ich
con stitu tes a m ultim odal sensorim otor coordinate transform er.
M ultilayered structu res have further virtues. It is plain that m aps of
several distinct m odalities, suitably deform ed and placed in collective
register w ithin a "club san d w ich ," provide a m ost effective m ean s of
cross-m od al integration and com parison. In the SC, for exam ple, this
m ultim odal arrangem ent is appropriate to the p roduction of a m otor
resp o n se to the jo in t receipt of faint but spatiotem porally coincident
auditory and visual stimuli, stimuli that, in isolation, w ould have
been su bthreshold for a m otor respon se. For exam ple, a faint sound
from a certain com pass point m ay be too faint to p rom p t the eyes into
a foveating saccade, and a tiny m ovem en t from a certain com pass
point m ay be similarly im potent, but if both the sound an d the m ove
m en t com e from the sam e com pass point (and are thus cod ed in the
SC along the sam e vertical axis), then their sim ultaneous conjunction
will indeed be sufficient to m ake the m otor layer direct the eyes
appropriately. This prediction is strongly corroborated by the recent
results of M eredith and Stein (1985).
F u rth er exploration reveals that m ultilayered sandw iches can sub
serve decidedly sophisticated cognitive functions. In an earlier pub
lication on these m atters (1986d), I have show n how a three-lay er
state-sp ace sandw ich can code, and project, the path of a m oving
object in such a fashion as to position the crab's arm to catch the
m oving target on the fly. Evidently, a m ultilayered cortex can offer
considerable advantages.
6 B eyond State-Space Sandwiches
The exam ples studied above are uniform in having an input state
space of only tw o dim ensions and an ou tp u t state space of only tw o
dim ensions. This allows the required coordinate transform ation to be
achieved by a contiguous pair of sheetlike m aps. But w h at of cases
w here the subsystem s involved each have m ore than tw o p aram
eters? W h at of cases w here the coordinate transform ations are from
an inp u t space of n dim ensions to an ou tp u t space of m dim ensions,
w h ere n and m are different and both g reater than tw o? C onsider, for
exam ple, the problem of coordinating the joint angles of a limb with
three or m ore joints, and the problem of coordinating several such
limbs w ith each other. O r consider the problem of coordinating the

even larger num ber of m uscles that collectively control such limbs. As
soon as one exam ines the problem s routinely faced and solved by real
creatu res, one appreciates that m any of them are far m ore com plex
than can be rep resented by a simple tw o-dim ensional to tw o-
dim ensional transform ation.
P erh ap s som e of these m ore com plex problem s m ight be solved by
dividing them into a set of sm aller ones, problem s that can be m an
aged after all by a set of distinct tw o-dim ensional state-space san d
w iches, each addressing som e slice or asp ect of the larger problem
(for som e specific suggestions in this vein, see Ballard 1986). The p re
d om inance of lam inar cortex in the brain certainly en cou rages sp e
culation along these lines. But such solutions, even approxim ate
ones, cannot in general be guaranteed. The brain badly needs som e
m echanism beyond the state-space sandw ich if it is routinely to h an
dle these higher-dim ensional problem s.
Pellionisz and Llinas have already outlined a m echanism adequate
to the task, and have found im pressive evidence of its im plem enta
tion w ithin the cerebellum . The cerebellum is the large structure at
the rear of the brain just und ern eath the cerebral hem ispheres. Its
principal function, divined initially from lesion studies, is the coor
dination of com plex bodily m ovem en ts, such as w ould be displayed
in preparing a dinner or in playing basketball. It displays a neural
organization quite different from that of the cerebral hem isp h eres, an
organization w hose significance m ay be ren d ered tran sp aren t by the
accou n t of Pellionisz and Llinas.
To illustrate this m ore general m echanism for coordinate tran s
form ation, let us consider an input system of four dim ensions w hose
inputs a, b, c, d, are transform ed into the values x , y, z, of a three-
dim ensional outp ut system . A s before, the inputs and outp u ts can
each be regard ed as points in a suitable state space. Since they are
n-tuples, each can also be regarded as a v ector (w hose base lies at the
origin of the relevant state space and w hose arrow head lies at the
point specified by the n-tuple).
A standard m athem atical operation for the system atic transform a
tion of v ectors into vectors is m atrix m ultiplication. H ere it is the
m atrix that em bodies or effects the desired coordinate transform ation.
To see how this w orks, consider the m atrix of figure 5.8 , w hich has
four row s and three colum ns.
To m ultiply the input vector <a, b, c, d> by this m atrix, w e multiply
a tim es р г, b tim es p2, c tim es p 3l d tim es p4, and then sum the four
results to yield x. W e then rep eat the p rocess w ith the second colum n
Figure 5.8
Vector-to-vector transformation by matrix multiplication
Figure 5.9
Effecting vector-to-vector transformations with a neural net
to yield y, and again w ith the third colum n to yield z. Thus results the
ou tp u t v ector <x, y, z>.
This algebraic operation can be physically realized quite sim ply by
the neural array of figure 5 .9 . The parallel input fibers at the right
each send a train of electrochem ical "sp ik es" tow ard the w aiting
dendritic trees. The numbers a, b, c, d represent the am ount by which
the m o m en tary spiking frequency of each of the four fibers is above
(positive num ber) or below (negative num ber) a certain baseline spik
ing frequency. The top -m ost input fiber, for exam ple, syn ap ses onto
each of the three outpu t cells, m aking a stim ulatory connection in
each case, one that tends to depolarize the cell body and m ake it send
a spike dow n its vertical output axon. The output frequency of spike
em issions for each cell is determ ined by the simple frequ en cy of input
stim ulations it receives from all incom ing synaptic con n ection s, and
by the w eight or strength of each synaptic connection, w hich is d eter

m ined by the placem ent of the synapses and by their cross-sectional
areas. These strength values are individually rep resented by the
coefficients of the m atrix of figure 5 .8 . The neural interconnectivity thus
im plem ents the m atrix. Each of the three cells of figure 5 .9 "s u m s " the
stim ulation it receives, and em its an appropriate train of spikes dow n
its outp u t axon. Those three outp u t frequencies differ from the back
ground or baseline frequencies of the three outp u t cells by positive
or negative am ounts, and these am ounts correspond to the output
vector <x, y, z>.
N ote that w ith state-space sandw iches the coding of inform ation is
a m atter of the spatial location of neural events. By contrast, w ith the
m atrix-m ultiplication style of com putation un d er discussion, input
and outp u t variables are coded by sets of spiking frequencies in the
relevant p ath w ays. The form er system uses spatial coding; the latter
system uses frequency coding. But both system s are engaged in the
coordinate transform ation of state-space positions.
The exam ple of figure 5 .9 con cern s a three-by-four m atrix. But it
is evident that neither the m athem atical operation n or its p hysi
cal realization suffers any dim ensional lim itations. In principle, a
Pellionisz-Llinas connectivity m atrix can effect transform ations on
state spaces of a dim ensionality into the thou san d s and beyond.
The schem atic architecture of figure 5.9 corresp on d s very closely to
the style of m icroorganization found in the cerebellum (figure 5 .1 0 ).
(For an accessible sum m ary of cerebellar architecture, see Llinas
1975). The horizontal fibers are there called 'parallel fibers', and they
input from the higher m otor cen ters. The bushy vertical cells are there
called 'Purkinje cells', and they output throu gh the cerebellar n u
cleus to the m otor periphery. In fact, it w as from the observation of the
cerebellum 's beautifully regular architecture, and from the attem p t to
re-create its functional properties by m odeling its large-scale physical
connectivity w ithin a com p u ter, that Pellionisz and Llinas (1979) w ere
originally led to the view that the cerebellum 's job is the system atic
transform ation of vectors in one neural hypersp ace into v ectors in
anoth er neural h yperspace.
G iven that view of the problem , the tensor calculus em erges as the
natural fram ew ork w ith w hich to add ress such m atters, especially
since w e can n ot exp ect the brain to limit itself to C artesian coordi
nates. In the exam ples discussed so far, variation in position along
any axis of the relevant state space is ind ep en d en t of variation along
any of the oth er axes, but this ind ependence will not characterize
state spaces w ith nonorthogonal axes. Indeed, this generalization of
the ap p roach , to include non-C artesian h yp ersp aces, is regarded by
P A R A L LE L F IB E R S MAKE
M U L TIP LE SYNAPSES
PUR KINJE A X O N O U TP U T
Figure 5.10
A schematic section of the cerebellum (cell population and fiber density reduced for
clarity)
Pellionisz and Llinas as one of the m ost im portant features of their

accoun t, a feature that is essential to understanding all b u t the sim
plest coordination problem s. U nfortunately, I cannot pursu e this fea
ture here.
Three final points about the neural m atrix of figure 5 .9 . First, it
need n ot be limited to com puting linear transform ations. The indi
vidual synaptic connections m ight rep resent any of a broad range of
functional properties. They need not be simple m ultipliers. In concert
then, they are capable of com puting a large variety of nonlinear tran s
form ations. Second, a neural m atrix will have the sam e extraordinary
speed displayed by a state-space sandw ich. A nd third, given large
m atrices an d /o r cell red u nd ancy, such structures will also display
a functional persistence despite the scattered loss of their cellular
com p onen ts.
These brief rem arks do not do justice to the very extensive w ork of
Pellionisz and Llinas, nor have I explored any criticisms. The reader
m ust tu rn to the literature for deeper instruction. (For criticism s, see
Arbib and A m ari 1985. For a reply, see Pellionisz and Llinas 1985.)
The principal lesson of this section is that the general functional
sch em a being advanced in this p ap er— the schem a of rep resen ta
tion by state-sp ace position, and com putation by coordinate
transform ation— does not encou n ter im plem entational difficulties
w hen the representational and com putational task exceeds the case of
tw o dim ensions. O n the contrary, the brain boasts neural m achinery
that is ideally suited to cases of v ery high dim ensionality. W e have,
then, at least tw o know n brain m echanism s for perform ing coordi
n ate transform ations: the state-space sandw ich specifically for
tw o-dim ensional cases, and the neural m atrix for cases of any
dim ensionality w hatever.
7 The R epresentational Pow er o f State Spaces
D iscussion so far has been concentrated on the im pressive com puta

tional pow er of coordinate transform ations of state spaces and on the
possible neural im plem entation of such activity. But it is im portant to
appreciate fully the equally pow erful representational capacity of n eu
ral state spaces. The global state of a com plex system of n distinct
variables can be econom ically represented by a single point in an ab
stract n-dim ensional state space. A nd such a state-space point can be
neurally im plem ented, in the sim plest case, by a specific distribution
of n spiking frequencies in a system of only n distinct fibres. M ore
over, a state-space representation em bodies the metrical relations
b etw een distinct possible positions w ithin it, and thus em bodies the
representation of sim ilarity relations betw een distinct item s thus rep
resented. Five exam ples will illustrate these claims, all of w hich m ay
be real, and three of w hich pose problem s familiar to philosophers.
The qualitative character of our sensations is com m only held to
pose an especially intractible problem for any neurobiological red u c
tion of m ental states (see N agel 1974; Jackson 1982; Robinson 1982).
A nd it is indeed hard to see m uch room for reductive purchase in the
subjectively discriminable but "objectively u n ch aracterizable" qualia
p resen t to consciousness.
Even so, a determ ined attem p t to find ord er rath er than m ystery in
this area u n covers a significant am ou nt of expressible inform ation.
For exam ple, w e will all agree that the color qualia of our visual sen sa
tions arrange them selves on a continuum . W ithin this continuum of
properties there are similarity relations (orange is similar to red), rela
tive similarity relations (orange is m ore similar to red than to purple),
and b etw een n ess relations (orange is b etw een red and yellow ). There
are also an indefinite num ber of distinct "p a th s " through continuous
ly similar colors that will take us from any given color to a different
color.
To this w e can add that people w ho suffer one or another of the
various types of color blindness appear to em body a significantly re
duced continuum of color qualia, one reduced in at least partially sp e
cifiable w ays (it fails to display a con trast betw een red and green, or
betw een blue and yellow , etc.). This question of the relative variety of
qualia displayed w ithin a given m odality raises the point that across the
familiar five m odalities there is notew orth y variation. F or exam ple,
thou gh the variety of discriminable color sensations is large, the variety
of discrim inable taste sensations is even larger, an d the vari
ety of discriminable smell sensations is larger still. Such variation
rem inds us further of the presum ed variation across species, as in
stanced in the canine's extraordinary ability to discrim inate, by smell
alone, any one of the 3.5 billion people on the planet. O ne presu m es
that the canine's continuum of olfactory sensations is som eh ow m uch
"la rg e r" than a h u m an 's, in the sense of containing a g reater variety
of discrim inable types of sensation.
H ere, then , are som e hu m dru m facts about the m anifold(s) of sub
jective sen sory qualia, facts w hich a reductive account of m ind m ight
attem p t to explain. It m ust do this by reconstructing these facts, in
som e revealingly system atic w ay, in neurobiological term s. (For a
general accoun t of the nature of cross-theoretic identities and inter
theoretic red u ction s, see chapter 3 and P. M. C hurchland 1979.)
This possibility will now be explored. For several of the relevant
m odalities, physiological and cognitive psychologists have already
sketched the outlines of such an account, and state-sp ace rep resen
tations play a p rom inent role in all of them .
C onsid er first the abstract three-dim ensional "co lo r cu b e" p ro
posed by Edwin Land (1977), within which every one of the m any hun
dreds of hum anly discriminable colors occupies a unique position or
small volum e (figure 5.11). Each axis rep resents the eye/brain's recon
struction of the objective reflectance of the seen object at one of the
three w avelengths to w hich our cones are selectively responsive. Two
colors are closely similar just in case their state-space positions w ithin
this cube are close to one another. A n d tw o colors are dissim ilar just
in case their state-space positions are distant. W e can even speak of
the d egree of the similarity, and of the dim ensions along w hich it is
reckon ed . (See also Zeki 1983.)
If the h u m an brain does possess an internal im plem entation of
such a state sp ace, it has purchased a great deal of representational
p ow er at a very low price. For exam ple, if our native discrim ination
along each axis of L an d 's color state space is only 10 distinct posi
tions, then a ternary system should be able to rep resent fully 103 dis
tinct colors. If anything, this underestim ates our capacities, so the
assu m p tion of 10-unit axial discrim ination is likely too low. In any
case, there is no trouble accounting for our broad discrim inatory p ow
ers: o n e's discrim ination w ithin L an d 's state space explodes as the
Figure 5.11
Color state space
third p ow er of one's discrim ination along each axis. A nd certainly our

peripheral m achinery tends to bear out the general h ypothesis. All
color perception arises from the inputs of exactly three kinds of retinal
cones.
All of w hich suggests the hypothesis that a visual sensation of any
specific color is literally identical w ith a specific triplet of spiking fre
quencies in som e triune brain system . If this is true, then the sim ilar
ity of tw o color sensations em erges as just the proxim ity of their
respective state-sp ace positions. Q ualitative betw een n ess falls out
as state-sp ace betw een ness. A n d , of cou rse, there are an indefinite
num ber of continuous state-space paths connecting any tw o state-
space points. Evidently, w e can reconceive the cube of figure 5.11 as
an internal "qualia cu b e." Just think of each axis as rep resenting the
instantan eou s activity level, or spiking frequency, of one of the three
internal p athw ays for reflectance inform ation.
Finally, if genetic m isadventure should deprive a h u m an of one of
the stan d ard three p athw ays, then that p erso n 's qualia space should
collapse to one of three possible tw o-dim ensional spaces according to
w hich of the three axes is rend ered inoperative. A specific and p re
dictable deficit in color discrim ination should accom p an y each loss.

A nd so it does. There are three principal types of color blindness,
each correspon d in g to the loss of one of the three types of cones in
the retina. W h at w e have here is the outline of a genuinely reductive
accou n t of one dom ain of sensory qualia.
O u r gustatory system appears to exploit a similar arran gem en t,
although here the dim ensionality of the state space w ould ap p ear to
be four, since that is the num ber of distinct kinds of taste receptors in
the m ou th . A n y hum anly possible taste sensation, it is therefore co n
jectu red , is a point som ew here w ithin a four-dim ensional gustatory
state sp ace. O r m ore literally, it is a quadruple of spiking frequencies
in the four p roprietary pathw ays carrying inform ation from the g usta
tory receptors for distribution to the rest of the brain. If our discrim
ination along each axis is com parable to that w ithin color space (10
or m ore units per axis), this m eans that the variety of different taste
sensations will be greater than the variety of different color sensations
by roughly an ord er of m agnitude. A nd so it seem s. This state-space
ap p roach to gustatory sensations appears in the neuroscience litera
ture as the "across-fiber pattern th eo ry " (Bartoshuk 1978; Smith 1983;
Pfaff 1985).
This insight into gustatory space allows us to say som ething d eter
m inate about "w h a t it is like" to be a rat, and a cat. Like h u m an s, rats
and cats are m am m als, and they also possess a four channelled g usta
tory sy stem . O ne difference m erits m ention, h ow ever. O ne of the
four p ath w ay s— som etim es labelled the "b itte r" p ath w ay, since the
four-value code for bitterness requires a high level of activity in that
p ath w ay — show s a different sensitivity across the three species. In
rats, this pathw ay show s a narrow er ran ge of evocable activity (it is
less discrim inating) than it does in hu m an s. In cats, it sh ow s a w ider
range of activity (it is m ore discrim inating).
C onsider now the slight contrast, in h u m an s, b etw een the taste
of su gar (sucrose) and the taste of saccharin. Sugar is generally p re
ferred b ecause saccharin has a faintly bitter aspect to it. The p reced
ing inform ation about rats and cats suggests that in rats this difference
in resp ect of bitterness will be sm aller than it is in h u m an s, and that
in cats the difference will be larger. Saccharin, that is, should taste
rath er m ore bitter to cats than it does to us. O r so the preceding
w ould su ggest. A s it hap pens, the choice behavior of rats does not
discrim inate betw een sugar and saccharin: they will eat either indis-
crim inantly. C ats, by con trast, will eat sugar, but reject saccharin
(Bartoshuk 1978).
A n accou n t of this sam e general kind m ay hold for our olfactory
system , w hich has six or m ore distinct types of recep tor. A six
dimensional space has greater volume still, and will permit even great
er feats of discrim ination. A six-dim ensional space, at 10-unit axial
discrim ination, will perm it the discrim ination of 106 odors. A nd if w e
im agine only a seven-dim ensional olfactory space, w ith only three
tim es the h um an axial discrim ination, w hich space a dog alm ost cer
tainly possesses, then w e are contem plating a state space w ith 307, or
22 billion, discriminable positions! W ith such a space, a canine's abil
ity to distinguish by smell any one of the 3 .5 billion people on the
planet no longer p resents itself as a m ystery.
I have neither the space nor the understan din g to discuss the com
plex case of auditory qualia, but here too a state-space ap proach is
claim ed to be illuminating (see Risset and W essel 1982). D epending
on the research ers and the m odality involved, the state-space
approach is variously called 'm ultivariate analysis', 'm ultidim en
sional scaling', 'across-fiber pattern coding', 'vector coding', and so
forth. But these are all alternative incarnations of the sam e thing:
state-space representations.
Evidently, this approach to understanding sensory qualia is both
theoretically and empirically m otivated, and it lends support to the
reductive position advanced in ch apter 3 on the ontological status of
sen sory qualia. In particular, it suggests an effective m eans of ex
pressing the allegedly inexpressible. The "ineffable" pink of one's
cu rrent visual sensation m ay be richly and precisely expressible as a
95H z/80H z/80H z "c h o rd " in the relevant triune cortical system . The
"u n co n v ey ab le" taste sensation produced by the fabled A ustralian
health tonic V egam ite m ight be quite poignantly conveyed as a 85/80/
90/15 "c h o rd " in one's four-channeled gustatory system (a dark co r
ner of taste-space that is best avoided). A n d the "indescribable" olfac
tory sensation produced by a new ly opened rose m ight be quite
accurately described as a 95/35/10/80/60/55 "c h o rd " in som e six
dim ensional system w ithin on e's olfactory bulb.
This m ore penetrating conceptual fram ew ork m ight even displace
the com m on sen se fram ew ork as the vehicle of intersubjective d e
scription and spontaneous introspection. Ju st as a m usician can learn
to recognize the constitution of heard m usical chords, after internaliz
ing the general theory of their internal stru ctu re, so m ay w e learn to
recognize, introspectively, the n-dim ensional constitution of our sub
jective sensory qualia, after having internalized the general theory of
their internal structure. This analogy has the further advantage of
preem p ting the predictable response that such a reconception of the
"in ternal w orld " w ould rob it of its beauty and peculiar identity. It
w ould do so no m ore than reconceiving m usical p h enom ena in term s
of harm onic theory robs m usic of its beauty and peculiar identity. On
the co n trary, such reconception opens m an y aesthetic doors that

w ould otherw ise have rem ained closed.
The familiar "ineffable qualia" are continuous, I believe, w ith fea
tures that clearly do divide into com p onen ts. C onsider the hum an
"m o d u le " for facial recognition. W e apparently have one, since the
specific ability to recognize faces can be destroyed by specific right
parietal lesions. H ere it is plausible to suggest that there is an internal
state-sp ace rep resentation of p erhaps 20 dim ensions, each coding
som e salient facial feature such as nose length, facial w idth, etc.
(Police Identi-kits attem p t to exploit such a system , w ith som e su c
cess.) E ven if discrim ination along each axis w ere limited to only 5
distinct positions, such a high-dim ensional space w ould still have an
en orm ou s volum e (520 positions), and it w ould perm it the discrim ina
tion an d recognition of billions of distinct faces. It w ould also em body
similarity relations, so that close relatives could be successfully
g rou p ed , and so that the sam e person could be reidentified in photos
taken at different ages. C onsider tw o p hotos of the y ou n g and the old
Einstein. W hat m akes them similar? They occupy proxim ate positions
in o n e's facial state space.
Finally, let us turn to a m otor exam ple, and let us consider one's
"b o d y im a g e ," on e's continuously updated sense of o n e's overall
bodily configuration in space. That configuration is constituted by the
sim ultaneous position and tension of several hu n dreds of m uscles,
and o ne m onitors it all quite successfully, to judge from the sm ooth
coordination of m ost of on e's m ovem ents. H ow does one do it? W ith
a high-dim ensional state space, according to the theories of Pellionisz
and Llinas, w ho ascribe to the cerebellum the job of com puting
appropriate transform ations am ong high-dim ensional codings of
actual and intended m otor circum stances, codings lodged in the in
put parallel fibers and the output Purkinje axons.
Som e of the possibilities here can be evoked by a very sim ple exam
ple. C on sid er a highly com plex and critically orch estrated periodic
m otion, such as occurs in feline locom otor activity (figure 5.12й). C on
sider n ow a three-dim ensional joint-angle m otor state space for the
cat's hind limb, a space in w hich every possible configuration of that
limb is represented by a point, and every possible m ovem en t is rep re
sented by a continuous path . The graceful step cycle of the galloping
cat will be very econom ically rep resented by a closed loop in that
joint-angle state space (figure 5.12b). If the relevant loop is specified
or "m a rk e d " in som e w ay, then the aw esom e task of coordinated
locom otion reduces to a simple tracking problem : m ake y ou r m otor
state-space position follow the path of that loop.
Figure 5.12
A state-space representation of complex locomotor activity
W h eth er anything in the brain an sw ers to this suggestion is m oot.

But exploration of this technique, w ith the ultim ate aim of using p ort
able m icrocom puters as an artificial m ean s of generating effective
locom otor activity in paraplegics, is und er w ay in the CNS Lab of
L arry Jordan at the University of M anitoba.
O nce w e have taken the step beyond the cognitive significance of
points in tw o-dim ensional state space to the cognitive significance of
lines and closed loops in «-dim ensional state sp aces, it seem s possible
that w e will also find cognitive significance in surfaces, and hyper sur
fa c es, and intersections of h ypersurfaces, and so forth. W hat w e have
opening before us is a "g eo m etrical,” as opp osed to a narrow ly sy n
tactic, conception of cognitive activity.
8 Concluding R em arks
W e have seen how a representational schem e of this kind can

accoun t, in a biologically realistic fashion, for a num ber of im portant
features of m otor control, sensory discrim ination, and sensorim otor
coordination. But has it the resou rces to accou n t for the so-called
higher cognitive activities, as represented by language use, for exam

ple, and by our propositional know ledge of the w orld in general?
C onceivably, yes. O ne m ight try to find, for exam ple, a w ay of
rep resentin g "an glop h on e linguistic h y p ersp ace" so that all gram
m atical sen ten ces turn out to reside on a proprietary hypersurface
w ithin that hyp ersp ace, w ith the logical relations b etw een them re
flected as spatial relations of som e kind. I do not know how to do
this, of course, b ut it holds out the possibility of an alternative to, or
potential reduction of, the familiar C hom skyan picture.
A s for the set of beliefs that is com m only supposed to constitute
a p erso n 's know ledge, it m ay be that a geom etrical rep resentation
of sen tences will allow us to solve the severe problem of tacit belief
(D ennett 1975; Lycan 1985). Just as a hologram does not "c o n ta in " a
large num ber of distinct three-dim ensional im ages, curiously
arranged so as to p resen t a sm oothly changing picture of a real ob
ject as the hologram is view ed from different positions; so m ay
h u m an s not "co n ta in " a large num ber of distinct beliefs, curiously
arranged so as collectively to p resent a coh eren t accou n t of the
w orld.
Perh ap s the truth is rather that in both cases a specific im age or
belief is just an arbitrary projection or "slice " of a d eep er set of data
stru ctu res, and the collective coherence of such sam ple slices is a sim
ple consequence of the m an n er in w hich the global inform ation is
stored at the deep er level. It is not a consequence of, for exam ple, the
b usyw ork of som e fussy inductive m achine applying inductive rules
for the accep tan ce or rejection of discreet slices taken singly. W hich
m eans that, to un d erstan d learning, w e m ay have to u n d erstan d the
forces that dictate directly the evolution of the global d ata structures
at the d eep er level.
These highly speculative rem arks illustrate one direction of re
search suggested by the theory outlined in this paper. Ju st w h at are
the abstract representational and com putational capacities of a system
of state spaces interacting by coordinate transform ations? C an w e use
it to articulate m odels for the "h ig h e r" form s of cognitive activity?
The theory also begs research in the opposite direction, tow ard the
neurophysiology of the brain. Since the brain is definitely not a
general-purpose m achine in the w ay that a digital com p u ter is, it m ay
often turn out that, once w e are prim ed to see them , the brain's local
ized com putational tactics can simply be read off its m icrostru ctu re.
There is point, therefore, to studying that m icrostructure. (For an ac
cessible review of cognitive neurobiology, see P. S. C hurchland 1986.)
Taken jointly, the prodigous representational and com putational
capacities of a system of state spaces interacting by coordinate trans
form ations suggest a pow erful and highly general m eans of u n d er
standing the cognitive activities of the n ervou s system . Especially
since the physical m echanism s appropriate to im plem ent such a sys
tem are w idespread throughou t the brain.
Chapter 6
Folk Psychology and the Explanation of
Human Behavior
Folk p sych ology, insist som e, is just like folk m echanics, folk therm o
dynam ics, folk m eteorology, folk chem istry, and folk biology. It is a
fram ew ork of concepts, roughly adequate to the d em an d s of every
day life, w ith w hich the hum ble adept com p reh en d s, explains, pre
dicts, and m anipulates a certain dom ain of p h enom en a. It is, in short,
a folk theory. A s w ith any theory, it m ay be evaluated for its virtues or
vices in all of the dim ensions listed. A n d as w ith any theory, it m ay be
rejected in its entirety if it fails the m easure of such evaluation. Call
this the "th eoretical v iew " of our self understanding.
Folk p sych ology, insist oth ers, is radically unlike the exam ples
cited. It does n ot consist of laws. It does n ot support causal explana
tions. It does not evolve over tim e. Its central purpose is norm ative
rath er than descriptive. A nd thus, it is not the sort of fram ew ork that
m ight be show n to be radically defective by sheerly em pirical find
ings. Its assim ilation to theories is just a m istake. It h as nothing to
fear, therefore, from advances in cognitive theory or the neurosci
en ces. Call this the "antitheoretical v iew " of our self u n d erstanding.
Som ebody here is deeply m istaken. The first burden of this p aper is
to argue that it is the antitheoretical view that harbors m ost, though
not all, of those m istakes. In the thirty y ears since the theoretical view
w as introduced (see especially Sellars 1956; Feyerabend 1963a; Rorty
1965; P. M . C hurchland 1970, 1979; and ch apter 1), a variety of objec
tions have b een leveled against it. The m ore interesting of those will
be ad d ressed shortly. M y current view is that these objections m oti
vate no ch an ges w h atever in the theoretical view .
The secon d and m ore im portant burd en of this pap er, h ow ever, is
to outline and repair a serious failing in the traditional expressions of
the theoretical view , m y ow n expressions included. The failing, as I
see it, lies in representing on e's com m on sen se understan din g of hu-
An abridged version of this paper first appeared in Proceedings o f the Aristote

lian Society, supplem entary vol. 62 (1988).
m an nature as consisting of an internally stored set o f gen eral sentences,

and in rep resenting one's predictive and explanatory activities as
being a m atter of deductive inference from those sentences plus o cca
sional prem ises about the case at hand.
This certainly sounds like a m ajor concession to the antitheoretical
view , but in fact it is not. For w hat m otivates this reappraisal of the
ch aracter of our self understanding is the gathering conviction that
little or none of h um an understanding consists of stored sentences,
not even the prototypically scientific understan din g em bodied in a
practicing physicist, chem ist, or astron om er. The familiar conception
of know ledge as a set of propositional attitudes is itself a central
aspect of the fram ew ork of folk p sych ology, according to the re
appraisal at h and, and it is an aspect that needs badly to be replaced.
O ur self u nd erstanding, I continue to m aintain, is no different in
character from our understan din g of any oth er empirical dom ain. It is
speculative, system atic, corrigible, and in principle replaceable. It is
just not so specifically linguistic as w e have chronically assum ed.
The speculative and replaceable ch aracter of folk p sychology is now
som ew hat easier to defend than it w as in the sixties and seventies,
b ecause recent advances in connectionist AI and com putational
neuroscien ce have provided us w ith a fertile new fram ew ork with
w hich to un derstand the perception, cognition, and behavior of in
telligent creatu res. W heth er it will eventually prove adequate to the
task of replacing folk psychology rem ains to be seen, but the m ere
possibility of system atic alternative conceptions of cognitive activity
and intelligent behavior should no longer be a m atter of dispute.
A lternatives are already in the m aking. L ater in the p ap er I shall out
line the m ain features of this novel fram ew ork and explore its signif
icance for the issues here at stake. For n ow , let m e acquiesce in the
folk-psychological conception of know ledge as a system of beliefs or
similar propositional attitudes, and try to m eet the objections to the
theoretical view already outstanding.
1 Objections to the T heoretical View
A s illustrated in m y 1970, 1979, and 1984, a thorough perusal of the

explanatory factors that typically appear in our com m onsense ex
planations of our internal states and our overt behavior sustains the
quick "reco n stru ctio n " of a large num ber of universally quantified
conditional statem en ts, conditionals w ith the conjunction of the
relevant explanatory factors as the an teced en t and the relevant
exp lanand u m as the consequent. It is these universal statem ents that
are supposed to constitute the "la w s" of folk psychology.
Folk Psychology and the Explanation of Behavior 113
A perennial objection is that these generalizations do n ot have the

ch aracter of genuine causal/explanatory law s; rath er, they have som e
other, less em pirical status (e .g ., that of norm ative principles or rules
of language or analytic truths). W ithout confronting each of the m any
alternatives in turn, I think w e can m ake serious difficulties for any
objection of this sort.
N ote first that the concepts of folk p sychology divide into two
broad classes. O n the one hand, there are those fully intentional con
cepts exp ressing the various propositional attitudes, such as belief
and desire. A nd on the other hand, there are those nonintentional or
quasi-intentional concepts expressing all of the other m ental states,
such as grief, fear, pain, hun ger, and the full range of em otions and
bodily sensations. W h ere states of the latter kind are con cern ed , I
think it is hardly a m atter for dispute that the com m on homilies
in w hich they figure are causal/explanatory laws. C onsider the
following.
• A p erson w ho suffers severe bodily dam age will feel pain.

• A p erson w ho suffers a sudden sharp pain will w ince.
• A p erson denied food for any length will feel hun ger.
• A h u n gry p erso n 's m ou th will w ater at the smell of food.
• A p erson w ho feels overall w arm th will tend to relax.
• A p erson w ho tastes a lem on will have a puckering sensation.
• A p erson w ho is angry will tend to be im patient.
Clearly these hum ble generalizations, and thousands m ore like them ,
are causal/explan atory in character. They will and regularly do sup
p ort sim ple explanations, sustain subjunctive and counterfactual
conditionals, and underw rite predictions in the standard fashion.
M oreover, con cep ts of this simple sort carry perhaps the m ajor p art of
the folk-psychological burden. The com paratively com plex explana
tions involving the propositional attitudes are of central im portance,
but th ey are surrounded by a quotidean whirl of simple explanations
like these, all quite evidently of a causal/explanatory cast.
It w o n 't do, then, to insist that the generalizations of folk psycholo
gy are on the w hole nonem pirical or noncausal in ch aracter. The bulk
of them , and I m ean thousands upon thou san d s of them , are tran s
parently causal or nom ological. The best one can hope to argue is that
there is a central core of folk-psychological concepts w h ose explana
tory role is som ehow discontinuous w ith that of their fellows. The
propositional attitudes, especially belief and desire, are the perennial
candidates for such a nonem pirical role, for explanations in their
term s typically display the explanandum ev en t as "ra tio n a l." W hat

shall w e say of explanations in term s of beliefs and desires?
W e should tell essentially the sam e causal/explanatory story, and
for the following reason. W h atever else h u m an s do with the concepts
for the propositional attitudes, they do u se them successfully to p re
dict the future behavior of oth ers. This m ean s that, on the basis of
p resum ed inform ation about the current cognitive states of the rel
evant individuals, one can nonaccidentally predict at least som e of
their future behavior som e of the time. But any principle that allows
us to do this— that is, to predict one em pirical state or event on the
basis of an other, logically distinct, em pirical state or event— has to be
em pirical in ch aracter. A nd I assum e it is clear that the event of m y
ducking m y head is logically distinct both from the event of m y p er
ceiving an incom ing snowball, and from the states of m y desiring to
avoid a collision and m y belief that ducking is the best w ay to achieve
this.
Indeed, one can do m ore than m erely predict: one can control and
m anipulate the behavior of others by controlling the inform ation
available to them . H ere one is bringing about certain behaviors by
steering the cognitive states of the subject, by relating opportunities,
d an gers, or obligations relevant to that subject. H ow this is possible
w ithout an understanding of the objective empirical regularities that
con n ect the internal states and the overt behaviors of norm al people
is som ething that the antitheoretical position needs to explain.
The confused tem ptation to find som ething special about the case
of intentional action derives prim arily from the fact that the central
elem ent in a full-blooded action explanation is a configuration of
propositional attitudes in the light of w hich the explanandum be
havior can be seen as sensible or rational, at least from the agent's
n arrow point of view . In this rational-in-the-light-of relation w e seem
to have som e sort of supercausal logical relation betw een the expla-
nans and the explanandum , w hich is an invitation to see a distinct
and novel type of explanation at w ork.
Y et while the prem ise is true— there is indeed a logical relation b e
tw een the explanandum and certain elem ents in the explan an s— the
conclusion does not begin to follow. Students of the subject are still
regularly misled on this point, for they fail to appreciate that a cir
cum stance of this general sort is typical of theoretical explanations. Far
from being a sign of the nonem pirical and h en ce nontheoretical ch ar
acter of the generalizations and explanations at issue, it is one of the
su rest signs available that w e are here dealing w ith a high-grade
theoretical fram ew ork. Let m e explain.
The electric current I in a wire or any con du ctor is causally d eter

m ined by tw o factors: it tends to increase w ith the electrom otive force
or voltage V that m oves the electrons d ow n the w ire, an d it tends to
be red u ced according to the resistance R the wire offers against their
m otion. Briefly, I = V/R. Less cryptically and m ore revealingly,
(x)(V )(R )[(x is subject to a voltage o f(V )) & (x offers a resistance o f(R ))
^ ( 3 I ) ( ( x has a current o f (I)) & (I = V/R))]
The first point to notice here is that the crucial p red icates— has a resist
ance o f(R ), is subject to a voltage o f(V ), and has a current o f (I)— are w hat
m ight be called "n u m erical attitu d es": they are predicate-form ing
functors that take singular term s for num bers in the variable position.
A com plete predicate is form ed only w hen a specific num eral appears
in the relevan t position. The second point to notice is that this elec
trodynam ical law exploits a relation holding on the dom ain of n u m
bers in ord er to exp ress an im portant empirical regularity. The
cu rren t I is the quotient of the voltage V and the resistance R , w hose
values will be cited in explanation of the cu rrent. A nd the third point
to notice is that this law and the explanations it sustains are typical of
law s and explanations throughou t science. M ost of our scientific p re
dicates exp ress num erical attitudes of the sort displayed, and m ost of
our law s exploit and display relations that hold prim arily on the
abstract dom ain of num bers. N or are they limited to num bers. O ther
laws exploit the abstract relations holding on the abstract dom ain of
vectors, or on the dom ain of sets, or groups, or m atrices. But none of
this m ean s they are nonem pirical, or noncausal, or nonnom ic.
A ction explanations, and intentional explanations in general, fol
low the sam e pattern. The only difference is that here the dom ain of
abstract objects being exploited is the dom ain of propositions, and the
relations displayed are logical relations. A nd like the num erical and
vectorial attitudes typical of theories, the expressions for the proposi
tional attitudes are predicate-form ing functors. Believes that P, for
exam ple, form s a com plete predicate only w hen a specific sentence
app ears in the variable position P. The principles that com p reh en d
these p red icates have the sam e abstract and highly sophisticated
stru cture displayed by our m ost typical theories. They just exploit the
relations holding on a different dom ain of abstract objects in order to
exp ress the im portant empirical regularities com p reh en d in g the
states and activities of cognitive creatures. That m akes folk psychol
ogy a v ery interesting theory, p erh ap s, but it is hard ly a sign of
its being nontheoretical. Q uite the reverse is true. (This m atter is dis
cu ssed at greater length in Churchland 1979, section 14, and 1981a,
pp. 8 2 -8 4 .)
In sum , the simpler parts of folk psychology are transparently cau

sal or nom ic in ch aracter, and the m ore com plex parts have the sam e
sophisticated logical structure typical of our m ost pow erful theories.
But w e are not yet done w ith objections. A recurren t com plaint is
that in m any cases the recon stru cted conditionals that p u rp ort to be
sam ple "law s” of folk psychology are either strictly speaking false, or
they border on the trivial by reason of being qualified by various ceteris
paribus clauses. A first reply is to point out that m y position does
not claim that the laws of folk psychology are either true or com plete.
I agree that they are a m otley lot. M y hope is to see them replaced
entirely, and their ontology of states with them . But this reply is not
w holly responsive, for the point of the objection is that it is im plausi
ble to claim the status of an en trench ed theoretical fram ew ork for a
bunch of "la w s" that are as vague, as loose, and as festooned w ith
ceteris paribus clauses as are the exam ples typically given.
I will m ake no attem pt here to defend the ultim ate integrity of the
laws of folk p sychology, for I have little confidence in them m yself.
But this is not w hat is required to m eet the objection. W h at needs
pointing out is that the "law s” of folk theories are in gen eral sloppy,
vague, and festooned with qualifications and ceteris paribus clauses.
W hat the objectors need to do, in order to rem ove the relevant system
of generalizations from the class of empirical theories, is to show that
folk psychology is significantly w orse in all of these respects than are
the principles of folk m echanics, or folk therm od yn am ics, or folk biol
ogy, and so forth. In this they are sure to be disappointed, for these
oth er folk theories are even w orse than folk p sychology (see M cKlos-
key 1983). In all, folk p sychology m ay be a fairly ram shackle theory,
but a theory it rem ains. N or is it a point against this that folk psychol
ogy has changed little or none since ancient tim es. The sam e is true of
other theories near and dear to us. The folk physics of the tw entieth
cen tu ry, I regret to say, is essentially the sam e as the folk physics of
the ancient G reeks (M cKloskey 1983). O ur conceptual inertia on such
m atters m ay be en orm ou s, but a theory rem ains a theory, h ow ever
m any centuries it m ay possess us.
A quite different objection directs our attention to the great m any
things beyond explanation and prediction for which we use the vocab
ulary and concepts of folk p sychology. Their prim ary function, runs
the objection, is not the function served by explanatory theories, but
rath er the m yriad social functions that constitute hum an culture and
com m erce. W e use the resources of folk psychology to p rom ise, to
en treat, to congratulate, to tease, to joke, to intim ate, to threaten , and
so on. (See W ilkes 1981, 1984).
The list of functions is clearly both long and genuine. But m ost of
these functions surely com e un d er the head in g of control or m an
ipulation, w hich is just as typical and central a function of theories as

is either explanation or prediction, but w hich is n ot m entioned in the
list of theoretical functions supplied by the objectors. Though the
im age m ay be popular, the idle m usings of an im potent stargazer
provide a p oor exam ple of w hat theories are and w hat theories do.
M ore typically, theories are the conceptual vehicles w ith w hich w e
literally com e to grips w ith the w orld. The fact that folk psychology
serves a w ealth of practical purp oses is no evidence of its being non-
theoretical. Quite the reverse.
M anipulation aside, w e should not underestim ate the im portance
for social com m erce of the explanations and predictions that folk
p sych ology m akes possible. If one cannot predict or anticipate the
behavior of one's fellows at all, then one can engage in no useful
com m erce w ith them w hatever. A nd finding the right explanations
for their past behavior is often the key to finding the appropriate prem
ises from w hich to anticipate their future behavior. The objection's
attem pt to paint the functions of folk p sychology in an exclusively
nontheoretical light is simply a distortion born of tunnel vision.
In an y case, it is irrelevant. For there is no inconsistency in saying
th at a theoretical fram ew ork should also serve a great m an y n on
theoretical p u rp oses. To use an exam ple I have used before (1986b),
the theory of w itches, dem onic possession, exorcism , and trial by ordeal,
w as also used for a variety of social p u rposes beyond strict explana
tion and prediction. For exam ple, its vocabulary w as used to w arn , to
cen su re, to abjure, to accuse, to badger, to sentence, and so forth. But
none of this m ean t that dem ons and w itches w ere anything other
than theoretical entities, and none of this saved the ontology of d e
m on theory from elim ination w hen its em pirical failings b ecam e acute
and different conceptions of h um an pathology arose to replace it. Be
liefs, desires, and the rest of the folk-psychological ontology all are in
the sam e position. Their integrity, to the extent that th ey have any,
derives from the explanatory, predictive, and m anipulative p row ess
they display.
It is on the topic of explanation and prediction that a further objec
tion finds fault w ith the theoretical view . Precisely w h at, begins the
objection, is the observable behavior that the ontology of folk
psychology is postulated to explain? Is it bodily behavior as kinem ati
cally described? In som e cases, perh ap s, but not in general, certainly,
because m any quite different kinematical sequences could count as
the sam e intentional action, and it is generally the action that is
p roperly the object of folk-psychological explanations of behavior.
In general, the descriptions of hum an behavior that figure in folk-
psychological explanations and predictions are descriptions that
already im ply perception, intelligence, and person h ood on the part of

the agent. Thus, it m ust be w ron g to see the relation b etw een one's
psychological states and on e's behavior on the m odel of theoretical
states postulated to explain the behavior of som e conceptually inde
p en d en t dom ain of p henom ena (H aldane 1988).
The prem ise of this objection is fairly clearly true: a large class of
behavior descriptions are n ot conceptually independent of the con
cep ts of folk p sychology. But this affords no grounds for denying
theoretical status to the ontology of folk psychology. The assum ption
that it does reflects a naive view of the relation b etw een theories and
the dom ains they explain and predict. The naive assum ption is that
the concepts used to describe the dom ain to be explained m u st
alw ays be conceptually independent of the theory used to explain the
p h en om en a w ithin that dom ain. That assum ption is know n to be
false, and w e need look no farther than the special theory of relativity
(STR) for a living counterexam ple.
The introduction of STR b rough t w ith it a system atic reconfigura
tion of all of the basic observational con cep ts of m echanics: spatial
length, tem poral duration, velocity, m ass, m om en tu m , etc. These are
all one-place predicates within classical m echanics, but they are all
replaced by tw o-place predicates w ithin STR. E ach ostensible "p ro p
e rty " has turned out to be a relation, and each has a definite value only
relative to a ch osen reference fram e. If STR is true, and since the early
y ears of this cen tu ry it has seem ed to be, then one cannot legitim ately
describe the observational facts of m echanics save in term s that are
d raw n from STR itself.
M odern chem istry provides a second exam ple. It is a rare chem ist
w h o does n ot use the taxon om y of the periodic table and the co m
binatorial lexicon of chem ical com pounds to describe both the observ
able facts and their theoretical underpinnings alike. For starters, one
can just smell hydrogen sulphide, taste sodium chloride, feel any
b ase, and identify copper, alum inum , iron, and gold by sight.
These cases are not unusual. O ur theoretical convictions typically
reshap e the w ay w e describe the facts to be explained. Som etim es it
h ap p ens im m ediately, as w ith STR, but m ore often it h ap p ens after
long familiarity w ith the successful theory, as is evidenced by the
idiom s casually em ployed in any w orking laboratory. The prem ise of
the objection is true. But it is no point at all against the theoretical
view . Given the great age of folk psychology, such conceptual inva
sion of the explanandum dom ain is only to be expected.
A different critique of the theoretical view proposes an alternative
accou nt of our understanding of h um an behavior. A ccording to this
view , on e's capacity for anticipating and understan din g the behavior
of o th ers resides not in a system of nom ically em bedded concepts,

but rath er in the fact that one is a norm al person oneself, and can
draw o n o n e's ow n reactions, to real or to im agined circum stances, in
order to gain insight into the internal states and the overt behavior of
oth ers. The key idea is that of em pathy. O ne uses oneself as a simula
tion (usually im agined) of the situation of an oth er and then extra
polates the results of that simulation to the person in question (see
G ordon 1986; G oldm an 1989).
M y first response to this line is simply to agree that an enorm ous
am ou nt of o n e's appreciation of the internal states and o vert behavior
of oth er h u m an s derives from one's ability to exam ine an d to extrap o
late from the facts of on e's ow n case. All of this is quite consistent
w ith the theoretical view , and there is no reason that one should
attem pt to deny it. O ne learns from every exam ple of h u m an ity one
en cou n ters, and one encoun ters oneself on a system atic basis. W hat
w e m ust resist is the suggestion that extrapolating from the partic
ulars of on e's ow n case is the fundam ental ground of o n e's u n d er
standing of oth ers, a ground that rend ers possession of a nom ic
fram ew ork u n necessary. Problem s for this stron ger position begin to
app ear im m ediately.
For one thing, if all of o n e's understanding of others is closed un d er
extrapolation from on e's ow n case, then the m od est con ten ts of one's
ow n case m ust form an absolute limit on w hat one can exp ect or ex
plain in the inner life and external behavior of others. But in fact we
are n ot so limited. People w ho are congenitally deaf or blind know
quite well that norm al people have perceptual capacities b eyon d w hat
they them selves p ossess, and they know in som e detail w hat those
capacities entail in the w ay of know ledge and behavior. People
w ho h ave never felt profound grief, say, or love, or rejection, can
n onetheless provide appropriate predictions and explanations of
the b ehavior of people so afflicted. A nd so on. In general, on e's im
m ediately available understanding of h u m an p sych ology and b e
havior goes substantially beyond w hat one has experienced in one's
ow n case, either in real life or in pointed sim ulations. First-person
experience or sim ulation is plainly not necessary for u n d erstan din g the
behavior of others.
N or is it sufficient. The problem is that sim ulations, even if they
m otivate predictions about others, do not by them selves provide any
explanatory understan din g of the behavior of others. To see this, con
sider the following analogy. Suppose I w ere to possess a m arvellous
m iniature of the physical universe, a m iniature I could m anipulate in
order to sim ulate real situations and thus predict and retrod ict the
b ehavior of the real universe. Even if m y m iniature unfailingly p ro
vided accurate sim ulations of the outcom es of real physical processes,

I w ould still be no further ahead on the business of explaining the
behavior of the real w orld. In fact, I w ould then have tw o u niverses,
both in n eed of explanation.
The lesson is the sam e for first-person and third-person situations.
A sim ulation itself, even a successful one, provides no explanation.
W h at explanatory understanding requires is an appreciation of the
gen eral patterns that com p rehend the individual events in both cases.
A n d that brings us back to the idea of a m oderately general theory.
W e should have com e to that idea directly, since the em pathetic
accou n t of our understanding of others d epends crucially on one's
having an initial understanding of oneself. To extrapolate o n e's ow n
cognitive, affective, and behavioral intricacies to others requires that
one be able to conceptualize and spontaneously to recognize those
intricacies in oneself. But on e's ability to do this is left an un ad dressed
m ystery by the em pathetic accou nt. Self-understanding is not seen as
a problem ; it is other-u nderstand in g that is held up as the problem .
But the form er is no less problem atic than the latter. If one is to be
able to appreh en d even the first-person intricacies at issue, then one
m u st possess a conceptual fram ew ork that draw s all of the n ecessary
distinctions, a fram ew ork that organizes the relevant categories into
the appropriate stru ctu re, a fram ew ork w h ose taxon om y reflects at
least the m ore obvious of the rough nom ic regularities holding across
its elem ents, even in the first-person case. Such a fram ew ork is
already a theory.
The fact is, the categories into w hich any im portant dom ain gets
divided by a learning creature em erge jointly w ith an appreciation of
the rou gh nom ic regularities that con n ect them . A nascent taxon om y
th at su p ports the expression of no useful regularities is a taxon om y
that is soon replaced by a m ore insightful one. The divination of u se
ful regularities is the single m ost dom inant force shaping the taxon o
m ies developed by any learning creature in any dom ain. A nd it is an
essential force, even in perceptual dom ains, since our observational
taxonom ies are alw ays radically u n derdeterm ined by our untrained
p erceptual m echanism s. To suppose that on e's conception of one's
own m ental life is innocent of a netw ork of system atic expectations is
just naive. But such a netw ork is already a theory, even before one
ad d resses the issue of others.
This is the cash value, I think, of P. F. S traw son's insightful claim,
n ow thirty years old, that to be in a position to pose any question
about other m inds, and to be in a position to try to construct argu
m en ts from analogy w ith one's ow n case, is already to possess at least
the rudim ents of w hat is sought after, nam ely, a general conception
of m ental p h en om en a, of their general connections w ith each other

and w ith behavior (Straw son 1958). W hat Straw son m issed w as the
further insight that such a fram ew ork is nothing other than an em pir
ical theo ry , one justified not by the quasi-logical character of its prin
ciples, as he attem pted unsuccessfully to show , but by its im personal
success in explaining and predicting h u m an behavior at large. There
is no special justificational story to be told here. Folk p sych ology is
justified by w h at standardly justifies any conceptual fram ew ork:
nam ely, its explanatory, predictive, and m anipulative su ccess.
This concludes m y survey of the outstanding objections to the
theoretical view outlined in the opening paragrap h of the p resent
ch ap ter. But in defending this view there is a m ajor difference be
tw een m y strategy in earlier w ritings and m y strategy h ere. In m y
1970 p ap er, for exam ple, the question w as fram ed as follow s: "A re
action explanations deductive-nom ological explan ation s?" I w ould now
prefer to fram e the question thus: "A re action explanations of the
sam e general type as the explanations typically found in the sci
en ce s?" I continue to think that the answ er to this secon d question
is p retty clearly yes. My reasons are given above. But I am no longer
confident that the deductive-nom ological (D-N) m odel itself is an
adequate accoun t of explanation in the sciences or an yw h ere else.
The difficulties w ith the D -N m odel are detailed elsew here in the
literature, so I shall not pause to sum m arize them here. M y diagnosis
of its failings, h ow ever, locates the basic problem in its attem p t to
rep resen t know ledge and understanding by sets of sen ten ces or
propositional attitudes. In this, the fram ers of the D -N m odel w ere
resting on the basic assum ptions of folk psychology. Let m e close this
p aper b y briefly exploring how w e m ight conceive of know ledge, and
of explan atory un d erstanding, in a system atically different w ay. This
is an im portan t undertaking relative to the concerns of this chapter,
for there is an objection to the theoretical view , as traditionally ex
p ressed, that seem s to m e to have som e real bite. It is as follows.
If o n e's capacity for understanding and predicting the behavior of
others derives from on e's internal storage of thousands of law s or
nom ic generalizations, how is it that one is so p oor at enunciating the
law s o n w hich on e's explanatory and predictive p row ess d ep en ds? It
seem s to take a trained philosopher to recon stru ct them ! H ow is it
that children are so skilled at understanding and anticipating the be
havior of h u m an s in advance of ever acquiring the com plex linguistic
skills n ecessary to express those laws? H ow is it that social h unters
such as w olves and lions can com prehend and anticipate each oth er's
behavior in great detail w h en they presum ably store n o internal
sen ten ces at all?
W e m u st resist the tem ptation to see in these questions a renew ed

m otivation for counting folk p sychology as special, for the very sam e
problem s arise w ith respect to any other folk theory you m ight care to
m ention: folk physics, folk biology, w hatever. It even arises for
theories in the highly developed sciences, since, as Kuhn has pointed
out, very little of a scientist's understan din g of a theory consists in his
ability to state a list of laws. It consists, rather, in the ability to apply
the conceptual resou rces of the theory to new cases, and thus to
anticipate and p erhaps m anipulate the behavior of the relevant
em pirical dom ain. This m eans that our problem here concerns the
ch aracter of know ledge and understan din g in general. Let us finally
ad d ress that problem .
2 A n A lternative Form o f K now ledge Representation
O ne alternative to the notion of a universal generalization about F is

the notion of a prototype of F, a central or typical exam ple of F w hich
all other exam ples of F resem ble, m ore or less closely, in certain
relevant respects. P rototyp es have certain obvious advantages over
universal generalizations. Just as a picture can be w orth a thousand
w ord s, so a single com plex prototype can em body the sam e breadth
of inform ation concerning the organization of cooccurrent features
that w ould be contained in a long list of com plex generalizations.
F u rth erm ore, p rototypes allow us a w elcom e degree of looseness that
is precluded by the strict logic of a universal quantifier: not all F s need
be Gs, but the standard or norm al ones are, and the nonstandard
ones m u st be related by a relevant similarity relation to those that
properly are G. Various theorists have independently found m otive
to introduce such a notion in a num ber of cognitive fields: they have
b een called 'paradigm s' and 'exem plars' in the philosophy of sci
ence (K uhn 1962), 'stereotypes' in sem antics (Putnam 1970, 1975),
'fram es' (M insky 1981) and 'scripts' (Schank 1977) in AI research , and
finally 'p rototyp es' in p sychology (Rosch 1981) and linguistics (Lakoff
1987).
Their ad vantages aside, p rototypes also have certain familiar p rob
lem s. The first problem is how to determ ine just w hat clutch of ele
m en ts or properties should constitute a given prototyp e, and the
secon d problem is how to determ ine the m etric of similarity along
w hich closeness to the central prototyp e is to be m easured. Though
they pose a problem for notions at all levels, these problem s are
especially keen in the case of the so-called "b a sic" or "sim p le" p ro p er
ties, because com m on sense is there unable even to articulate any
deeper constituting elem ents (for exam ple, w hat elem ents "m ak e u p "
a p urple color, a sour taste, a floral sm ell, or the p h on em e /a /?).

A final problem concerning prototypes is a familiar one: how m ight
p rototyp es be effectively represented in a real cognitive creature?
This last question brings m e to a possible answ er, and to a p ath that
leads to further answ ers. The relevant research co n cern s the o pera
tions of artificial neural netw orks, netw orks that m im ic som e of the
m ore obvious organizational features of the brain. It co n cern s how
they learn to recognize certain types of com plex stimuli, and how
they rep resen t w hat they have learned. U pon repeated presentation
of various real exam ples of the several features to be learned (F, G, H ,
e tc.), and un d er the steady pressure of a learning algorithm that
m akes small adjustm ents in the netw ork 's synaptic con n ection s, the
netw ork slowly but spontaneously generates a set of internal rep re
sentations, one for each of the several features it is required to rec
ognize. Collectively, those representations take the form of a set or
system of similarity spaces, and the central point or volum e of
such a space constitutes the netw ork's rep resentation of a p rototypi
cal F, G, or H . After learning is com pleted, the system resp on d s to
any F-like stim ulus w ith an internal p attern of neuronal activity that
is close to the prototypical p attern in the relevant sim ilarity space.
The netw ork consists of an initial “ sen so ry " layer of n eurons,
w hich is m assively connected to a second layer of n eu ron s. The sizes
or “w eig h ts" of the m any connections determ ine how th e n eu ron s at
the secon d layer collectively respond to activity across the input layer.
The n eu ron s at the second layer are connected in turn to a third layer
(and p erh ap s a fourth layer, etc., but I shall limit the discussion here
to three-layer networks). During learning, w hat the system is search
ing for is a configuration of w eights that will turn the n eu ron s at
the second layer into a set of com plex fea tu re detectors. W e then w ant
the n eu ron s at the third or "o u tp u t" layer to respond in turn to the
secon d layer, given any F-like stimuli at the input layer, w ith a ch ar
acteristic p attern of activity. All of this is achieved by presen ting the
n etw ork w ith diverse exam ples of Fs, and slowly adjusting its con
nection w eights in the light of its initially chaotic respon ses.
Such netw orks can indeed learn to recognize a w ide variety of sur
prisingly subtle features: phonem es from voiced speech, the shapes
of objects from grey-scale photos, the correct pronunciation of
printed English text, the presence of metallic m ines from sonar re
turns, and gram m atical categories in novel sentences. G iven a su c
cessfully trained netw ork, if w e exam ine the behavior of the neurons
at the secon d or interm ediate layer during the process of recognition,
w e discover that each neuron has com e to rep resent, b y its level of
activity, som e distinct aspect or dim ension of the input stim ulus.
124 The' Nature of Mind
Taken togeth er, their joint activity constitutes a m ultidim ensional

analysis of the stimuli at the input layer. The trained netw ork has
succeeded in finding a set of dim ensions, an abstract space, such that
all m ore-or-less typical Fs produce a characteristic profile of neuronal
activity across those particular dim ensions, while deviant or de
g raded Fs produce profiles that are variously close to that central p ro
totyp e. The job of the third and final layer is then the relatively simple
one of distinguishing that profile-region from other regions in the
larger space of possible activation p attern s. In this w ay do artificial
neural netw orks generate and exploit p rototyp es. It is now m ore than
a suggestion that real neural netw orks do the sam e thing. (For a su m
m ary of these results and how they bear on the question of theoretical
know ledge, see Churchland 1989a. For a parade case of successful
learning, see Rosenberg and Sejnowski 1987. For the locus classicus
concerning the general technique, see R um elhart et al. 1986.)
N otice that this picture contains an sw ers to all three of the prob
lem s about prototyp es noted earlier. W h at dim ensions go into a
p rototyp e of F ? Those that allow the system to respon d to diverse ex
am ples of F in a distinctive and uniform w ay, a w ay that red u ces the
error m essages from the learning algorithm to a m inim um . H ow is
similarity to a prototype m easured? By geom etrical proxim ity in the
relevant param eter space. H ow are prototyp es rep resented in real
cognitive creatu res? By canonical activity p attern s across an ap propri
ate population of neurons.
N ote also that the objective features recognized by the netw ork can
also have a tem poral com p onent: a netw ork can just as well be
trained to recognize typical sequences and processes as to recognize
atem poral p attern s. W hich brings m e to m y final suggestion. A n or
m al h u m an 's understan din g of the springs of h u m an action m ay re
side n ot in a set of stored generalizations about the hidden elem ents
of m ind and how they conspire to produ ce behavior, but rath er in one
or m ore p rototypes of the deliberative or purposeful p rocess. To
u n d erstan d or explain som eon e's behavior m ay be less a m atter of
deduction from implicit law s, and m ore a m atter of recognitional sub
sum ption of the case at issue un d er a relevant prototyp e. (For a m ore
detailed treatm ent of this view of explanation, the prototype activation
m odel, see ch apter 10.)
Such p rototyp es are no doubt at least m odestly com plex, and p re
sum ably they depict typical configurations of desires, beliefs, prefer
en ces, and so forth, roughly the sam e configurations that I have
earlier attem pted to express in the form of universally quantified
sen ten ces. Beyond this, I am able to say little about them , at least
on this occasion. But I hope I have su cceed ed in m aking intelligible
to you a novel approach to the problem of explanatory u n d erstan d

ing in h u m an s. This is an approach that is grounded at last in w hat
w e know about the brain. A nd it is an ap proach that ascribes to us
n either ream s of universally quantified prem ises, nor deductive ac
tivity on a heroic scale. Explanatory understan din g turns out to be
n ot quite w hat w e thought it w as, b ecause cognition in general gets
ch aracterized in a new w ay. A nd yet explanatory understan din g
rem ains the sam e sort of process in the case of h u m an behavior as
in the case of natural ph enom en a generally. A n d the question of the
adequacy of our com m onsen se understanding rem ains as live as ever.
3 A dden du m : Com m entary on D ennett
I focus here on one of the relatively few issues that still divide
Dennett and m e: the ontological status of intentional states. W e both
accept the prem ise that neuroscience is unlikely to find "se n te n ce s in
the h e a d ," or anything else that answ ers to the structure of individual
beliefs and desires. O n the strength of this shared assum ption, I am
willing to infer that folk p sychology is false, and that its ontology is
chim erical. Beliefs and desires are of a piece w ith phlogiston, caloric,
and the alchem ical essen ces. W e therefore need an entirely new
kinem atics and dynam ics w ith w hich to com p reh en d h u m an cogni
tive activity, one d raw n, perhap s, from com putational neuroscience
and connectionist AI. Folk psychology could then be p u t aside in
favor of this descriptively m ore accurate and explanatorily m ore
pow erful portrayal of the reality w ithin. C ertainly, it will be p ut aside
in the lab and in the clinic, and eventually, p erh ap s, in the m arket
place as well.
But D enn ett declines to draw this eliminativist conclusion, despite
his firm accep tan ce of the prem ise cited, and despite his willingness
to contem plate u n orthod ox form s of cognitive theory. H e prefers to
claim a special status for the various intentional states, a status that
will perm it us to be "realists" about beliefs and desires despite their
projected absence from our strict scientific ontology.
This im pulse in D ennett continues to strike m e as arbitrary p ro tec
tionism , as ill m otivated special pleading on behalf of the old and
familiar. H is initial rationale for exem pting folk psychology from the
usual scientific standards involved assigning it a purely instrum ental
This note is a short com m entary on Daniel C. D en nett's Intentional Stance. It

first appeared in Behavioral and Brain Sciences 11 (1989), no. 3, under the title,
"O n the O ntological Status of Intentional States: Nailing Folk Psychology to
Its P erch ."
status, but this swiftly brought him all kinds of grief, as he him self
explains (1987, pp. 7 1 -7 2 ). Instrum entalism is first and forem ost an
anh'realist position, hardly a w elcom e p ort given D ennett's aim s, a
fact Dennett now appreciates in m ore detail. Accordingly, his current
rationale draw s a m uch m ore narrow ly focused analogy b etw een in
tentional states and geom etrical abstracta such as the centers of grav
ity, axes of rotation, equators, etc., that are postulated to such good
effect in m echanics. A s D ennett sees it, these latter are not real in the
sam e sense that concreta like bricks and trees are real (you can 't trip
over them , for exam ple), but they can reasonably be said to be real
even so. Intentional states are real in this sam e sen se, claims D ennett.
The reality of equators, cen ters, and rotational axes I am h ap p y to
g ran t. They are all places or loci of som e sort that are decisively sp e
cifiable by reference to the shape or behavior of the relevant concrete
object. But the alleged similarity of these item s to beliefs, desires, and
oth er intentional states escapes m e entirely. In w hat respects are they
similar, and w hy should they be grouped togeth er in advance of the
issue here at stake? That is, in advance of any hopes of finding an
exculpatory status for intentional states?
D enn ett is quick to point out that folk psychology has som e n on
trivial predictive pow er, especially in its central dom ain of norm al
h um an behavior, despite the lack of any neural concreta answ ering to
the propositional attitudes. H e em phasizes, quite correctly, that it is
an objective fact about hum ans that a significant am ount of their b e
havior is accurately predictable in intentional term s.
But I think he overvalues this fact wildly. W e m ust not forget that
all sorts of false theories, with w holly chim erical ontologies, can boast
v ery im pressive predictive pow er in various proprietary dom ains. But
this buys their ontology no special status. It is an objective fact that
m u ch of the behavior of m etals and ores is predictable in term s of the
alchem ical essen ces, that m ost of the behavior of the visible heavens
is predictable in term s of nested crystal sp heres, that m uch of the
behavior of m oving bodies is predictable in term s of im petus, and so
forth, A nd yet there are no alchem ical essen ses, nor any crystal
sp heres, nor any im petuses. W e could, of course, set about insisting
that these three "th in g s” are real and genuine after all, though m ere
abstracta to be sure. But none of us is tem p ted to salvage their reality
by such a tortu red and tran sp aren t ploy. W h y should w e be tem pted
in the case of the propositional attitudes?
This disagreem ent betw een us on the status of folk p sychology
dates from several letters now a full decad e old. H ow ever, one point
on w hich w e then agreed w as that neither of us could clearly im agine
a system atic alternative to folk psychology. A t the time I ascribed this
inability to the natural poverty of our im aginations. D enn ett w as in

clined to suspect a deeper reason. But since then the m aterials avail
able to im agination have im proved dram atically. The m icrostructure
of the brain and the recent successes of connectionist A I both sug
gest that our principal form of representation is the high-dimensional
activation vector, and that our principal form of com putation is the
v ector-to-vecto r transform ation, effected by a m atrix of differently
w eighted synapses. In place of propositional attitudes an d logical in
ferences from one to another, therefore, w e can conceive of p ersons
as the seat of vectorial attitudes and various nonlinear transformations
from o ne vector to an other. W e can already see how su ch a vectorial
system can do m any of the things that h u m an s and oth er anim als do
swiftly and easily, such as recognize faces and other highly com plex
stimuli, or control a com plex body w ith both relevance and grace. The
possibility of a real alternative now seem s beyond dispute: w e are
already building it.
W h at rem ains an issue is how our familiar folk p sych ology will fare
in light of w h at the new conception will reveal. Retention through
reduction rem ains a real possibility, though the ch aracter of the
theoretical developm ents just cited m ake this seem increasingly u n
likely. If w e rule out reduction, then elimination em erges as the only
coh eren t alternative, D enn ett's resistance notw ithstanding.
In the end, D enn ett's steadfast insistence that folk p sych ology is
not just an oth er a false theory, but rath er an "ab stract stan ce" of
som e kind, one w ith striking predictive pow ers, rem inds m e of the
shop keeper in the M onty Python sketch about the distraught cu stom
er trying to retu rn a recently purchased b ut very dead p arro t. Python
fans will rem em ber the shopkeeper's deliciously shifty-eyed insis
tence. "N a w , n aw , it's not d ea d ! It's just restin g ! It's ju st pin in g for
the fio rd s !. . . Lovely plum age, the N orw egian B lu e."
Chapter 7
Reductionism, Connectionism, and the Plasticity
of Human Consciousness
It rem ains a m atter of keen debate w heth er h um an natu re in general,

and the h um an sciences in particular, will ever be b rough t un d er the
exp lan atory um brella of the natural sciences. The resou rces of the
latter continue to exp an d, and their accum ulated successes reach ever
closer to the dom ain of the form er. Yet certain salient features of the
h u m an condition continue to elude naturalistic explanation, and they
p resen t them selves, to som e eyes, as forever beyond th at form of
u n d erstanding.
M y concern in this ch apter is to exam ine one type of antinaturalist
argu m en t, a type that starts from the prem ise that h u m an nature is
both highly plastic and culturally configured. The first part of the
challenge posed to naturalism is the claim that w hat constitutes a
h u m an consciousness is not just the intrinsic ch aracter of the crea
ture itself, but also the rich m atrix of relations it bears to the other
h u m an s, practices, and institutions of its em bedding culture. A n atu r
alistic accou n t of hu m an consciousness and behavior, insofar as it is
limited to such things as the m icroscopic activities in an individual's
brain, for exam ple, cannot hope to capture m ore than a small part of
w hat is explanatorily im portant.
The second part of the challenge arises directly from the first. Given
that (possible) cultures are endlessly various, so also is "h u m an
n a tu re ." H um an beings, unlike the objects of the natural sciences,
are plastic in character. M oreover, hu m an s are self-defining or self-
constituting entities. A s h um an cultures evolve, so also does the
ch aracter of individual h u m an consciousness. Simply p u t, there is no
stable or lasting "h u m an n atu re" that a reductive or naturalistic
accou n t m ight hope to explicate or capture. For this reason , the argu
m ent concludes, any such project m ust be m isconceived from the
outset.
This chapter is an abridged version of a paper that first appeared in Cultural

Dynamics 1 (1988), no. 1.
This style of argum ent has a considerable history, but one of its
clearest and m ost accessible con tem porary exponents is Charles
Taylor (1970, 1987). In w hat follows, I hope to reverse the antinatural
ist, antireductionist conclusion of the argum ent. Y et this undertaking
h as a certain poignancy for m e, since I accept both of the prem ises
from w hich the argum ent proceeds. Indeed, in earlier w ritings
(C hurchland 1979, 1981a, 1985b) I have vigorously defended both the
radical plasticity of the h um an m ind, and the dram atic extent to
w hich the character of hu m an consciousness is determ ined by the
cultural surround, specifically, by the ideological, linguistic, and
practical surround. In this ch apter I shall try to show how a naturalist
can em brace the insights of the tw o prem ises cited, w ithout co m
prom ising the possibility of a strongly naturalistic, and even red u c
tionistic, science of hum an consciousness.
M y strategy here is not m erely critical, b ut positive. I propose to
illustrate the w eakness of the argum ents cited above by outlining a
recently developed theoretical and experim ental approach to the p h e
n om en a of h u m an cognition that is at once (a) naturalistic, (b) red u c
tionistic, and (c) capable of explaining both the radical plasticity of
h u m an consciousness, and its intricate depen den ce on the extended
cultural surround.
This new app roach , called 'connectionism ', resides at the interface
of com putational neuroscience, cognitive psychology, and artificial
intelligence (AI). It represents a radical break w ith the approach es
that have dom inated all three of these disciplines for the last three
decad es. It even m akes firm contact with som e of the m ajor them es of
the continental tradition in philosophy, such as the inarticulate or
nonpropositional character of the bulk of h u m an know ledge, and the
prim acy of being an endlessly active agent in a w orld of practical ex
igencies. But it does not support the antinaturalist them es to be found
in recent continental and analytical philosophies, as w e are about to
see. Let m e begin w ith a brief introduction to the central con cep ts of
this novel and very fertile approach. After w e have seen the unusual
p roperties displayed by connectionist system s, w e shall return to the
philosophical issues with w hich w e began.
[I h ere ask the read er to turn to ch apter 9, sections 3 and 4 (pp.
1 5 9 -1 7 1 ), for the m aterial on learning in brainlike netw orks. A fter
reading that m aterial, the p resen t essay m ay be picked up below .]
E nou gh exam ples. You have seen som ething of w hat netw orks of
this kind can do, and of how they do it. In both respects they contrast
sharply w ith the kinds of representational and processing strategies
th at analytical philosophers, cognitive psychologists, and AI w orkers
have traditionally ascribed to us (nam ely, sentencelike rep resen ta
The Plasticity of Human Consciousness 131
tions manipulated by formal rules.) You can see also w hy this theoret
ical and experim ental approach has captured the interest of those
w ho seek to u n derstand how the m icroarchitecture of the biological
brain p rodu ces the p h en om en a of h um an consciousness. Let us now
retu rn to philosophical issues.
1 The P lasticity A rgum ent
W e b egan this discussion w ith the objection that h u m an conscious

ness is plastic, that it has no stable or essential character that a red u c
tionist program m ight hope to capture, and that w hat character it
does h ave derives m ainly from the com plex culture that su rrou n d s it.
Let us exam ine the first p art first: the plasticity objection.
The plasticity of h u m an nature is a problem for naturalistic and
reductionistic theories only if they are unable to explain it. To explain
our cognitive plasticity will require that the reductionist give an
accou n t of the underlying m echanism s that sustain such plasticity,
the dim ensions of variation along w hich change can o ccu r, and the
forces that can drive change from one cognitive configuration to
an oth er. The m ere fact of conceptual plasticity is not a fatal problem
for reductionists, since they are in no w ay com m itted to the idea that
h u m an s adm it of only one cognitive configuration, only one style of
con sciou sn ess. O n the contrary, the radical plasticity of h um an na
ture is a m ajor them e of som e reductionist w riters (P. M . Churchland
1979 an d ch apters 11 and 12 below ).
But can a naturalistic theory have a realistic hope of ev er explaining
our plasticity? Certainly it can, and w e m ay observe the outlines of a
possible explanation in the approach detailed in chap ter 9 . A ccording
to th at ap proach , on e's basic cognitive apparatus consists of a very
large netw ork of interconnected units, w hich adm its of variation in
the w eights of its m yriad connections. The character of o n e 's p ercep
tion, o n e's cognition, and one's behavior is determ ined by the p ar
ticular configuration of w eights w ithin that netw ork. It is the m any
w eights that determ ine w hat features in the w orld o ne respon d s
to, w hich concepts one uses to process them , w hich values one
em b races, and w hich range of behaviors one com m ands.
H ow m u ch plasticity will this buy us? A spectacular am ou nt. A n
alm ost incom prehensible am ount. A typical h um an brain contains
som ething close to 100 billion neurons. A typical n eu ron receives
synaptic connections from roughly 3000 other n eu ron s. A hum an
brain, therefore, will typically contain som ething on the order of
1011 x 103 = 1014 synaptic connections. Let us suppose, conservative-
4 ly, that the w eight of each connection can assum e any 1 of only 10
possible values. Given these figures, how m any distinct cognitive

configurations can a hum an brain assum e? W ell, there are 10 possible
w eights for the first of its 1014 connections, tim es 10 possible w eights
for the second connection, tim es 10 for the third, and so on. The total
n um ber of distinct possible configurations is therefore 1010 , or
K)ioo,ooo,ooo,ooo,ooo 7 0 g e t SOm e idea of the size of this num ber, recall
th at the total num ber of elem entary particles in the entire universe is
generally estim ated to be about 1087.
To be sure, this m easure of hum an plasticity is in one respect an
overestim ation, since m any possible settings of the w eights in a brain
will not produ ce an "in terestin g ly " functional creature. If w e set all
1014 of the connection w eights at zero, for exam ple, the resulting crea
ture w ould be useless, as w ould a creature with every w eight set at
m axim um . Deletion of such cases will certainly reduce our estim ate
by several orders of m agnitude, but w e are still left w ith a spectacular
num ber. O ur m easure is also an underestimation in at least one im
p ortan t respect. It assum es that connection w eights rep resent the
only dim ension of variation. But real brains not only adjust their
existing connection w eights, they can and do grow millions of en
tirely new connections.
I will m ake no further attem pt to fine tune this rough num erical
estim ate of the range of cognitive characters possible for h u m an s.
W h at is im portant here is that the num ber, w hatever it is, is very
large. Reductionists need not deny the plastic character of h um an
con sciou sn ess. O n the contrary, they are in an excellent position to
explain it, conceivably in som e detail.
2 T he Cultural-Em bedding Objection
This final objection begins w ith the claim that the features of the
w orld that are m axim ally im portant for explaining the behavior of
h u m an s are n ot the simple ones that can be defined in the vocabulary
of a naturalistic physics, but rath er the m u ch m ore subtle and co m
plex features that constitute our social culture. O ne does indeed dis
crim inate and respon d to light, to w arm th , to sound, and so forth.
But one also discrim inates and respond s to the phonem es peculiar to
o n e's language, to the m eanings of its w ord s, to the m oral of a story,
to the significance of gestures and facial expressions, to challenges
and obligations and social opportunities, and to all of the intricacies
that m ake up a functioning culture.
All of this is no doubt true. But again, these things are w holly con
sistent w ith a reductionist p rogram for understanding the natu re of
h u m an cognition. W hat the reductionist m u st do is explain how a
physical system can com e to address and m anipulate su ch subtle and

culturally configured features. W hile this is certainly a challenge, it no
longer appears to be a problem in principle, for w e have already seen
how suitably trained netw orks can com e to rep resent and discrimi
nate features of great subtlety and abstraction.
W h at is im p ortant is that the netw ork have a "te a c h e r" of som e
ap p rop riate sort, a teach er w ho can help to shape the ch aracter of its
internal representations and the p attern of its environm ental re
sp onses. W e have seen how relatively simple netw orks can be trained
to recognize com plex and culturally-im posed auditory features like
m ine ech oes and ph on em es. W e have seen (from ch apter 9) how they
can learn the highly irregular and highly context-sensitive tran s
form ation rules that are essential to the task of reading English text.
A nd w e have seen how they can be trained to recognize the endlessly
variable three-dim ensional structures of external objects from two-
dim ensional gray-scale pictures of them . In each case, the netw ork is
forced tow ard acknow ledgm ent of som e feature or dom ain of features
selected from the infinity of possible features to w hich it m ight re
sp ond , and it is not an insuperable problem that the features at issue
are com plex, subtle, con text sensitive, or stim ulus tran scen d en t. W ith
suitable teaching, the netw ork generates an internal rep resentation of
them regardless. This does not m ean that the features ad d ressed are
m agical, or superphysical, or beyond the realm of natural science. It
just m ean s that the sim plest possible definition or rep resentation of
them m ay well be the entire configuration of the successfully trained
netw ork! Given a netw ork of any nontrivial size, those rep resen ta
tions can indeed be extraordinarily com plex, but it is no part of the
reductionist's p rogram to deny this.
It therefore should not be a surprise that a hum an infant com es to
recognize and respon d to cultural features that resist definition in
term s of notions like m ass, charge, length, and so forth, b ecause the
m ost dom inant "te a ch e r" in the local environm ent is the culture into
w hich the infant is born. The set of w eights that constitutes a child's
developing consciousness is continually being shaped by the linguis
tic, concep tu al, and social surround. The developing brain com es to
reflect the elem ents and structure of that surround in great detail. For
that is w h at netw orks do. W h at shapes them is the stimuli they typi
cally receive, and the subsequent corrections in their respon ses to
w hich th ey are typically subject. Small w on d er w e b ecom e attu n ed to
the categories of the culture that raises us.
This m ay sound like a m ajor concession to the view th at the m ost
im p ortan t factors in the explanation of h um an behavior are to be
found at the cultural level rather than at the m icrophysical level. But
to say this w ould be to m iss the point, and to acquiesce in a false

dichotom y betw een "cu ltu ral" and "p h y sica l" features. The real dis
tinction is b etw een features that are simple and context free on the
one h an d , and features that are extraordinarily com plex and have
m an y layers of context d ep en den cy on the other. The latter are in
d eed of m ajor im portance for explaining h u m an behavior. But neither
they, n or our capacity for recognizing them , is forever beyond a
naturalistic understanding. For w e can already und erstan d, right
here and now , how a system of m any simple physical elem ents can
com e to rep resent them , and to respon d to them , and to undergo
d ram atic conceptual developm ent in the p rocess. That is the p rosp ect
that a com putational neuroscience holds out to us, and I think it
w ould be a tragedy if antireductionist convictions should frustrate or
slow our pursuit of this p rosp ect. Such conservative convictions, to
judge from the preceding discussion, m ay be based on true and im
p ortan t prem ises about h um an nature. But they underestim ate the
capacity of em erging science to provide a highly revealing and n atu r
alistic explanation of those very sam e prem ises. A n d such explanation,
after all, is w h at scientific reduction am ou nts to.
3 Conclusion
The plasticity of h um an consciousness is real, and precious, and

theoretically im portant. Y et it is not beyond naturalistic explanation.
The diffuse and culturally-constituted ch aracter of m any of the fea
tures to w hich h u m an s respond is also real and im portant. But those
features are still a p art of the n atural w orld, and it is not impossible to
u n d erstan d either how they are constituted, or how a purely physical
system is able to discrim inate them . This conclusion does leave intact
the idea that m any of the im portant d eterm inants of h um an behavior
are to be found at the very high level of com plexity that constitutes a
culture. But that is all right. P erh ap s a reductionist is bound to say
that w e are com posed of simple things. But he is not bound to say that
either w e, or our environm ent, are simple things.
I close on a cautionary note. The story of the preceding pages has
show n how a naturalistic approach to h u m an consciousness is not
autom atically and fundam entally opposed to the explanatory cate
gories of our existing culture. O n the con trary, it m ay vindicate them
by providing a naturalistic explanation of them . But it rem ains p os
sible that this d eep er understanding of our cognitive m icrostructure
will frequently reveal m istakes and m isconceptions in our com m on
sense categories. Indeed, given the great com plexity of h u m an s and
hum an culture, it is only to be expected that our com m onsense ideas
con cern in g the dim ensions of h um an consciousness an d the springs

of h u m an action will turn out to be pinched and defective in various
w ays. A n d w e m ay hope that fields such as connectionist AI and
com putational neurobiology will be a source of new an d better ideas
on these m atters. I em phasize this point because, although I w ish to
concede the dynam ical relevance of high-level culturally-em bedded
features, it rem ains an open scientific question w hich features at that
level are the truly real and genuinely im portant ones. P erh ap s folk
p sych ology has a firm if partial grasp of them . A n d p erh ap s they
rem ain largely to be discovered.
PA RT II
The Structure of Science

Chapter 8
The Ontological Status of Observables:
In Praise of the Superempirical Virtues
A t several points in the reading of van F raassen 's book (1980), I feared
I w ould no longer be a realist by the time I com pleted it. Fortunately,
sheer doxastic inertia has allowed m y convictions to survive its search
ing critique, at least tem porarily, and as w e address you today, van
F raassen and I still hold different view s. I am a scientific realist, of
u n orth od o x persuasion, and van Fraassen is a constructive em piri
cist, w h ose persuasions currently define the doctrine. I assert that
global excellence of theory is the ultim ate m easure of tru th and ontol
ogy at all levels of cognition, even at the observational level. Van
Fraassen asserts that descriptive excellence at the observational level
is the only genuine m easure of any th eory's truth, and that one's
accep tan ce of a theory should create no ontological com m itm ents
w h atev er beyond the observational level.
A gain st his first claim I shall m aintain that observational excellence
or "em p irical ad eq u acy " is only one epistem ic virtue am on g others,
of equal or com parable im portance. A nd against his secon d claim I
shall m aintain that the ontological com m itm ents of any theory are
w holly blind to the idiosyncratic distinction betw een w h at is and
w hat is n ot hum anly observable, and so should be our o w n ontologi
cal com m itm ents. Criticism will be directed primarily at van Fraassen's
selective skepticism in favor of observable ontologies o ver unobserv
able ontologies; and against his view that the superem pirical th eoret
ical virtu es (simplicity, coherence, explanatory p ow er) are m erely
pragm atic virtues, irrelevant to the estim ate of a th eo ry 's truth. My
aim s are n ot m erely critical, h ow ever. Scientific realism does need
rew orking, and there are good reasons for m oving it in the direction
of v an F raassen 's constructive em piricism , as will be d iscussed in the
This paper first appeared in the Pacific Philosophical Quarterly 63 (1982), no. 3.
As the prose suggests, it was first presented at a sym posium on scientific
realism (Halifax, 1981) that focused on my 1979 book and Bas van Fraassen's
1980 book.
140 The Structure of Science
closing section of this paper. But those reason s do not support the
skeptical theses at issue.
1 H ow van Fraassen's Problem Collapses into H um e's Problem
Before pursuing our differences, it will p rove useful to em phasize

certain convictions that w e share. Van Fraassen is already a scientific
realist in the minim al sense that he interprets theories literally and he
con ced es them a truth value. Furth er, w e agree that the observable/
unobservable distinction is entirely distinct from the nontheoretical/
theoretical distinction, and w e agree as well that all observation sen
tences are irredeem ably laden w ith theory.
A dditionally, I absolutely reject m any sanguine assum ptions com
m on am ong realists. I do not believe that on the w hole our beliefs
m u st be at least roughly true; I do not believe that the term s of
"m a tu re " sciences m ust typically refer to real things; and I very
m u ch doubt that the R eason of H om o sapien s, even at its best and
even if allowed infinite tim e, w ould eventually encom p ass all an d /or
only true statem ents.
This skepticism is born partly from a historical induction: so m any
p ast theories, rightly judged excellent at the tim e, have since proved
to be false. And their current successors, though even better founded,
seem but the next step in a probably endless and not obviously
con vergen t jou rn ey. (For a m ost thorough and insightful critique of
typical realist theses, see the recent p aper by Lau d an [1981].)
Evolutionary considerations also counsel a healthy skepticism . H u
m an reason is a hierarchy of heuristics for seeking, recognizing, stor
ing, and exploiting inform ation. But those heuristics w ere invented at
ran d om , and they w ere selected for w ithin a very narrow evolution
ary environ m en t, cosm ologically speaking. It w ould be m iraculous
if h um an reason w ere com pletely free of false strategies and fun
d am ental cognitive limitations, and doubly m iraculous if the theories
w e accept failed to reflect those defects.
Thus som e very realistic reasons for skepticism with respect to any
theory. W h y, then , am I still a scientific realist? Because these reason s
fail to discrim inate betw een the integrity of observables and the in
tegrity of unobservables. If anything is com prom ised by these consid
eration s, it is the integrity of theories generally. That is, of cognition
generally. Since our observational con cep ts are just as theory laden as
any others, and since the integrity of those con cep ts is just as contin
g en t on the integrity of the theories that em bed them , our observa
tional ontology is rendered exactly as dubious as our nonobservational
ontology.
The Ontological Status of Observables 141
This parity should not seem surprising. O ur history contains real

exam ples of m istaken ontological com m itm ents in both d om ains. For
exam ple, w e have had occasion to banish phlogiston, caloric, and the
lum iniferous ether from our ontology— b ut w e have also had occa
sion to banish w itches, and the starry sphere that turns about us
daily. These latter item s w ere as "o b serv ab le" as you please, and
they w ere w idely "o b serv ed " on a daily basis. W e are too often
m isled, I think, by our casual use of 'observes' as a su ccess verb:
w e tend to forget that, at any stage of our history, the ontology p re
su p p osed by our observational judgm ents rem ains essentially specu
lative and wholly revisable, h ow ever entrench ed and familiar it m ay
have b ecom e.
A ccord in gly, since the skeptical considerations ad d u ced above are
indifferent to the distinction b etw een w h at is and is n o t observable,
they provide no reason for resisting a com m itm ent to unobservable
ontologies w hile allow ing a com m itm ent to w hat w e take to be observ
able ontologies. The latter appear as no better off than the form er. For
m e, th en , the "em pirical su ccess" of a theory rem ains a reason for
thinking the theory to be true, and for accepting its overall ontology.
The inference from success to truth should no doubt be severely tem
pered by the skeptical considerations adduced , but the inference to
unobservable ontologies is n ot rendered selectively dubious. Thus, I re
m ain a scientific realist. M y realism is highly circum spect, but the
circum spection is uniform for unobservables and observables alike.
Perh ap s I am w ron g in this. P erhaps w e should be selectively skep
tical in the fashion van Fraassen recom m en d s. D oes he have other
argu m ents for refusing factual belief and ontological com m itm ent
b eyond the observational dom ain? Indeed he does. In fact, he does
not appeal to historical induction or evolutionary hum ility at all.
These are m y reasons for skepticism (and they will rem ain , even if
w e m anage to underm ine van Fraassen 's). They have been intro
d uced here to show that, while there are som e pow erful reason s for
skepticism , those reasons do not place unobservables at a selective
disadvantage.
V ery well, w hat are van F raassen's reasons for skepticism ? They
are very interesting. To sum m arize quickly, he does a com pelling job
of deflating certain stan d ard realist argu m ents (from Sm art, Sellars,
Salm on, Boyd, and others) to the effect that, given the aim s of scien
ce, w e have no alternative but to bring unobservables (not just into
our calculations, but) into our literal ontology. H e also argu es, rather
com pellingly, that the superem pirical virtues, such as simplicity and
com prehen sive explanatory pow er, are at bottom m erely pragm atic
4virtues, having nothing essential to do w ith any theory's truth. This
leaves only empirical adequacy as a genuine m easure of any theory's

truth. R oughly, a theory is empirically adequate if and only if ev ery
thing it says about observable things is true. Empirical adequacy is thus
a n ecessary condition on a th eo ry 's truth.
H ow ever, claims van Fraassen, the truth of any theory w hose
ontology includes unobservables is alw ays radically u n d erd eter
m ined by its empirical adequacy, since a great m any logically incom
patible theories can all be empirically equivalent. A ccordingly, the
inference from empirical adequacy to truth now appears p resu m p
tuou s in the extrem e, especially since it h as just been disconnected
from additional selective criteria such as simplicity and explanatory
p ow er, criteria w hich m ight have reduced the arbitrariness of the p ar
ticular inference draw n. Fortunately, says van Fraassen, w e do not
need to m ake such w anton inferences, since w e can perfectly well
u nd erstand science as an enterprise that n ever really draw s them .
H ere w e arrive at his positive conception of science as an enterprise
w hose sole intellectual aims are empirical adequacy and the satisfac
tion of certain h um an intellectual needs.
The central elem ent in this argu m en t is the claim that, in the case of
a theory w hose ontology includes unobservables, its empirical ade
quacy underdeterm ines its truth. (W e should notice that in the case of
a theory w hose ontology is com pletely free of unobservables, its
em pirical adequacy does not u n d erdeterm ine its truth: in that case,
truth and empirical adequacy are obviously identical. Thus van F raas
sen 's selective skepticism with respect to unobservables.) That is, for
any theory T inflated with unobservables, there will alw ays be m any
oth er such theories incom patible with T, b u t empirically equivalent
to it.
In m y view , the notions of "em pirical ad eq u acy " and its cognate
relative term "em pirically equivalent" are extrem ely thorny notions
of doubtful integrity. If w e attem pt to explicate a th eory's "em pirical
co n ten t" in term s of the observation sen ten ces it entails (or entails if
conjoined w ith available background inform ation, or w ith possible
future background inform ation, or with possible future theories), w e
generate a variety of notions that are variously em p ty, con text rela
tive, ill defined, or flatly incom patible w ith the claim of u n d erd eter
m ination. V an Fraassen expresses aw aren ess of these difficulties and
p roposes to avoid them by giving the notions at issue a m odel-
theoretic rath er than a syntactic explication. I am unconvinced that
this im proves m atters decisively (on this issue, see W ilson 1980; also
M usgrave 1985, H ooker 1985, G lym our 1985, and W ilson 1985). In
particular, I think van Fraassen has not dealt at all adequately with
the problem of how the so-called "em pirical equivalence" of tw o in
com patible theories rem ains relative to w hich background theories are
added to the evaluative context, especially background theories that
in som e w ay revise our conception of w hat hum ans can observe. I
intend to sidestep this issue for now , h ow ever, since the m atter is
com plex and there is a m uch simpler objection to be voiced.
Let m e approach m y objection by first pointing out th at the em pir
ical adequacy of any theory is itself som ething that is radically u n d er
determ ined by any evidence conceivably available to us. Recall that,
for a theory to be empirically adequate, w hat it says about observable
things m u st be true— all observable things, in the p ast, in the indefi
nite fu ture, and in the m ost distant corners of the cosm os. But since
any actual data possessed by us m ust be finite in its scope, it is plain
that w e here suffer an underdeterm ination problem no less serious
than th at claimed above. This is H u m e's problem , and the lesson is
that even observation-level theories m u st suffer radical u n d erd eter
m ination by the evidence. A ccordingly, theories about observables
and theories about unobservables appear on a p ar again, so far as
skepticism is con cerned .
V an Fraassen thinks there is an im portant difference betw een the
tw o cases, and on e's first im pulse is to agree w ith him . W e are all
willing to concede the existence of H u m e's problem , the problem of
justifying the inference to unobserved entities. But the inference to
entities that are dow nright unobservable app ears as a different and
addition al problem .
The ap p earance is an illusion, as the following considerations will
show . C onsider som e of the different reasons w hy entities or p ro
cesses m ay go unobserved by us. First, they m ay go u n observed be
cause, relative to our natural sensory app aratus, they fail to enjoy an
appropriate spatial or tem poral position. They m ay exist in the U pper
Jurassic Period, for exam ple, or they m ay reside in the A n d rom ed a
G alaxy. Second, they m ay go unobserved becau se, relative to our
natural sensory app aratu s, they fail to enjoy the appropriate spatial or
tem poral dim ensions. They m ay be too sm all, or too brief, or too large,
or too p rotracted . Third, they m ay fail to enjoy the appropriate energy,
being too feeble, or too pow erful, to perm it useful discrim ination.
Fou rth and fifth, they m ay fail to have an appropriate w avelength, or
an ap p rop riate m ass. Sixth, they m ay fail to “ feel" the relevan t fun
dam ental fo rces our sensory apparatus exploits, as w ith our inability
to observe the background neutrino flux, despite the fact th at its en er
gy density exceeds that of light itself.
This list could be lengthened, but it is long enough to su ggest that
being spatially or tem porally distant from our sen sory ap p aratu s is
4 only one am ong m any w ays in w hich an entity or p rocess can fall
outside the com pass of h um an ob serv ation , a w ay distinguished by

no relevant epistem ological or ontological features.
There is clearly som e practical point in o u r calling a thing "o b serv -
able" if it fails on ly the first test (sp atiotem p oral proxim ity), and “ un
o b s e r v a b le " if it fails any of the oth ers. B ut that is only because of the
contingent practical fact that h u m ans g enerally have som ew hat m ore
control over the spatiotem poral p ersp ectiv e of their sensory system s
than they have over their size, or reaction time, or m ass, or
w avelen gth sensitivity, or chem ical con stitu tion . H ad w e b een less
mobile than w e are— rooted to the earth like D ouglas firs, say— yet
b een m ore voluntarily plastic in our se n so ry constitution, the distinc
tion betw een the "m erely u n o b serv ed " and the "d o w n righ t u n
observable" w ould have been very differently draw n. It m ay help to
im agine h ere a suitably rooted arboreal philosopher nam ed (w hat
else?) D ouglas van Fiirrsen, w ho, in his sed en tary w isdom , u rges an
antirealist skepticism concerning the spatially very distant entities
p ostulated by his fellow trees.
A dm ittedly, for any distant entity o n e can in principle alw ays
change the relative spatial position of o n e 's sensory apparatus so that
the entity is observed: one can go to it. B ut equally, for any m icro
scopic entity one can in principle alw ays ch an ge the relative spatial
size or configuration of one's sen sory ap p aratu s so that the entity is
o bserved. Physical law im poses certain lim itations on such plasticity,
but so also does physical law limit h ow far one can travel in a lifetime.
To em phasize the im portance of th ese considerations, let m e
u nd erscore the structure of m y objection h ere. C onsider the distinc
tion betw een
(1) things observed by som e h u m an (w ith unaided senses),

(2) things thus observable by h u m a n s, but not in fact observed,
(3) things not observable by h u m an s at all.
Van Fraassen 's position w ould exclude (3) from our rational ontology.
This h as at least som e initial plausibility. But his position w ould not
be at all plausible if it w ere com m itted to excluding both (3) and (2)
from our rational ontology. N o p arty to the p resen t discussion is will
ing to restrict rational ontology to (1) alon e. Van F raassen's position
thu s requires a principled distinction b etw een (2) and (3), a distinction
adequate to the radical difference in epistem ic attitude he w ould have
us ad op t tow ard them . The burden of m y argu m en t is that the dis
tinction betw een (2) and (3), once it is u n earth ed , is only very feebly
principled, and is w holly inadequate to b ear the great w eight that van
Fraassen puts on it.
The point of all this is that there is no special or novel problem

about inferences to the existence of entities com m only called " u n
o b serv ab les." Such entities are m erely those that go unobserved by
us for reason s other than their spatial or tem poral distance from us.
But w h eth er the " g a p " to be bridged is spatiotem poral, or one of the
m any o th er gaps, the logical/epistem ological problem is the sam e in
all cases: am pliative inference and u nderdeterm ined h yp oth eses. I
therefore fail to see h ow van Fraassen can justify tolerating an am plia
tive inference w hen it bridges a gap of spatial distance, while refusing
to tolerate an ampliative inference w hen it bridges a gap of, for exam
ple, spatial size. H um e's problem and van F raassen 's problem
collapse into one.
Van Fraassen attem p ts to m eet such w orries about the inescapable
ubiquity of speculative activity by observing that "It is n o t an episte-
m ological principle that one m ay as well hang for a sheep as for a
lam b" (1980, p. 72). A greed . But it is a principle of logic th at one m ay
as well h an g for a sheep as for a sheep, and van Fraassen 's lamb
(em pirical adequacy) is just an oth er sheep.
Sim ply to hold few e r beliefs from a given set is, of cou rse, to be less
ad ven tu rou s, but it is not necessarily to be applauded. O ne m ight
decide to relinquish all on e's beliefs save those about objects
w eighing less than 500 kilogram s, and perhap s one w ould then be
logically safer. But in the absence of som e relevant epistem ic differ
ence b etw een on e's beliefs about such objects and on e's beliefs about
other objects, that is perversity, not parsim ony.
L t m e sum m arize. A s van Fraassen sets it up, and as the in
strum entalists set it up before him , the realist looks m ore gullible than
the nonrealist, since the realist is willing to extend belief beyond the
observable, while the nonrealist insists on confining belief w ithin that
dom ain. I su ggest, h ow ever, that it is really the nonrealists w ho are
being the m ore gullible in this m atter, since they sup p ose that the
epistem ic situation of our beliefs about observables is in som e w ay
superior to that of our beliefs about unobservables. But in fact, their
epistem ic situation is not superior. They are exactly as dubious as
their nonobservational cousins. Their causal history is different (they
are occasioned by activity in the sensory p ath w ays), but the ontology
they p resu p p o se enjoys no privilege or special credibility.
2 The Prim acy o f the Superem pirical Virtues
Let m e n ow try to ad d ress the question of w heth er the theoretical

virtues such as simplicity, coh erence, and explanatory p ow er are epis-
' tem ic virtu es genuinely relevant to the estim ate of a th eory's truth, as
tradition says, or m erely pragm atic virtues, as van Fraassen u rges. His
view p rom otes empirical adequacy, or evidence of empirical adequa
cy, as the only genuine m easure of a th eory's truth, the other virtues
(insofar as they are distinct from these) being cast as purely pragm atic
virtues, to be valued only for the h u m an needs they satisfy. D espite
certain com pelling features of the account of explanation that van
Fraassen provides, I rem ain inclined tow ard the traditional view .
M y reason is simplicity itself. Since there is no w ay of conceiving or
representing "th e empirical facts" that is com pletely independent of
speculative assum ptions, and since w e shall occasionally confront
theoretical alternatives on a scale so com prehensive that w e m u st also
choose betw een com peting m od es of conceiving w hat the empirical
facts before us are, it is clear that the epistem ic choice betw een
these global alternatives cannot be m ade by com paring the extent to
w hich they are adequate to som e com m on tou ch ston e, "th e em
pirical fa c ts ." In such a case, the choice m u st be m ade on the com
parative global virtues of the tw o global alternatives, 7 ,-p lu s-
the-observational-evidence-therein-construed, versu s T2-plus-the-
observational-evidence-therein-(differently)-construed. That is, it
m u st be m ad e on superem pirical grou nd s such as relative coh eren ce,
simplicity, and explanatory unity.
V an Fraassen has said that to "s a v e the ap p earan ces" is to exhibit
them as a fragm ent of a larger unity. W ith this I w holly agree. But I
am here pointing out that it is a decision b etw een com peting larger
unities that determ ines w hat w e count as "th e true ap p earan ces" in
the first place. There is no independent w ay to settle that question.
A n d if such global decisions can only be m ad e on w hat van Fraassen
calls "p ra g m a tic" grounds, then it w ould seem to follow that any d e
cision concerning w h at the observable w orld contains m ust be essen
tially "p ra g m a tic" also. Inflationary m etaphysics and "p ra g m a tic"
decisions begin, it seem s, as soon as w e open our eyes.
Global issues such as these are rem iniscent of C arn ap 's "e x te rn a l"
questions, and I think it likely that van Fraassen , like C arnap, does
n ot regard them as decidable in any but a second-rate sense, since
th ey can only be decided by second-rate (i.e ., by pragm atic) con sid
erations. If so, how ever, it is difficult to see how van Fraassen can
justify a selectively realist attitude tow ard "o b se rv a b le s," since, as w e
h ave seen , pragm atic considerations m ust attend their selection also.
(These issues receive extended treatm ent in C hurchland 1979, sec
tions 2, 3, 7, and 10.) W hat all of this illustrates, I think, is the poverty
of van F raassen 's crucial distinction betw een factors that are " e m
pirical, and therefore tru th -relev an t," and factors that are "su p e r-
em pirical, and therefore not tru th -relev an t."
A s I see it then, values such as ontological simplicity, coh eren ce,

and explanatory p ow er are som e of the brain's m ost basic criteria for
recognizing inform ation, for distinguishing inform ation from noise.
A nd I think they are even m ore fundam ental values than is "em p iri
cal adequacy," since collectively they can overthrow an entire concep
tual fram ew ork for representing the empirical facts. Indeed, they
even dictate how such a fram ew ork is constructed by the questing
infant in the first place. O ne's observational taxon om y is n o t read off
the w orld directly; rather, one com es to it piecem eal and by stages,
and one settles on that taxonom y w hich finds the greatest coherence
and simplicity in the w orld, and the m ost and the sim plest lawful
connections.
I can bring togeth er m y protective concerns for unobservables and
for the superem pirical virtues by w ay of the following th ou gh t experi
m ent. C on sid er a m an for w hom absolutely nothing is observable. All
of his sen sory m odalities have been surgically d estroyed, and he has
no visual, tactile, or oth er sensory experience of any kind. Fortu n ate
ly, he has m ounted on top of his skull a m icrocom p u ter fitted out
w ith a variety of environm entally-sensitive transd u cers. The com p u
ter is con n ected to his association cortex (or p erhaps the frontal lobe,
or W ernicke's area) in such a w ay as to cause in him a continuous
string of singular beliefs about his local environm ent. These "intellec
tual in tu itions" are not infallible, but let us suppose that th ey provide
him w ith m u ch the sam e inform ation that our perceptual jud gm en ts
provide us.
For su ch p erson , or for a society of such p ersons, the observable
w orld is an em pty set. There is no question, therefore, of their
evaluating any theory by reference to its "em pirical a d e q u a cy ," as
ch aracterized by van Fraassen (i.e., isom orphism betw een som e
observable features of the w orld and som e "em pirical su b stru ctu re"
of one of the th eory's m odels). But such a society is still capable of
science, I assert. They can invent theories, con stru ct explanations
of the facts-as-represented-in-past-spontaneous-beliefs, hazard p re
dictions of the facts-as-represented-in-future-spontaneous-beliefs,
and so forth. In principle, there is no reason they could n ot learn as
m uch as w e have (com pare Feyerabend 1969).
But it is plain in this case that the global virtues of simplicity, coh er
ence, and explanatory unification are w hat m ust guide the continuing
evolution of their collected beliefs. A nd it is plain as well that their
ontology, w hatever it is, m ust consist entirely of unobservable en
tities. To invite a van Fraassenean disbelief in unobservable entities
is, in this case, to invite the suspension of all beliefs beyond tautolo
gies! Surely reason does n ot require them to be so abstem ious.
It is time to consider the objection that those aspects of the w orld

th at are successfully m onitored by the transducing m icrocom puter
should count as "ob servab les" for the folk described, despite the lack
of any appropriate field of internal sensory qualia to m ediate the ex
ternal circum stance and the internal jud gm en t it causes. Their tables-
and-chairs ontology, as expressed in their spontaneous jud gm en ts,
could then be conceded legitim acy.
I will be the first to accept such an objection. But if w e do accept it,
then I do not see how w e can justify van F raassen 's selective skepti
cism w ith respect to the w ealth of "u n o b serv ab le" entities and p rop
erties reliably m onitored by our transducing m easuring instrum ents
(electron m icroscopes, cloud cham bers, ch rom atograp h s, e tc.). The
sp ontan eou s singular judgm ents of the w orking scientist, at hom e in
his theoretical vocabulary and deeply familiar w ith the m easuring in
stru m ents to w hich his conceptual system is responding, are not
w orse off, causally or epistem ologically, than the spontaneous singu
lar jud gm ents of our transducer-laden friends. If skepticism is to be
p u t aside above, it m ust be p ut aside here as well.
M y concluding thought experim ent is a com plem ent to the one just
outlined. C onsider som e folk w ho observe, not less of the w orld than
w e do, but m ore of it. Suppose them able to observe a dom ain n or
mally closed to u s: the m icrow orld of virus particles, D NA strands,
and large protein m olecules. Specifically, sup p ose a race of hum anoid
creatures each of w hom is born w ith an electron m icroscope p er
m anently in place over his left eye. The scope is biologically co n
stituted, let us suppose, and it projects its im age onto a hum an-like
retina, w ith the rest of their neurophysiology paralleling our ow n.
Science tells us, and I take it that van Fraassen w ould agree, that
virus particles, D N A strands, and m ost oth er objects of com parable
dim ensions count as observable entities for the hum anoids described.
The h u m an oids, at least, w ould be justified in so regarding them and
in including them in their ontology.
But w e hum ans m ay not include such entities in our ontology,
according to van Fraassen's position, since they are not observable
w ith our unaided perceptual ap p aratus. W e m ay n ot include such
entities in our ontology even though w e can con stru ct, and even if w e
do con stru ct, electron m icroscopes of identical function, place them
over our left eyes, and enjoy exactly the sam e m icroexperience as the
hum anoids!
The difficulty for van F raassen's position, if I u n derstand it correct
ly, is that his position requires that a hum anoid and a scope-equipped
h u m an m ust em brace different epistem ic attitudes tow ard the m icro
w orld, even thou gh their causal connections to the w orld and their
continuing experience of it are identical: the hum anoid is required to

be a realist w ith respect to the m icrow orld, and the h u m an is required
to be an antirealist (i.e., an agnostic) w ith respect to the m icrow orld.
But this distinction betw een w hat w e and they m ay p roperly em
b race as real seem s to m e to be highly arbitrary and radically u n d er
m otivated . For the only difference betw een the hum anoid and a
scope-equipped hu m an lies in the causal origins of the transducing
instru m en ts feeding inform ation into their respective brains. The
h u m an oid 's scope ow es its existence to inform ation cod ed in his
genetic m aterial. The h u m an 's scope ow es its existence to inform ation
coded in his cortical m aterial, or in technical libraries. I do not see
w hy this should m ake any difference in their respective ontological
com m itm en ts, w h atever they are, and I m u st decline to em brace any
philosophy of science w hich says that it m ust.
3 Tow ard a M ore R ealistic Realism
I now tu rn from critic of van Fraassen's position to ad vocate. O ne of

the m o st central elem ents in his view seem s to m e to be well m oti
vated and urgently deserving of further developm ent. A s he explains
in his introd uctory chapter, his aim is to reconceive the relation of
theory to w orld, and the units of scientific cognition, and the virtue of
those units w hen successful. H e says, "I use the adjective 'co n stru c
tive' to indicate m y view that scientific activity is one of construction
rath er than d iscovery: construction of m odels that m ust be adequate
to the p h en om en a, and not discovery of truth concerning the u n
observable" (1980, p. 5).
The traditional view of h um an know ledge is that the unit of cogni
tion is the sen tence or proposition, and the cognitive virtue of such
units is tru th . V an Fraassen rejects this overtly linguistic guise for his
em piricism . H e invites us to reconceive a theory as a set of m odels
(rather than as a set of sentences), and he sees em pirical adequacy
(rather than truth) as the principal virtue of such units.
T hough I reject his particular reconception, and the selective skep
ticism h e draw s from it, I think the m ove aw ay from the traditional
conception is entirely correct. The criticism to w hich I am inclined is
that v an Fraassen has n ot m oved quite far enou gh . Specifically, if w e
are to recon sid er truth as the aim or p roduct of cognitive activity, I
think w e m u st reconsider its applicability right across the board, and
n ot just in som e arbitrarily or idiosyncratically segregated dom ain of
"u n o b serv ab les." That is, if w e are to m ove aw ay from the m ore
naive form ulations of scientific realism , w e should m ove in the direc
tion of pragm atism rather than in the direction of a positivistic in

strum entalism . Let m e elaborate.
W h en w e consider the great variety of cognitively active creatures
on this planet— sea slugs and octop u ses, bats, dolphins, and h u
m an s; and w hen w e consider the ceaseless reconfiguration in w hich
their brains or central ganglia en gage— adjustm ents in the respon se
potentials of single neurons m ad e in the m icrosecond ran ge, changes
in the respon se characteristics of large system s of neurons m ad e in
the secon ds-to-h ou rs range, dendritic grow th and new synaptic con
nections and the selective atrophy of old connections effected in the
day-plus ran ge; then van F raassen 's term 'construction' begins to
seem highly appropriate. There is endless construction and recon
struction, both functional and structural. Furth er, it is far from ob
vious that truth is either the prim ary aim or the principal p rodu ct of
this activity. R ather, its function w ould app ear to be the ever m ore
finely tuned adm inistration of the o rgan ism 's behavior. N atural selec
tion does not care w heth er a brain has or tends tow ard true beliefs, so
long as the organism reliably exhibits reproductively advantageous
behavior. Plainly, there is going to be som e connection betw een the
faithfulness of the brain's w orld m odel and the propriety of the
o rgan ism 's behavior. But just as plainly, the connection is not going
to be direct.
W hile w e are considering cognitive activity in biological term s and
in all branch es of the phylogenetic tree, w e should note that it is far
from obvious that sentences or propositions or anything rem otely like
them constitute the basic elem ents of cognition in creatures generally.
Indeed, as I have argued at length elsew here (1979, in chapters 1 and
5 above), it is highly unlikely that the sentential kinem atics em braced
by folk psychology and o rth od ox epistem ology rep resents or captures
the basic p aram eters of cognition and learning even in hu m an s. That
fram ew ork is part of a com m on sense theory that threatens to be
either superficial or false. If w e are ever to u n d erstan d the dynam ics of
cognitive activity, therefore, w e m ay have to reconceive our basic unit
of cognition as som ething other than the sen ten ce or proposition, and
reconceive its virtue as som ething other than truth.
Success of this sort on the descriptive-explanatory front w ould like
ly have norm ative consequences. Truth, as currently conceived,
m ight cease to be an aim of science. N ot b ecause w e had low ered our
sights and red u ced our epistem ic stan d ard s, as van F raassen 's co n
structive em piricism w ould suggest, but b ecause w e had raised our
sights, in pursuit of som e epistem ic goal even m ore w orth y than truth.
I can n ot now elucidate such goals, but w e should be sensible of their
possible existence. The notion of truth, after all, is but the central
elem ent in a clutch of descriptive and norm ative theories (folk psych ol
ogy, folk epistem ology, folk sem antics, classical logic), and w e can
exp ect conceptual p rogress here as appropriately as anyw h ere else.
The notion of truth is susp ect on purely m etaphysical grou n d s any
w ay. It suggests straightaw ay the notion of The C om plete and Final
True Th eory: at a m inim um , the infinite set of all true sen ten ces. Such
a theory w ould be, by epistem ic criteria, the best theory possible. But
nothing w h atever guarantees the existence of such a unique theory.
Ju st as there is no largest positive integer, it m ay be that there is no
best theory. It m ay be that for any theory w hatsoever, there is always
an even b etter theory, and so ad infinitum. If w e w ere thu s unable to
speak of the set of all true sentences, w hat sense could w e m ake of
truth sen ten ce-b y-sen ten ce?
These considerations do invite a "co n stru ctiv e " conception of
cognitive activity, one in w hich the notion of truth plays at best a
highly derivative role. The form ulation of such a conception, ade
quate to all of our epistem ic criteria, is the outstanding task of epis
tem ology. I do not think w e shall find that conception in van
F raassen 's m odel-theoretic version of "positivistic in stru m en talism ,"
nor do I think w e shall find it quickly. But the empirical brain begs
unraveling, and w e have plenty of time.
Finally, there is a question pu t to m e by Stephen Stich. If ultim ately
m y view is even m ore skeptical than van F raassen 's concerning the
relevance or applicability of the notion of truth, w h y call it scientific
realism at all? For at least tw o reasons. The term 'realism ' still m arks
the principal contrast w ith its traditional adversary, positivistic in
strum entalism . W h atever the integrity of the notion of tru th , theories
about unobservables have just as much a claim to truth, epistem ologi-
cally an d m etaphysically, as theories about observables. S econd, I re
m ain com m itted to the idea that there exists a w orld, ind ep en d en t of
our cognition, with w hich w e interact, and of w hich w e con stru ct
rep resentation s: for varying p urp oses, w ith varying penetration , and
w ith varyin g success. Lastly, our best and m ost penetrating grasp of
the real is still held to reside in the rep resentations provided by our
b est theories. Global excellence of theory rem ains the fundam ental
m easure of rational ontology. A nd that h as always b een the central
claim of scientific realism .
Chapter 9
On the Nature of Theories:
1 The C lassical View o f Theories
N ot long ago, w e all knew w hat a theory w as: it w as a set of sentences

or propositions, expressible in the first-order predicate calculus. A nd
w e had w h at seem ed to be excellent reason s for that view . Surely any
theory had to be statable. A nd after it h ad been fully stated , as a set of
sen ten ces, w h at residue rem ained? Furth erm ore, the sentential view
m ade system atic sense of how theories could perform the prim ary
business of theories, nam ely, prediction, explanation, and inter-
theoretic reduction. It w as basically a m atter of first-order deduction
from the sen tences of the theory conjoined w ith relevant prem ises
about the dom ain at hand.
Equally im portant, the sentential view prom ised an accou n t of the
nature of learning, and of rationality. Required w as a set of formal
rules to dictate appropriate changes or u p dates in the overall set of
believed sen ten ces as a function of new beliefs supplied by observa
tion. O f cou rse, there w as substantial disagreem ent about w hich
rules w ere appropriate. Inductivists, falsificationists, hyp oth etico-
deductivists, and Bayesian subjectivists each proposed a different
accou n t of them . But the general approach seem ed clearly correct.
Rationality w ould be captured as the proper set of form al rules
em erged from logical investigation.
Finally, if theories are just sentences, then the ultim ate virtue of a
theory is tru th . A n d it w as w idely expected that an adequate accou n t
of rational m ethodology w ould reveal w h y h u m an s m u st tend , in the
long ru n , tow ard theories that are true.
H ardly anyone will now deny that there are serious problem s w ith
every elem en t of the preceding picture, difficulties I shall discuss be
low . Y et the m ajority of the profession is n ot y et willing to regard
This essay w as first published in C. W . Savage, ed ., The N ature o f Theories,

M innesota Studies in the Philosophy of Science, vol. 14 (M inneapolis: Uni-
Nversity of M innesota Press, 1989).
them as fatal. I profess myself am ong the m inority that does so regard
them . In urging the poverty of "senten tial epistem ologies" for over a
decad e now (1975b, 1979, chapters 1 and 5), I h ave been m otivated
prim arily by the pattern of the failures displayed by that approach.
Those failures suggest to m e that w hat is defective in the classical
approach is its fundam ental assum ption that languagelike structures
of som e kind constitute the basic or m ost im portant form of rep re
sentation in cognitive creatures, and the correlative assum ption that
cognition consists in the m anipulation of those representations by
m ean s of structure-sensitive rules.
To be sure, not everyone saw the sam e p attern of failure, nor w ere
they prepared to draw such a strong conclusion even if they did. For
any research program has difficulties, and so long as w e lack a com
parably com pelling alternative conception of rep resentation and com
putation, it m ay be best to stick w ith the familiar research p rogram of
sen tences and rules for their m anipulation.
H ow ever, it is no longer true that w e lack a com parably com pelling
alternative approach. W ithin the last five years, there have been som e
striking theoretical developm ents and experim ental results w ithin
cognitive neurobiology and connectionist AI. These have provided us
w ith a pow erful and fertile fram ew ork w ith w hich to address p rob
lem s of cognition, a fram ew ork that ow es nothing to the sentential
paradigm of the classical view . M y m ain p u rp ose in this essay is to
m ake the rudim ents of that fram ew ork available to a w ider audience,
and to explore its far-reaching consequences for traditional issues in
the philosophy of science. Before turning to this task, let m e prep are
the stage by briefly sum m arizing the principal failures of the classical
view and the m ost prom inent responses to them .
2 P roblem s and A lternative A pproaches
The depiction of learning as the rule-governed updating of a system

of sen ten ces or propositional attitudes encou n tered a wide range of
failures. For starters, even the best of the rules p roposed failed to
rep rod uce reliably our preanalytic jud gm en ts of credibility, even in
the artificially restricted or toy situations in w hich they w ere asked to
function. P aradoxes of confirm ation plagued the hypoth etico-d ed u c-
tive accou nts (H em pel 1965; Scheffler 1963). The indeterm inacy of fal
sification plagued the Popperian accoun ts (Lakatos 1970; Feyerabend
1970; C hurchland 1975b). Law s w ere assigned negligible credibility
on C arnapian accounts (Salmon 1966). Bayesian accounts, like C arna-
pian ones, p resupposed a given probability space as the epistem ic
playground w ithin w hich learning takes place, and they could not
On the Nature of Theories 155
accou nt for the rationality of m ajor shifts from one probability space
to an other, w hich is w hat the m ost interesting and im p ortan t cases of
learning am ou nt to. The rationality of large-scale conceptual change,
accordingly, seem ed beyond the reach of such ap p roach es. Fu rth er
m ore, simplicity em erged as a m ajor determ inant of theoretical
credibility on m ost accounts, but none of them could provide an
adequate definition of simplicity in syntactic term s, or give a co n
vincing explanation of w hy it w as relevant to truth or credibility in
any case. O ne could begin to question w heth er the basic factors
relevant to learning w ere to be found at the linguistic level at all.
Beyond these an noyan ces, the initial resources ascribed to a learn
ing subject by the sentential approach plainly p resu p p osed the su c
cessful com pletion of a good deal of sophisticated learning on the part
of that subject already. For exam ple, reliable observation judgm ents
do n ot just app ear out of now here. Living subjects h ave to learn to
m ake the com plex perceptual discrim inations that m ake perceptual
jud gm en ts possible. A nd they also have to learn the linguistic or prop-
ositional system within w hich their beliefs are to be constituted.
Plainly, both cases of learning will have to involve som e p rocedure
quite distinct from that of the classical account. For that accou n t p re
su p p oses an teced ent possession of both a determ inate propositional
system and a capacity for determ inate perceptual jud gm en t, w hich is
precisely w h at, prior to extensive learning, the h u m an infant lacks.
A ccordingly, the classical story cannot possibly accou n t for all cases
of learning. There m u st exist a type of learning that is prior to, and
m ore basic than, the p rocess of sentence m anipulation at issue.
Thus are w e led rath er swiftly to the idea that there is a level of
rep resentation beneath the level of the sentential or propositional atti
tud es, and to the correlative idea that there is a learning dynam ic that
operates prim arily on sub linguistic factors. This idea is reinforced by
reflection on the problem of cognition and learning in n on h u m an
anim als, none of w hich appear to have the benefit of langu age, either
the external speech or the internal stru ctures, but all of w hich en
gage in sophisticated cognition. P erhaps their cognition p roceeds
entirely w ithout benefit of any system for processing sentencelike
rep resentation s.
E ven in the h u m an case, the depiction of on e's know ledge as an
im m ense set of individually stored "se n te n c e s" raises a severe prob
lem concern in g the relevant retrieval or application of those internal
rep resentation s. H ow is it one is able to retrieve, from the millions of
sen ten ces stored, exactly the handful that is relevant to o n e 's current
predictive or explanatory problem , and how is it one is generally able
4 to do this in a few tenths of a second? This is know n as the "fram e
p roblem " in AI, and it arises b ecause, from the point of view of fast
and relevant retrieval, a long list of sen ten ces is an appallingly ineffi
cient w ay to store inform ation. A nd the m ore inform ation a creature
h as, the w orse its application problem becom es.
A further problem w ith the classical view of learning is that it finds
no essential connection w hatever b etw een the learning of fa cts and
the learning of sk ills. This is a problem in itself, since one m ight have
h op ed for a unified accou nt of learning, b ut it is doubly a problem
w hen one realizes that so m u ch of the business of u n d erstanding a
theory and being a scientist is a m atter of the skills one has acquired.
M em orizing a set of sentences is n ot rem otely sufficient: one m ust
learn to recognize the often quite various instances of the term s they
contain; one m ust learn to m anipulate the peculiar form alism in w hich
th ey m ay be em bedded; one m u st learn to apply the form alism to
novel situations; one m ust learn to control the instrum ents that typi
cally produ ce or m onitor the p h en om en a at issue. A s T. S. K uhn first
m ad e clear (1962), these dim ensions of the scientific trade are only
artificially separable from on e's understan din g of its current theories.
It begins to ap p ear that even if w e do harbor internal sen ten ces, they
capture only a small part of h um an know ledge.
These failures of the classical view over the full range of learning,
both in hu m an s and in nonh u m an anim als, are the m ore suspicious
given the total disconnection of the classical view from any theory
concerning the structure of the biological brain, and the m an n er in
w hich it m ight im plem ent the kind of representations and com p u ta
tions p roposed . M aking acceptable con tact w ith neurophysiological
theory is a long-term constraint on any epistem ology: a schem e of
representation and com putation that can n ot be im plem ented in the
m achinery of the h um an brain cannot be an adequate accou n t of
h um an cognitive activities.
The situation on this score used to be m u ch better than it is n ow : it
w as clear that the classical accou n t of rep resentation and learning
could easily be realized in typical digital com p u ters, and it w as
thou gh t that the h u m an brain w ould turn out to be relevantly like a
digital com p u ter. But quite aside from the fact that com p u ter im ple
m entations of sentential learning chronically produced disappointing
results, it has becom e increasingly clear that the brain is organized
along com putational lines radically different from those em ployed in
conventional digital com puters. The brain, as w e shall see below , is a
m assively parallel p rocessor, and it perform s com putational tasks of
the classical kind at issue only v ery slowly and com paratively badly.
Loosely speaking, it does not ap p ear to be designed to perform the
tasks the classical view assigns to it.
I conclude this survey by returning to specifically philosophical

m atters. A final problem w ith the classical ap proach h as b een the
failure of all attem pts to explain w hy the learning p rocess m u st tend,
at least in the long run, to lead us tow ard true theories. Surprisingly,
and p erh ap s distressingly, this Panglossean hope has p roved very
resistant to vindication (van Fraassen 1980; L audan 1981). A lthough
the history of h u m an intellectual en deavor does support the view that
over the centuries our theories have becom e dram atically better in
m any dim ensions, it is quite problem atic w heth er they are su cces
sively "c lo se r" to " tru th ." Indeed, the notion of tru th itself has
recently com e in for critical scrutiny (Pu tn am 1981; C hurchland,
ch apter 8 above; Stich 1989). It is no longer clear that there is any
unique and unitary relation that virtuous belief system s m ust bear
to the nonlinguistic w orld. W hich leaves us free to recon sid er the
great m an y different dim ensions of epistem ic and pragm atic virtue
that a cognitive system can display.
The problem s of the preceding pages have not usually been p re
sen ted in concert, and they are not usually regarded as conveying a
unitary lesson. A few philosophers, h ow ever, have been m oved by
them , or by som e subset of them , to su ggest significant m odifications
in the classical fram ew ork. O ne ap p roach that has captured som e
ad h eren ts is the "sem an tic v iew " of theories (Suppe 1974; van F raas
sen 1980; Giere 1988). This ap proach attem pts to drive a w edge be
tw een a theory and its possibly quite various linguistic form ulations
by characterizing a theory as a set o f m odels, those that will m ake a
first-order linguistic statem ent of the theory com e out true u n d er the
relevant assignm ents. The m odels in the set all share a co m m o n ab
stract structu re, and that structure is w hat is im portant about any
theory, according to the sem antic view , n ot any of its idiosyncratic
linguistic expressions. A theory is tru e, on this view , just in case it
includes the actual w orld, or som e part of it, as one of the m odels in
the set.
This view buys us som e ad van tages, p erh ap s, but I find it to be a
relatively narrow respon se to the panoply of problem s ad d ressed
a b o v e. In particular, I think it strange that w e should be asked, at this
stage of the debate, to em brace an accou n t of theories th at h as abso
lutely nothing to do w ith the question of how real physical system s
m ight em body representations of the w orld, and how they m ight
execu te principled com putations on those rep resentation s in such a
fashion as to learn. Prim a facie, at least, the sem antic ap p roach takes
theories even farther into Plato's H eaven, and aw ay from the buzzing
brains that use them , than did the view that a theory is a set of sen t
en ces. This com plaint does not do justice to the positive virtues of the
sem antic ap proach (see especially Giere 1988, w hose version does
m ake som e contact with cu rrent cognitive psychology). But it is clear
that the sem antic approach is a response to only a small subset of the
extant difficulties.
A m ore celebrated response is em bodied in K uhn's Structure o f Sci
entific R evolutions (1962). Kuhn cen ters our attention not on sets of
sen ten ces, n or on sets of m odels, but on w h at he calls paradigm s or
exem p lars, w hich are specific applications of our conceptual, m ath e
m atical, and instrum ental resou rces. M astering a theory, on this
view , is m ore a m atter of being able to perform in various w ays, of
being able to solve a certain class of problem s, of being able to recog
nize diverse situations as relevantly similar to that of the original or
paradigm atic application. K uhn's view brings to the fore the histori
cal, sociological, and psychological factors that structure our theoreti
cal cognition. Of central importance is the m anner in which one comes
to perceive the w orld as one internalizes a theory. The perceptual
w orld is redivided into new categories, and while the theory m ay
be able to provide n ecessary and sufficient conditions for being an
instance of any of its categories, the p erceptual recognition of any
instance of a category does not generally p roceed by reference to
those conditions, w hich often transcend perceptual experience.
R ather, perceptual recognition proceeds by som e inarticulable p ro
cess that registers sim ilarity to one or m ore perceptual prototypes of the
category at issue. The recognition of new applications of the ap p ar
atus of the entire theory displays a similar dynam ic. In all, a su ccess
ful theory provides a prototypical beachh ead that one attem pts to ex
p and by analogical extensions to new dom ains.
R eaction to this view has been deeply divided. Som e applaud
K u h n 's m ove tow ard naturalism , tow ard a perform ance conception
of know ledge, and aw ay from the notion of truth as the guiding co m
p ass of cognitive activity (M unevar 1981; Stich 1989). O thers deplore
his neglect of norm ative issues, his instrum entalism and relativism ,
and his alleged exaggeration of certain lessons from perceptual and
developm ental p sychology (Fod or 1984). I shall address these issues
later in this chapter.
A third and less visible reaction to the classical difficulties has
sim ply rejected the sentential or propositional attitudes as the m ost
im portan t form of representation used by cognitive creatu res, and
h as insisted on the necessity of empirical and theoretical research
into brain function in order to answ er the question of w hat are the
m ost im portant form s of representation and com putation w ithin
cognitive creatu res. Early statem ents can be found in P. M. C hurch-
land 1975b and H ooker 1975; extend ed argu m en ts app ear in P. M.
C h urchland 1979 and ch apter 1; and further argu m en ts app ear in

P. S. C h urchland 1980a and 1986; and in H ooker 1987.
W hile the antisentential diagnosis could be given som e consider
able su p p ort, as the opening sum m ary of this section illustrates,
neuroscience as the recom m end ed cure w as alw ays m ore difficult to
sell, given the functional opacity of the biological brain. Recently,
how ev er, this has changed dram atically. W e now have som e p ro
visional insight into the functional significance of the brain's m icro
stru ctu re, and som e idea of how it represents and com p u tes. W h at has
been discovered so far appears to vindicate the claims of philosophi
cal relevance and the expectations of fertility in this area, and it
app ears to vindicate som e central elem ents in K uhn's p erspective as
well. This neurofunctional fram ew ork prom ises to sustain wholly
new directions of cognitive research. In the sections below I shall try
to outline the elem ents of this fram ew ork and its applications to som e
familiar problem s in the philosophy of science. I begin w ith the
physical structure and the basic activities of the brainlike system s at
issue.
3 E lem entary Brainlike N etw orks
The functional atom s of the brain are cells called neu ron s (figure 9.1 ).
These have a natural or default level of activity, w hich can , h ow ever,
be m odulated up or dow n by external influences. From each neuron
Figure 9.1
x A schematic neuron
Figure 9.2
A neuronlike processing unit
there extends a long, thin output fiber called an axon, w hich typically
b ran ches at the far end so as to m ake a large num ber of synaptic con
nections w ith either the central cell body or the bushy dendrites of o ther
n eu ron s. Each n euron thus receives inputs from a great m an y other
n eu ron s, w hich inputs tend to excite (or to inhibit, according to the
type of synaptic connection) its norm al or default level of activation.
The level of activation induced is a function of the num ber of con n ec
tions, of their size or w eight, of their polarity (stim ulatory or inhibi
tory), and of the strength of the incom ing signals. F urth erm ore, each
n eu ron is constantly em itting an output signal along its ow n axon, a
signal w hose stren gth is a direct function of the overall level of activa
tion in the originating cell body. That signal is a train of pulses or
spikes, as they are called, w hich are p ropagated swiftly along the
axon. A typical cell can emit spikes along its axon at anything b e
tw een 0 and p erh ap s 200 h ertz. N euron s, if you like, are hum m ing to
one an other, in basso notes of varying frequency.
The netw orks to be explored attem p t to sim ulate natural neurons
w ith artifical units of the kind depicted in figure 9 .2 . These units
adm it of various levels of activation, w hich I shall assum e to vary
b etw een 0 and 1. Each unit receives input signals from other units via
"s y n a p tic " connections of various w eights and polarities. These are
rep resented in the diagram as small end plates of various sizes. For
simplicity I dispense w ith dendritic trees: the "a x o n a l" end b ranches
from oth er units all m ake connections directly to the "cell b o d y " of
the receiving unit. The total m odulating effect E im pacting on that
unit is just the sum of the contributions m ade by each of the co n n ec
tions. The contribution of a single connection is just the p rodu ct of its
w eight Wj tim es the strength s, of the signal arriving at that connec-
Figure 9.3
The sigmoid axonal output function
tion. Let m e em phasize that if for som e reason th e connection

w eigh ts w ere to change over time, then the unit w ould receive a quite
different level of overall excitation or inhibition in resp o n se to the
very sam e configuration of input signals.
Turn now to the outp ut side of things. A s a function of the total
input E, the unit m odulates its activity level and em its an output sig
nal of a certain stren gth s0 along its "a x o n a l" outp u t fiber. But s0 is not
a direct or linear function of E. R ather, it is an S -sh ap ed function as in
figure 9 .3 . The reasons for this small wrinkle will em erge later. I
m ention it here b ecause its inclusion com pletes the story of the
elem en tary units. Of their intrinsic properties, there is n othing left to
tell. They are very simple indeed.
It rem ains to arrange them into netw orks. In the brain, neurons
frequently consitute a population all of w hich send their axons to the
site of a second population of neurons, w here each arriving axon
divides into term inal end branches in order to m ake synaptic con
n ections w ith m an y different cells w ithin the target population.
A xon s from cells in this second population can then project to a
third population of cells, and so on. This is the inspiration for the
arran gem en t of figure 9.4.
The units in the b ottom or input layer of the n etw ork m ay be
th ou gh t of as "s e n so ry " units, since the level of activation in each is
directly determ ined by aspects of the environm ent (or p erh ap s by the
exp erim enter, in the p rocess of sim ulating som e environm ental in
put). The activation level of a given input unit is designed to be a
respon se to a specific aspect or dim ension of the overall input stim u
lus th at strikes the b ottom layer. The assem bled set of sim ultaneous
activation levels in all of the input units is the netw ork 's representation
N of the input stim ulus. W e m ay refer to that configuration of stim ula
tion levels as the input vector, since it is just an ordered set of num bers
Figure 9.4
A simple network
or m agnitudes. For exam ple, a given stim ulus m ight produce the
v ector < .5, .3, .9, .2>.
These input activation levels are then p ropagated u p w ard , via the
ou tp u t signal in each u nit's axon, to the m iddle layer of the netw ork,
to w h at are called the "h id d en u n its." A s can be seen in figure 9 .4 ,
any unit in the input layer m akes a synaptic connection of som e
w eight or other w ith every unit at this interm ediate layer. Each hid
den unit is thus the target of several inputs, one for each cell at the
input layer. The resulting activation level of a given hidden unit is
essentially just the sum of all of the influences reaching it from the
cells in the low er layer.
The result of this upw ard propagation of the input vector is a set of
activation levels across the three units in the hidden layer, called the
"h id d en unit activation v e cto r." The values of that three-elem ent
vector are strictly determ ined by
(a) the m akeup of the input vector at the input layer,

(ib) the various values of the connection w eights at the ends of the
term inal branches of the input units.
W h at this b ottom half of the netw ork does, evidently, is convert or

transform one activation vector into another.
The top half of the netw ork does exactly the sam e thing, in exactly
the sam e w ay. The activation vector at the hidden layer is p ropagated
u p w ard to the outp u t (topm ost) layer of units, w here an output vector
is p ro d u ced , w hose character is determ ined by
(a) the m akeup of the activation vector at the hidden layer,

(b) the various values of the connection w eights at the ends of
the term inal branches of the hidden units.
Looking n ow at the w hole netw ork, w e can see that it is just a device
for transform ing any given input-level activation vector into a unique
ly corresp on d in g output-level activation vector. A n d w h at d eter
m ines the character of the global transform ation effected is the
peculiar set of values possessed by the m an y connection w eights.
This m u ch is easy to grasp. W hat is n ot so easy to grasp , prior to
exploring exam ples, is just how very pow erful and useful those tran s
form ations can be. So let us explore som e real exam ples.
4 R epresentation and Learning in Brainlike N etw orks
A great m an y of the environm ental features to w hich h u m an s re

spond are difficult to define or characterize in term s of their purely
physical properties. Even som ething as m u nd an e as the vow el sound
/а /, as in 'rain ', resists such characterization, for the ran ge of acou s
tical variation am ong acceptable and recognizable /а /s is en orm ou s. A
fem ale child at tw o y ears and a basso m ale at fifty will p rodu ce quite
different sorts of atm ospheric excitations in pronou n cin g this vow el,
but each sound will be easily recognized as an /а / by oth er m em bers
of the sam e linguistic culture.
I do n ot m ean to suggest that the m atter is utterly intractable from a
physical point of view , for an exam ination of the acoustical pow er
sp ectru m of voiced vow els begins to reveal som e of the similarities
that unite /a /s. A nd yet the analysis continues to resist a sim ple list of
necessary and sufficient physical conditions on being an /а /. Instead,
being an /а / seem s to be a m atter of being close enough to a typical /а /
sound along a sufficient num ber of distinct dim ensions o f relevance,
w h ere each notion in italics rem ains difficult to characterize in a
nonarbitrary w ay. M oreover, som e of those dim ensions are highly
contextual. A sound type that w ould not norm ally be counted or rec
ognized as an /а / w hen voiced in isolation m ay be unproblem atically
so cou n ted if it regularly occurs, in som eon e's m odestly accented
4 sp eech , in all of the phonetic places that w ould norm ally be occupied
by /а /s. Evidently, w hat m akes som ething an /а / is in part a m atter of

the entire linguistic surround. In this w ay do w e very quickly ascend
to the abstract and holistic level for even the sim plest of culturally
em bedded properties.
W hat holds for ph on em es holds also for a great m any oth er im
p ortan t features recognizable by us: colors, faces, flow ers, trees,
anim als, voices, sm ells, feelings, songs, w ord s, m eanings, and even
m etaphorical m eanings. A t the o utset, the categories and resou rces of
ph ysics, and even n euroscience, look p u n y and im potent in the face
of such subtlety.
A n d yet it is a purely physical system that recognizes such intrica
cies. Short of appealing to m agic, or of sim ply refusing to confront the
problem at all, w e m ust assum e that som e configuration of purely
physical elem ents is capable of grasping and m anipulating these
features, and by m eans of purely physical principles. Surprisingly,
netw orks of the kind described in the p reced in g section have m any
of the properties needed to address precisely this problem . Let m e
explain.
S uppose w e are subm arine engineers confronted with the problem
of designing a sonar system that will distinguish betw een the sonar
echoes retu rn ed from explosive m ines, such as m ight lie on the b ot
tom of sensitive w aterw ays during w artim e, and the sonar echoes
retu rn ed from rocks of com parable sizes that dot the sam e u n d er
w ater landscapes. The difficulty is tw ofold: ech oes from both objects
sound indistinguishable to the casual ear, and echoes from each type
show w ide variation in sonic ch aracter, since both rocks and m ines
com e in various sizes, shap es, and orientations relative to the probing
son ar pulse.
E n ter the netw ork of figure 9 .5 . This one h as thirteen units at the
input layer, since w e need to code a fairly com plex stim ulus. A given
son ar echo is run through a frequency analyzer and is sam pled for its
relative en ergy levels at thirteen frequencies. These thirteen values,
expressed as fractions of 1, are then entered as activation levels in the
respective units of the input layer, as indicated in figure 9 .5 . From
h ere they are p ropagated throu gh the netw ork, being transform ed as
th ey go, as explained earlier. The result is a pair of activation levels in
the tw o units at the output layer. W e need only tw o units here, for w e
w an t the netw ork eventually to produce an output activation vector
at or near < 1, 0> w hen a mine echo is en tered as input, and an output
activation vector at or near <0, 1> w hen a rock echo is entered as
input. In a w ord , w e w ant it to distinguish m ines from rocks.
It w ould, of course, be a m iracle if the netw ork m ade the desired
discrim ination im m ediately, since the connection w eights that d eter
Figure 9.5
Perceptual recognition with a large network
m ine its transform ational activity are initially set at ran d om values. A t
the beginning of this experim ent, then, the outp u t v ectors are sure to
disappoint u s. But w e proceed to teach the netw ork by m ean s of the
following procedure.
W e p rocu re a large set of recorded sam ples of various (genuine)
mine ech o es, from m ines of various sizes and orientations, and a
com parable set of genuine rock ech oes, keeping careful track of w hich
is w hich. W e then feed these echoes into the netw ork, one b y one,
and observe the output vector p rodu ced in each case. W h at interests
us in each case is the am ou nt by w hich the actual outp u t v ecto r differs
from w h at w ould have been the correct vector, given the identity of
the specific echo that produced it. The details of that erro r, for each
elem ent of the output vector, are then fed into a special rule that
co m p u tes a set of small changes in the values of the various synaptic
w eights in the system . The idea is to identify those w eights m ost

responsible for the error, and then to nudge their values in a direction
that w ould at least reduce the am ou nt by w hich the output vector is
in error. The slighty m odified system is then fed an oth er echo from
the training set, and the entire p rocedure is rep eated.
This provides the netw ork w ith a "te a c h e r." The process is called
"train in g up the n etw ork " and it is standardly execu ted by an auxili
ary com p u ter p rogram m ed to feed sam ples from the training set into
the netw ork, m onitor its respon ses, and adjust the w eights according
to the special rule after each trial. U n d er the pressure of such re
p eated corrections, the behavior of the netw ork slowly con verges on
the behavior w e desire. That is to say, after several thou san d s of
p resentations of recorded echoes and subsequent adjustm ents, the
n etw ork starts to give the right answ er close to ninety p ercen t of
the tim e. W h en fed a mine echo, it generally gives som ething close
to a <1, 0> outp u t. A nd w hen fed a rock echo, it generally gives
som ething close to a <0, 1>.
A useful w ay to think of this is captured in figure 9 .6 . Think of an
Figure 9.6
Learning: gradient descent in weight/error space. Axes are shown for only 2 of 105
synaptic weights
abstract space of m any dim ensions, one for each w eight in the n et
w ork (105 in this case), plus one dim ension for rep resentin g the o ver
all erro r of the outpu t vector on any given trial. A n y point in that
space rep resen ts a unique configuration of w eights, plus the perform
ance error that that configuration produ ces. W h at the learning rule
does is steadily nudge that configuration aw ay from erron eou s posi
tions and tow ard positions that are less erroneou s. The system inches
its w ay d ow n an "e rro r g radien t" tow ard a global error m inim um .
O nce there, it respon d s reliably to the relevant kinds of echoes. It
even respo n d s well to echoes that are similar to m ine ech oes, by
giving outp ut vectors that are closer to <1, 0> than to <0, 1>.
T here w as no guarantee the netw ork w ould succeed in learning to
discrim inate the tw o kinds of echoes, because there w as no guarantee
that rock ech oes and mine echoes w ould differ in any system atic or
detectable w ay. But it turns out that m ine echoes do indeed have
som e com p lex of relational or structural features that distinguishes
them from rock ech oes, and under the pressu re of rep eated error
corrections, the netw ork m an ages to lock onto, or b ecom e "tu n e d "
to, th at subtle but distinctive w eave of features.
W e can test w heth er it has truly succeeded in this by n ow feeding
the n etw ork som e mine and rock echoes n ot included in the training
set, ech o es it has never encountered before. In fact, the n etw ork does
alm ost as well classifying the new echoes as it does w ith the sam ples
in its training set. The "k n o w led ge" it h as acquired generalizes quite
successfully to new cases. (This exam ple is a highly simplified
accou nt of som e striking results from G orm an and Sejnowski 1988a,
1988b.)
All of this is m odestly am azing, because the problem is quite a diffi
cult o ne, at least as difficult as learning to discrim inate the p honem e
/а /. H u m an sonar operators, during a long tour of subm arine duty,
eventually learn to distinguish the tw o kinds of ech oes w ith som e
u ncertain but nontrivial regularity. But they never perform at the
level of the artificial netw ork. Spurred on by this su ccess, w ork is
currently u n d erw ay to train up a netw ork to distinguish the various
p h on em es characteristic of English speech (Zipser and Elm an 1987).
The idea is to produ ce a speech-recognition system th at will not be
troubled by the acoustic idiosyncracies of diverse speakers, as
existing speech-recognition system s are.
The su ccess of the m ine/rock netw ork is further intriguing because
the "k n o w led g e" the netw ork has acquired, concerning the distinc
tive ch aracter of m ine echoes, consists of nothing m ore than a careful
ly orch estrated set of connection w eights. A n d it is finally intriguing
4 b ecau se there exists a learning algorithm — the rule for adjusting the
w eights as a function of the error displayed in the output v ector—

that will eventually produce the required set of w eights, given suffi
cient exam ples on w hich to train the netw ork (Rum elhart et al.
1986b).
H ow can a set of connection w eights possibly em body know ledge
of the desired distinction? Think of it in the following w ay. Each of
the thirteen input units rep resents one aspect or dim ension of the
incom ing stim ulus. Collectively, they give a sim ultaneous profile of
the input echo along thirteen distinct dim ensions. N ow p erh ap s there
is only one profile that is roughly characteristic of mine ech oes; or
p erh ap s there are m any different profiles, united by a com m on re
lational feature (e .g ., that the activation value of unit num ber 6 is
alw ays three tim es the value of unit num ber 12); or perhaps there is a
disjunctive set of such relational features; and so forth. In each case it
is possible to rig the w eights so that the system will respond in a
typical fashion, at the output layer, to all and only the relevant
profiles.
The units at the hidden layer are very im p ortan t in this. If w e co n
sider the abstract space w hose seven axes rep resent the possible
activation levels of each of the seven hidden units, then w hat the
system is searching for during the training period is a set of w eights
that partitions this space so that any m ine input produ ces an activation
vector across the hidden units that falls som ew here w ithin one large
subvolum e of this abstract space, while any rock input produ ces a
vector that falls som ew here into the com plem ent of that subvolum e
(figure 9 .7 ). The job of the top half of the netw ork is then the relative
ly easy one of distinguishing these tw o subvolum es into w hich the
abstract space has been divided.
V ectors n ear the center of (or along a certain path in) the m ine-
vector subvolum e represent prototypical m ine ech oes, and these will
p rodu ce an outp u t vector very close to the desired <1, 0>. V ectors
n earer to the surface (strictly speaking, the hypersurface) that parti
tions the abstract space rep resent atypical or problem atic m ine
ech oes, and these produce m ore am biguous output vectors, such as
< .6 , ,4> . The netw ork's discrim inative respon ses are thus graded
respon ses: the system is sensitive to sim ilarities along all of the rel
evant dim ensions, and especially to rou gh conjunctions of these
subordinate similarities.
So w e have a system that learns to discrim inate hard-to-define p er
ceptual features, and to be sensitive to similarities of a com parably
diffuse but highly relevant ch aracter. A n d once the netw ork is trained
u p, the recognition task takes only a split second, since the system
processes the input stim ulus in parallel. It finally gives us a dis-
Figure 9.7
Learned partition on hidden-unit activation-vector space. Axes are shown for only
three of seven hidden-unit activation levels
crim inatory system that perform s som ething like a living creatu re,
both in its speed and in its overall character.
I have explained this system in som e detail, so that the read er will
have a clear idea of how things w ork in at least one case. But the
netw ork described is only one instance of a general technique that
w orks well in a large variety of cases. N etw orks can be co n stru cted
w ith a larger num ber of units at the outp u t layer, so as to be able to
exp ress n ot just tw o, but a large num ber of distinct discrim inations.
O ne netw ork, aptly called NETtalk by its authors (Rosenberg and
Sejnowski 1987), takes vector codings for seven-letter segm en ts of
printed w ord s as inputs and gives vector codings for p h on em es as
ou tp u ts. These outp u t vectors can be fed directly into a sound syn
thesizer as they occur, to produce audible sounds. W hat this network
learns to do is to transform printed w ord s into audible speech.
T hough it involves no understanding of the w ords that it " r e a d s ," the
n etw ork's feat is still very im pressive, b ecause it w as given no rules
w h atev er concerning the phonetic significance of stan d ard English
-sp ellin g. It began its training period by producing a stream of unintel
ligible babble in response to text entered as input. But in the course

of m an y thou san d s of w ord presen tation s, and und er the stead y
pressure of the w eight-nudging algorithm , the set of w eights slowly
m ean d ers its w ay to a configuration that red u ces the m easured
error close to zero. After such training, it will then produce as output,
given arbitrary English text as input, perfectly intelligible speech
w ith only rare and m inor errors.
This case is significant for a num ber of reason s. First, the trained
netw ork m akes a large num ber of discrim inations (79, in fact), not
just a binary one. Second, it contains no explicit representation of any
rules, h ow ever m uch it m ight seem to be following a set of rules.
Third, it has m astered an in p u t/o u tp u t transform ation that is n otor
iously irregular, and it m ust be sensitive to lexical context in order to
do so. (Specifically, the p honem e it assigns to the cen ter or focal letter
of its seven-letter input is in large part a function of the identity of the
three letters on either side.) A nd fourth, it p ortray s som e aspects of a
sensorim otor skill, rather than a purely sen sory skill: it is producing
highly com plex behavior.
N ETtalk has som e limitations, of course. Pronunciations that de
pen d on specifically sem antical or gram m atical distinctions will
generally elude its grasp — unless they h ap p en to be reflected in som e
w ay in the corpus of its training w ord s, as occasionally they are—
since N ETtalk know s neither m eanings n or syntax. But such d ep en
d encies affect only a very small p ercen tage of the transform ations
appropriate to any text, and they are in any case to be exp ected . To
o vercom e them com pletely w ould require a netw ork that actually
u n d erstan d s the text being read. A nd even then m istakes w ould
occur, for even hum ans occasionally m isread w ords as a result of
gram m atical or sem antic confusion. W hat is arresting about NETtalk
is just how very m u ch of the com plex and irregular business of text-
based pronunciation can be m astered by a simple netw ork w ith only a
few hun dred neuronlike units.
A n o th er rath er large netw ork (by Lehky and Sejnowski 1988a,
1988b) addresses problem s in vision. It takes codings for sm oothly
varying gray-scale pictures as input; and after training, it yields as
o utp uts surprisingly accurate codings for the cu rvatures and orienta
tions of the physical objects portrayed in the pictures. It solves a form
of the shape-from -shading problem long familiar to theorists in the
field of vision. This netw ork is of special interest because a subse
quent exam ination of the receptive fields of the trained hidden units
show s them to have acquired som e of the sam e response properties
as are displayed by cells in the visual cortex of m ature anim als. Speci
fically, they show a m axim al sensitivity to sp ots, ed ges, and bars in
specific orientations. This finding echoes the sem inal w ork of Hubei
and W iesel (1962), in w hich cells in the visual cortex w ere discovered
to h ave receptive fields of this sam e ch aracter. Results of this kind are
very im p ortan t, for if w e are to take these artificial netw orks as m od
els for h ow the brain w orks, then they m u st display realistic behavior
n ot ju st at the m acro level; they m ust also display realistic behavior at
the m icro level.
E n ou g h exam ples. You have seen som ething of w hat netw orks of
this kind can do and of how they do it. In both respects th ey contrast
sharply w ith the kinds of representational and processing strategies
that philosophers of science, inductive logicians, cognitive p sych o
logists, and AI w orkers have traditionally ascribed to us (nam ely, sen
tencelike rep resentations m anipulated by formal rules). Y ou can see
also w h y this theoretical and experim ental approach has cap tu red the
interest of those w ho seek to understand how the m icroarchitecture
of the biological brain produ ces the p h en om en a displayed in h um an
and anim al cognition. Let us now explore the functional properties of
these netw orks in m ore detail, and see how they bear on som e of the
traditional issues in epistem ology and the philosophy of science.
5 Som e F unctional Properties o f B rainlike N etw orks
The netw orks described above are d escended from a device called the
P ercep tron (Rosenblatt 1959), w hich w as essentially just a tw o-layer
netw ork as opposed to a three-layer netw ork. D evices of this con
figuration could and did learn to discrim inate a considerable variety
of input p attern s. U nfortunately, having the input layer connected
directly to the outp u t layer im poses v ery severe limitations on the
range of possible transform ations a netw ork can perform (Minsky
and P ap ert 1969), and interest in Perceptron-like devices w as soon
eclipsed b y the m uch faster-m oving developm ents in standard
"p ro g ram -w ritin g " AI, w hich exploited the high-speed general-
p u rp ose digital m achines that w ere then starting to b ecom e widely
available. Throu ghout the seventies, research in artificial "n eu ral
n e ts" w as an u nd ergrou n d p rogram by com parison.
It has em erged from the shadow s for a num ber of reason s. O ne
im p ortan t factor is just the troubled doldrum s into w hich m ainstream
or program -w riting AI h as fallen. In m an y respects, th ese doldrum s
parallel the infertility of the classical approach to theories and learn
ing w ithin the philosophy of science. This is not surprising, since
m ain stream AI w as proceeding on m any of the sam e basic assu m p
tions about cognition, and m any of its attem p ts w ere just m achine
-im p lem en tation s of learning algorithm s proposed earlier by philo
sop hers of science and inductive logicians (G lym our 1988). The
failures of m ainstream A I— unrealistic learning, p oor perform ance in
com plex perceptual and m otor tasks, w eak handling of analogies,
and snail-like cognitive perform ance despite the use of very large and
fast m achin es— teach us even m ore dram atically than do the failures
of m ainstream philosophy that w e need to rethink the style of rep re
sentation and com putation that w e have b een ascribing to cognitive
creatures.
O th er reasons for the resurgen ce of interest in netw orks are m ore
positive. The introduction of additional layers of intervening or "h id
d en ” units produced a dram atic increase in the range of possible
transform ations that the netw ork could effect. A s Sejnowski et al.
describe it,
Only the first-order statistics of the input pattern can be captured

by direct connections betw een input and output units. The role
of the hidden units is to capture h igher-order statistical rela
tionships and this can be accom plished if significant underlying
features can be found that have stron g, regular relationships
w ith the p attern s on the visible units. The hard part of learning is
to find the set of w eights w hich turn the hidden units into useful
feature detectors. (1986, p. 264)
Equally im p ortan t is the S-shaped, nonlinear respon se profile (figure

9 .3 ) now assigned to every unit in the netw ork. So long as this re
sp onse profile rem ains linear, any netw ork will be limited to com p u t
ing purely linear transform ations. (A tra n sfo rm a tio n /(x ) is linear just
in case f ( n ■x) = n -f( x ) , and f ( x + y ) = f ( x ) + f ( y ) . ) But a nonlinear
respon se profile for each unit brings the entire range of possible
nonlinear transform ations w ithin reach of three-layer netw orks, a
dram atic expansion of their com putational potential. N ow there
are no transform ations that lie beyond the com putational p ow er of a
large en ou gh and suitably w eighted netw ork (H ornik et al. 1989).
A third factor w as the articulation, by Rum elhart, H inton, and
W illiam s (1986a), of the gen eralized delta rule (a generalization, to
three-layer netw orks, of R osenblatt's original teaching rule for adjust
ing the w eights of the P ercep tron), and the empirical discovery that
this new rule very rarely got p erm an en tly stuck in inefficient local
m inim a on its w ay tow ard finding the b est possible configuration of
connection w eights for a given netw ork and a given problem . This
w as a m ajor breakthrough, n ot so m u ch because "learn ing by the
back-propagation of e rro r," as it has com e to be called, w as just like
h u m an learning, but because it provided us w ith an efficient tech n ol
ogy for quickly training up various netw orks on various problem s, so

that w e could study their properties and explore their potential.
The w ay the generalized delta rule w orks can be m ade fairly intui
tive given the idea of an abstract w eight space as rep resen ted in figure
9.6 . C on sider any output vector produced by a netw ork w ith a speci
fic configuration of w eights, a configuration rep resented b y a specific
position in w eight space. Suppose that this outp u t v ector is in error
by various degrees in various of its elem ents. C onsider n ow a single
syn ap se at the ou p u t layer, and consider the effect on the output
vector th at a small positive or negative change in its w eight w ould
have h ad . Since the output vector is a determ inate function of the
sy stem 's w eights (assum ing w e hold the input vector fixed), w e can
calculate w hich of these tw o possible changes, if either, w ou ld have
m ade the greater im provem ent in the outp ut vector. The relevant
change is m ade accordingly. (For m ore detail, see R um elhart et al.
1986b.)
If a similar calculation is perform ed over every synapse in the net
w ork, and the change in its w eight is then m ad e accordingly, w hat
the resulting shift in the position of the system 's overall point in
w eight space am ounts to is a small slide dow n the steep est face of the
local " e rr o r su rface." N ote that there is no guarantee th at this in
crem ental shift m oves the system directly tow ard the global position
of zero error (that is w hy perfection cannot be achieved in a single
jum p). O n the contrary, the descending path to a global error mini
m um m ay be highly circuitous. N or is there any guarantee that the
system m u st eventually reach such a global m inim um . O n the con
trary, the dow n w ard path from a given starting point m ay well lead
to a m erely local m inim um , from w hich only a large ch an ge in the
sy stem 's w eights will afford escape, a change beyond the reach of the
delta rule. But in fact this happens relatively rarely, for it turns out
that the m ore dim ensions (synapses) a system has, the sm aller the
probability of there being an intersecting local m inim um in every one
of the available dim ensions. The global point is usually able to slide
dow n som e narrow cleft in the local top ograph y. Em pirically, then,
the back-propagation algorithm is surprisingly effective at driving the
system to the global error m inim um , at least w here w e can identify
that global m inim um effectively.
The ad vantage this algorithm provides is easily appreciated. The
possible com binations of w eights in a netw ork increases exp on en
tially w ith the size of the netw ork. If w e conservatively assum e
that each w eight adm its of only 10 possible values, the num ber of
distinct positions in w eight space (i.e., the num ber of possible w eight
-configurations) for the simple rock/m ine netw ork of figure 9 .5 is
already 10105! This space is far too large to explore efficiently w ithout
som ething like the generalized delta rule and the back propagation of
error to do it for us. But w ith the delta rule, adm inistered by an auxili
ary com pu ter, research ers have show n that netw orks of the simple
kind described are capable of learning som e quite extraordinary skills,
and of displaying som e highly intriguing properties. Let m e now re
turn to an exploration of these.
A n im portant exploratory technique in cognitive and behavioral
neuroscience is to record , w ith an im planted m icroelectrode, the elec
trical activity of a single neuron during cognition or behavior in the
intact anim al. This is relatively easy to do, and it does give us tantaliz
ing bits of inform ation about the cognitive significance of neural activ
ity (recall the results of Hubei and W iesel m entioned earlier). Single
cell recordings give us only isolated bits of inform ation, how ever, and
w hat w e w ould really like to m onitor are the patterns of sim ultaneous
neural activation across large num bers of cells in the sam e subsystem .
U nfortunately, effective techniques for sim ultaneously recording
from large num bers of adjacent cells are still in their infancy. The task
is extrem ely difficult.
By con trast, this task is extrem ely easy w ith the artificial netw orks
w e have been describing. If the netw ork is real hard w are, its units are
far m ore accessible than the fragile and m icroscopic units of a living
brain. A n d if the netw ork is m erely being sim ulated within a standard
com p uter (as is usually the case), one can w rite the p rogram so that
the activation levels of any unit, or set of units, can be read out on
com m an d. A ccordingly, once a netw ork has been successfully trained
up on som e skill or other, one can then exam ine the collective b e
havior of its units during the exercise of that skill.
W e have already seen the results of one such analysis in the ro ck /
m ine netw ork. O nce the w eights have reached their optim um con
figuration, the activation vectors (i.e., the p attern s of activation) at
the hidden layer fall into tw o disjoint classes: the vector space is parti
tioned in tw o, as depicted schem atically in figure 9 .7 . But a m ere bin
ary discrim ination is an atypically simple case. The read er N ETtalk,
for exam p le, partitions its hidden-unit v ector space into fully 79 sub
spaces. The reason is simple. For each of the 26 letters in the alphabet,
there is at least one phonem e assigned to it, and for m any letters
there are several ph on em es that m ight be signified, depending on the
lexical context. A s it hap p ens, there are 79 distinct letter-to-phonem e
associations to be learned if one is to m aster the pronunciation of
English spelling, and in the successfully trained netw ork a distinct
hidden-unit activation vector o ccurs w hen each of these 79 possible
transform ations is effected.
In the case of the rock/m ine netw ork, w e noted a similarity m etric
w ithin each of its tw o hidden-unit subspaces. In the case of NETtalk,
w e also find a similarity m etric, this time across the 79 functional
hidden-unit vectors (by 'functional vector' I m ean a v ector that cor
respon d s to one of the 79 desired letter-to-phonem e transform ations
in the trained netw ork). R osenberg and Sejnowski did a cluster analy
sis of these vectors in the trained netw ork. Roughly, their p rocedure
w as as follow s. They asked, for every functional vector in th at space,
W hat o th er such vector is closest to it? The answ ers yielded about 30
v ector pairs. They then constructed a secondary vector for each such
pair, by averaging the tw o original vectors, and asked, for ev ery such
secon d ary vector, W hat oth er secondary v ector (or so far unpaired
prim ary vector) is closest to it? This p rodu ced a sm aller set of
secon d ary-vector pairs, on w hich the averaging proced u re w as re
p eated to produce a set of tertiary vectors. These w ere then paired in
turn, and so forth. This p rocedure produ ces a hierarchy of groupings
am ong the original transform ations, and it com es to an en d w ith a
grand division of the 79 original vectors into tw o disjoint classes.
A s it h ap p en s, that d eep est and m ost fundam ental division w ithin
the hidden-unit v ector space corresp on d s to the division b etw een the
con son an ts and the vow els! Looking further into this h ierarch y, into
the con son an t b ranch, for exam ple, w e find that there are sub
divisions into the principal consonant typ es, and that w ithin these
b ran ch es there are further subdivisions into the m ost similar con
sonants. All of this is depicted in the tree diagram of figure 9 .8 . W hat
the netw ork has m an aged to recover, from its training set of several
thou san d English w ord s, is the highly irregular phonological signif
icance of standard English spelling, plus the hierarchical organization
of the p h on etic structure of English speech.
H ere w e have a clear illustration of tw o things at once. The first
lesson is the capacity of an activation-vector space to em body a rich
and w ell-structured hierarchy of categories, com plete w ith a similar
ity m etric em bracing everything w ithin it. A n d the secon d lesson is
the capacity of such netw orks to em body rep resentation s of factors
and p attern s that are only partially or implicitly reflected in the cor
pus of inputs. Though I did not m ention it earlier, the rock /m in e n et
w ork provides another exam ple of this, in that the final partition
m ad e on its hidden-unit vector space correspon d s in fact to the objec
tive distinction betw een sonar targets m ade of m etal and son ar targets
m ade of nonm etal. That is the true uniform ity that lies behind the
ap p arently chaotic variety displayed in the inputs.
It is briefly tem pting to suggest that N ETtalk has the con cep t of a
tiard c, for exam ple, and that the rock/m ine netw ork has the concept
MEASURE OF DISSIMILARITY
( DISTANCE BETWEEN VECTORS )
Figure 9.8
Hierarchy of partitions on hidden-unit vector space of NETtalk
of m etal. But this w o n 't really do, since the vector-sp ace rep resen ta
tions at issue do not play a conceptual or com putational role rem otely
rich en ou gh to m erit their assimilation to specifically h u m an con
cep ts. N evertheless, it is plain that both netw orks have contrived a
system of internal representations that truly corresp on d s to im p or
tant distinctions and structures in the outside w orld, stru ctu res that
are n o t explicitly represented in the corpus of their sen sory inputs.
The value of those representations is that they and only they allow
the netw orks to "m ak e sen se" of their variegated and often noisy
input co rp u s, in the sense that they and only they allow the netw ork
to resp o n d to those inputs in a fashion that system atically red u ces the
error m essag es to a trickle. These, I need hardly rem ind, are the func
tions typically ascribed to theories.
W h at w e are confronting here is a possible conception of know l
edge o r understan din g that ow es nothing to the sentential categories
of cu rrent com m on sense. A n individual's overall theory-of-the-
w orld, w e m ight v entu re, is not a large collection or a long list of
stored sym bolic item s. R ather, it is a specific point in that individual's
synaptic w eight space. It is a configuration of connection w eights, a
configuration that partitions the system 's activation-vector space(s)
into useful divisions and subdivisions relative to the inputs typically
fed the system . 'U seful' here m ean s 'tend s to minim ize the error m es
sages'.
A possible objection here points to the fact that differently w eight
ed sy stem s can produ ce the sam e, or at least roughly the sam e, parti
tions on their activation-vector spaces. A ccordingly, w e m ight try to
abstract from the idiosyncratic details of a system 's connection
w eigh ts, and identify its global theory directly w ith the set of parti
tions th ey produ ce w ithin its activation-vector space. This w ould
allow for differently w eighted system s to h ave the sam e theory.
There is som e virtue in this suggestion, but also som e vice. W hile
differently w eighted system s can em body the sam e partitions and
thus display the sam e outpu t perform ance on any given input, they
will still learn quite differently in the face of a p rotracted sequence of
new and problem atic inputs. This is because the learning algorithm
that drives the system to new points in w eight space d oes n ot care
about the relatively global partitions that have been m ad e in activa
tion sp ace. All it cares about are the individual w eights and how they
relate to ap p reh en ded error. The law s of cognitive evolution, there
fore, do n ot operate prim arily at the level of the partitions, at least on
the view of things here being explored. R ather, they operate at the
level of the w eights. A ccordingly, if w e w ant our "u n it of cognition"
Ло figure in the laws of cognitive developm ent, the point in w eight
space seem s the w iser choice of unit. W e need only concede that
different global theories can occasionally p rodu ce identical short-term
behavior.
The level of the partitions certainly corresp on d s m ore closely to the
"co n cep tu al" level, as u n derstood in com m on sense and traditional
theory, but the point is that this seem s not to be the m ost im portant
dynam ical level, even w hen explicated in neurocom putational term s.
K now ing a creatu re's vector-sp ace partitions m ay suffice for the accu
rate sh ort-term prediction of its behavior, but that know ledge is in
adequate to predict or explain the evolution of those partitions over
the course of time and cruel experience. K now ledge of the w eights,
by con trast, is sufficient for this task. This gives substance to the con
viction, voiced earlier in section 2, that to explain the phen om en on
of conceptual change, w e need to unearth a level of subconceptual
com binatorial elem ents within w hich different concepts can be
articulated, evaluated, and then modified according to their p er
form an ce. The connection w eights provide a level that m eets all of
these conditions.
This general view of how know ledge is em bodied and accessed in
the brain has som e further appealing features. If w e assum e that the
brains of the higher anim als w ork in som ething like the fashion out
lined, then w e can explain a num ber of puzzling features of h um an
and anim al cognition. For one thing, the speed-of-relevant-access
problem sim ply disappears. A netw ork the size of a hum an brain—
w ith 1011 n eurons, 103 connections on each , 1014 total connections,
and at least 10 distinct layers of hidden units-— can be exp ected , in
the course of grow ing up, to partition its internal vector spaces into
m an y billions of functionally relevant subdivisions, each responsive
to a broad but proprietary range of highly com plex stimuli. W h en the
n etw ork receives a stim ulus that falls into one of these classes, the
netw ork produ ces the appropriate activation vector in a m atter of
only tens or hun dreds of m illiseconds, b ecau se that is all the time it
takes for the parallel-coded stim ulus to m ake its w ay through only
tw o or three or ten layers of the m assively parallel netw ork to the
functionally relevant layer that drives the appropriate behavioral re
sponse. Since inform ation is stored not in a long list that m ust som e
h ow be search ed , but rather in the m yriad connection w eights that
configure the network, relevant aspects of the creature's total infor
m ation are autom atically accessed by the coded stimuli them selves.
A third advantage of this m odel is its explanation of the functional
p ersistence of brains in the face of m inor dam age, disease, and the
norm al but steady loss of its cells w ith age. H um an cognition d e
grad es fairly gracefully as the physical plant deteriorates, in sharp
con trast to the behavior of typical com pu ters, w hich h ave a very low
fault tolerance. The explanation of this persistence lies in the m assive
ly parallel ch aracter of the com putations the brain perform s, and in
the v ery tiny contribution that each synapse or each cell m akes to the
overall com putation. In a large netw ork of 100,000 units, the loss or
m isbehavior of a single cell will not even be detectable. A nd in the
m ore dram atic case of w idespread cell loss, so long as the losses are
m ore o r less random ly distributed throu ghou t the netw ork, the gross
ch aracter of the netw ork's activity will rem ain u n ch an ged . W h at hap
p ens is that the quality of its com putations will be progressively
d egraded .
Turning now tow ard m ore specifically philosophical con cern s, w e
m ay n ote an u n expected virtue of this approach concerning the m at
ter of sim plicity. This im portant notion p resen ts tw o problem s: it is
robustly resistant to attem pts to define or m easure it, and it is not
clear w h y it should be counted an epistem ic virtue in any case. There
seem s no obvious reason , either a priori or a posteriori, w hy the
w orld should be simple rath er than com plex, and epistem ic decisions
based on the con trary assum ption thus ap p ear arbitrary and unjusti
fied. Sim plicity, conclude som e (van Fraassen 1980), is a m erely
pragm atic or aesthetic virtue, as opposed to a genuinely epistem ic
virtue. But consider the following story.
The rock /m in e netw ork of figure 9 .5 displayed a strong capacity for
generalizing beyond the sam ple echoes in its training set: it can accu r
ately discrim inate entirely new sam ples of both kinds. But trained
netw orks do n ot alw ays generalize so well, and it is interesting w hat
d eterm ines their success in this regard . H ow well the training gen er
alizes is in part a function of how m any hidden units the system p os
sesses, or u ses to solve the problem . There is, it turns out, an optim al
num ber of units for any given problem . If the netw ork to be trained is
given m o re than the optim al num ber of hidden units, it will learn to
respon d appropriately to all of the various sam ples in its training set,
but it will generalize to new sam ples only v ery poorly. O n the other
h and, w ith less than the optim al num ber, it never really learns to
respon d appropriately to all of the sam ples in its training set.
The reason is as follows. D uring the training period th e netw ork
gradually g enerates a set of internal rep resentation s at the level of the
hidden units. O ne class of hidden-unit activation v ectors is ch aracter
istic of rocklike input vectors; anoth er class is characteristic of m ine
like inp u t vectors. D uring this period the system is theorizing at the
level of the hidden units, exploring the space of possible activation
vectors, in hopes of finding som e partition or set of partitions on it
-that the outp u t layer can then exploit in turn, so as to draw the
n eeded distinctions and thus bring the process of error-induced

synaptic adjustm ents to an end.
If there are far too m any hidden units, then the learning p rocess
can be partially subverted in the following w ay. The lazy system
ch eats: it learns a set of unrelated representation s at the level of the
hidden units. It learns a distinct representation for each sam ple input
(or for a small group of such inputs) draw n from the very finite train
ing set, a rep resentation that does indeed prom pt the correct re
sponse at the outp u t level. But since there is nothing com m on to all of
the hidden-unit rock representation s, or to all of the hidden-unit
m ine representations, an input vector from outside the training set
p ro d u ces a hidden-unit representation that bears no relation to the
representations already form ed. The system has not learned to see
w hat is com m on w ithin each of the tw o stim ulus classes, w hich w ould
allow it to generalize effortlessly to new cases that shared that com
m on feature. It h as just knocked togeth er an ad h oc "ю о к -up table"
that allows it to deal successfully w ith the limited sam ples in the
training set, at w hich point the error m essages cease, the w eights
stop evolving, and the system stops learning. (I am grateful to Terry
Sejnowski for m entioning to m e this wrinkle in the learning behavior
of typical netw orks.)
There are tw o w ays to avoid this ad h oc, unprojectible learning.
O ne is to enlarge dram atically the size of the training set. This will
overload the system 's ability to just "m e m o riz e " an adequate re
sp onse for each of the training sam ples. But a m ore effective w ay is
just to reduce the num ber of hidden units in the netw ork, so that it
lacks the resources to cobble togeth er su ch w asteful and ungeneraliz-
able internal representations. W e m ust red u ce them to the point
w h ere it has to find a single partition on the hidden-unit vector space,
a partition that p u ts all of the sam ple rock representations on one
side, and all of the sam ple m ine representations on the other. A sys
tem constrained in this w ay will generalize far better, for the global
partition it h as been forced to find corresp on d s to som ething com m on
to each m em ber of the relevant stim ulus class, even if it is only a
unifying dim ension of variation (or set of such dim ensions) that u n
ites them all by a similarity relation. It is the generation of that similar
ity relation that allows the system to respon d appropriately to novel
exam ples. They m ay be new to the system , but they fall on a sp ec
tru m for w hich the system now h as an adequate representation.
N etw orks w ith only a few hidden units in excess of the optim al
n um ber will som etim es spontaneously achieve the m axim ally simple
"h y p o th esis" despite the excess units. The few unneeded units are
slowly shut dow n by the learning algorithm during the course of
training. They becom e zero-valued elem ents in all of the successful

vectors. N etw orks will not alw ays do this, h ow ever. The n eed ed sim
plicity m u st generally be forced from the outside, by a progressive
reduction in the num ber of available hidden units.
O n the oth er hand, if the netw ork has too few hidden units, then it
lacks the resou rces even to express an activation vector th at is ade
quate to characterize the underlying uniform ity, and it will never
m aster com pletely even the smallish corpus of sam ples in the training
set. In oth er w ord s, simplicity m ay be a virtue, but the system m ust
com m and sufficient com plexity at least to m eet the task at hand.
W e have just seen how forcing a neural netw ork to generate a small
er n um ber of distinct partitions on a hidden-unit vector space of few
er dim ensions can produ ce a system w hose learning achievem ents
generalize m ore effectively to novel cases. Ceteris paribus, the simpler
hyp oth eses generalize better. Getting by w ith few er resou rces is, of
course, a virtue in itself, though a pragm atic one, to be su re. But this
is n ot the principal virtue here displayed. Superior generalization is a
genuinely epistem ic virtue, and it is regularly displayed b y netw orks
constrain ed, in the fashion described, to find the sim plest hypothesis
concerning w h atever structures m ight be hidden in, or behind, their
input vectors.
Of cou rse, nothing guarantees successful generalization: a netw ork
is alw ays h ostage to the quality of its training set relative to the total
population. A n d there m ay be equally simple alternative h yp oth eses
that generalize differentially well. Y et from the p erspective of the
relevant m icrodynam ics, w e can see at least one clear reason w hy
simplicity is m ore than a m erely pragm atic virtue. It is an episte
mic virtue, not principally because simple hypotheses avoid the vice
of being com plex, but because they avoid the vice of being ad hoc.
6 H ow Faithfully Do These N etw orks D epict the Brain?
The functional properties so far observed in these m odel netw orks are
an encou ragin g rew ard for the structural assum ptions that w ent into
them . But just how accurate are these m odels, as depictions of the
brain's m icrostru cture? A wholly appropriate answ er h ere is u n cer
tain, for w e continue to be uncertain about w hat features of the
brain's m icrostructure are and are not functionally relevant, and w e
are therefore uncertain about w hat is and is not a "legitim ate" sim
plifying assum ption in the m odels w e m ake. Even so, it is plain that
the m odels are inaccurate in a variety of respects. The point of the
p resen t section is to sum m arize and evaluate these failings. Let m e
Figure 9.9
a) A schematic section of the cerebellum (cell population and fiber density reduced for
clarity). b) Neural matrix
begin by underscoring the basic respects in w hich the m odels appear

to be correct.
It is true that real nervous system s display, as their principal orga
nizing featu re, layers or populations of neurons that p roject their
axons en m asse to som e distinct layer or population of n eu ron s, w here
each arriving axon divides into multiple branch es w hose end bulbs
m ake synaptic connections of various w eights onto m any cells at the
target location. This description captures all of the sen sory m odalities
and their prim ary relations to the brain; it captures the ch aracter of
the various areas of the central brain stem ; and it cap tu res the stru c
ture of the cerebral cortex, w hich in hum ans contains at least six dis
tinct layers of neurons, w here each layer is the source an d /o r the
target of an orderly projection of axons to an d /o r from elsew here.
It cap tu res the character of the cerebellum as well (figure 9.9 a), a
structure discussed in ch apter 5 in connection w ith the problem of
m otor control. I there described the cerebellum as h aving the struc
ture of a v ery large "m atrix m ultiplier," as schem atized in figure 9 .9 b.
Follow ing Pellionisz and Llinas 1982, I ascribed to this neural m atrix
the function of perform ing sophisticated transform ations on incom
ing activation vectors. This is, in fact, the sam e function perform ed
betw een any tw o layers of the three-layered netw orks described ear
lier, and the tw o cases are distinct only in the superficial details of
their w iring diagram s. A three-layered netw ork of the kind discussed
earlier is equivalent to a pair of neural m atrices connected in series, as
is illustrated in figure 9.10. The only substantive difference is that in
figure 9.10a the end b ranches synapse directly onto the receiving cell
body itself, while in 9.10b they synapse onto som e dendritic filam ents
extending out from the receiving cell body. The actual connectivity
within the tw o netw orks is identical. The cerebellum and the m otor
Figure 9.10
The equivalence of nets and matrices
end of natural system s, accordingly, seem to be further instances of

the gross p attern at issue.
But the details p resen t all m an n er of difficulties. To begin with
small ones, note that in real brains an arriving axon m akes synaptic
contact w ith only a relatively small percen tage of the thou san d s or
millions of cells in its target population, n ot w ith every last one of
them as in the m odels. This is not a serious difficulty, since m odel
netw orks w ith com parably p ared connections still m an age to learn
the required transform ations quite well, though p erhaps not so well
as a fully connected netw ork.
M ore serious, real axons, so far as is know n, have term inal end
bulbs that are uniform ly inhibitory, or uniform ly excitatory, accord
ing to the type of neuron. W e seem not to find a m ixture of both kinds
of connections radiating from the sam e n eu ron , nor do w e find con
nections changing their sign during learning, as is the case in the
m odels. M oreover, that m ixture of positive and negative influences is
essential to successful function in the m odels: the sam e input cell
m u st be capable of inhibiting som e cells dow n the line at the sam e
time that it is busy exciting oth ers. Fu rth er, cell populations in the
brain typically show extensive "h o rizo n ta l" cell-to-cell connections
w ithin a given layer. In the m odels there are none at all (see, e .g .,
figure 9 .4 ). Their connections join units only to units in distinct
layers.
These last tw o difficulties m ight conceivably serve to cancel each
oth er. O ne w ay in w hich an excitatory end bulb m ight serve to inhibit
a cell in its target population is first to m ake an excitatory connection
onto one of the m any small interneurons typically scattered throu gh
out the target population of m ain n eurons, w hich interneuron has
m ad e an inhibitory synaptic connection onto the target m ain neuron.
Exciting the inhibitory interneuron w ould then have the effect of
inhibiting the m ain neu ron , as desired. A nd such a system w ould
display a large num ber of short horizontal intralayer connections, as
is observed. This is just a suggestion, h ow ever, since it is far from
clear that the elem ents m entioned are predom inantly connected in
the m an ner required.
M ore serious still, there are several m ajor problem s w ith the idea
that netw orks in the brain learn by m eans of the learning algorithm
so effective in the m odels, the proced u re of back propagating
ap p reh en ded errors according to the generalized delta rule. That p ro
cedure requires tw o things: (1) a com putation of the partial correction
needed for each unit in the outp u t layer, and via these corrections a
com putation of a partial correction for each unit in the earlier layers,
and (2) a m eth od of causally conveying these correction m essages
back throu gh the netw ork to the sites of the relevant synaptic con n ec
tions in such a fashion that each w eight gets nudged up or dow n
accordingly. In a com p uter simulation of the netw orks at issue (which
is currently the standard technique for exploring their properties),
both the com putation and the subsequent w eight adjustm ents are
easily d on e: the com putation is done outside the netw ork by the host
com p u ter, w hich has direct access to and control over every elem ent
of the n etw ork being sim ulated. But in the self-contained biological
brain, w e have to find som e real source of adjustm ent signals and
som e real p athw ays to convey them back to the relevant un its. U nfor
tunately, the empirical brain displays little that an sw ers to exactly
these requirem ents.
N ot that it contains nothing along these lines: the prim ary ascen d
ing p ath w ays already described are typically m atch ed by reciprocal or
"d e scen d in g " p ath w ays of com parable density. These allow higher
layers to have an influence on affairs at low er layers. Y et th e influence
ap p ears to be on the activity levels of the low er cells them selves,
rath er th an on the m yriad synaptic connections w hose w eights need
adjusting during learning. There m ay be indirect effects on the
syn apses, of course, but it is far from clear that the b rain's wiring
d iagram an sw ers to the dem ands of the back-propagation algorithm .
The case is a little m ore prom ising in the cerebellum (figure 9 .9 a),
w hich contains a second m ajor input system in the aptly nam ed
clim bing fib ers (not show n in the diagram for reasons of clarity). These
fibers envelop each of the large Purkinje cells from below in the sam e
fashion that a climbing ivy envelops a giant oak, w ith its filam entary
tendrils reaching well up into the bushy dendritic tree of the Purkinje
cell, w hich tree is the locus of all of the synaptic connections m ad e by
the incom ing parallel fibers. The climbing fibers are thu s at least
roughly positioned to do the job that the back-propagation algorithm
requires of them , and they are distributed one to each Purkinje cell, as
consistent delivery of the error m essage requires (T hom pson 1986).
Equally, they m ight serve som e other quite different learning algor
ithm , as advocated by Pellionisz and Llinas (1985). U nfortunately,
there is as y et no com pelling reason to believe that the m odification of
the w eights of the parallel-fiber-to-Purkinje-dendrite syn ap ses is even
w ithin the causal p ow er of the climbing fibers. N or is there any
clear reason to see either the climbing fibers in the cerebellum , or the
d escen ding p ath w ays elsew here in the brain, as the bearers of any
appropriately com p u ted error-correction m essages appropriate to
need ed synaptic change.
O n the h ard w are side, therefore, the situation does not su p p ort the
idea that the specific back-propagation p roced u re of R um elhart et al.
is the brain's central m echanism for learning. (N or, it should be m en

tioned, did they claim that it is.) A nd it is implausible on som e func
tional groun d s as well. First, in the p rocess of learning a recognition
task, living brains typically show a progressive reduction in the reac
tion time required for the recognitional outp u t respon se. W ith the
delta rule, h ow ever, learning involves a progressive reduction in
error, but reaction tim es are con stan t throu ghou t. A second difficulty
w ith the delta rule is as follows. A n ecessary elem ent in its calculated
ap p ortionm en t of error is a representation of w hat w ould have been
the correct vector in the output layer. That is w hy back propagation is
said to involve a global teacher, an inform ation source that always
know s the correct answ ers and can therefore provide a perfect m ea
sure of outp u t error. Real creatures generally lack any such perfect
inform ation. They m u st struggle along in the absence of any sure
com p ass tow ard the truth, and their synaptic adjustm ents m u st be
based on m uch p oorer inform ation.
A nd yet their brains learn. W hich m ean s that som ehow the con
figuration of their synaptic w eights m u st u ndergo ch an ge, change
steered in som e w ay by error or related dissatisfaction, change that
carves a path tow ard a regim e of decreased error. Know ing this
m u ch , and know ing som ething about the m icrostructure and m icro
d ynam ics of the brain, w e can explore the space of possible learning
p rocedu res with som e idea of w hat features to look for. If the general
ized delta rule is not the brain's p roced u re, as it seem s not to be, there
rem ain other possible strategies for back-propagating sundry error
m easu res, strategies that m ay find m ore detailed reflection in the
brain. If these prove unrealizable, there are other proced u res that do
n ot require the organized distribution of any global error m easures at
all; they d epend primarily on local constraints (H inton and Sejnowski
1986; H opfield and Tank 1985; Barto 1985; Bear et al. 1987).
O ne of these is w orth y of m ention, since som ething along these
lines does appear to be displayed in biological brains. H ebbian learn
ing (so called after D. O. H ebb, w ho first p roposed the m echanism ) is
a process of w eight adjustm ent that exploits the tem poral coinci
d en ce, on either side of a given synaptic junction, of a strong signal in
the incom ing axon and a high level of excitation in the receiving cell.
W h en such conjunctions occu r, Hebb p roposed , som e physical or
chem ical change is induced in the synapse, a change that increases its
w eight. Of cou rse, high activation in the receiving cell is typically
caused by excitatory stim ulation from m an y oth er incom ing axons,
and so the im p ortan t tem poral coincidence here is really betw een
high activation am ong certain of the incom ing axons. Those w hose
high activation coincides w ith the activation of m an y others have
their subsequent influence on the cell increased. C rudely, those w ho

vote w ith w inners becom e w inners.
A H ebbian w eight-adjusting p rocedure can indeed p rodu ce learn
ing in artificial netw orks (Linsker 1986), although it does n ot seem to
be as general in its effectiveness as is back propagation. O n the other
hand, it has a m ajor functional advantage over back propagation. The
latter has scaling problem s, in that the p rocess of calculating and dis
tributing the relevant adjustm ents expan ds geom etrically w ith the
num ber of units in the netw ork. But Hebbian adjustm ents are locally
driven; they are independent of one another and of the overall size of
the netw ork. A large netw ork will thus learn just as quickly as a small
one. Indeed, a large netw ork m ay even show a slight ad van tage over
a sm aller one, since the tem poral coincidence of incom ing stim ula
tions at a given cell will be better and b etter defined w ith increasing
num bers of incom ing axons.
W e m ay also postulate "an ti-H ebbian " p rocesses, as a m ean s of
reducing synaptic w eights instead of increasing them . A nd w e need
to explore various possible flavors of each . W e still h ave very little
u n d erstan din g of the functional properties of these alternative learn
ing strategies. N or are w e at all sure that Hebbian learning, as de
scribed above, is really how the brain typically adjusts its w eights.
There d oes seem to be a good deal of activity-sensitive synaptic m o d
ification occurring in the brain, but w heth er its profile is specifically
Hebbian is not yet established. N or should w e exp ect the brain to
confine itself to only one learning strategy, for even at the behavioral
level w e can discern distinct types of learning. In sum , the problem
of w h at m echanism s actually produ ce synaptic change during
learning is an unsolved problem . But the functional success of the
generalized delta rule assures us that the problem is solvable in
principle, and oth er m ore plausible p roced u res are currently un d er
active exploration.
W hile the m atter of how real neural netw orks g en erate the right
configuration of w eights rem ains obscure, the m atter of how they
perform their various cognitive tasks once configured is a good deal
clearer. If even small artifical netw orks can perform the sophisticated
cognitive tasks illustrated earlier in this pap er, there is no m ystery
that real netw orks should do the sam e or better. W h at the brain dis
plays in the w ay of hardw are is not radically different from w h at the
m odels contain, and the differences invite exploration rath er than dis
ap pointm ent. The brain is, of course, very m uch larger and d enser
than the m odels so far constructed. It has m any layers rath er than just
tw o or three. It boasts p erh ap s a hun dred distinct and highly special
ized cell typ es, rather than just one. It is not a single и-layer n et
w ork, but rath er a large com m ittee of distinct but parallel netw orks,
interacting in sundry w ays. It plainly com m an d s m any spaces of
stunning com plexity, and m any skills in consequence. It stands as a
glow ing invitation to m ake our hum ble m odels yet m ore and m ore
realistic, in h op es of unlocking the m any secrets rem aining.
7 C om putational N euroscience: The N aturalization o f E pistem ology
O ne test of a new fram ew ork is its ability to throw a new and unifying
light on a variety of old ph en om ena. I will close this essay w ith an
exploration of several classic issues in the philosophy of science. The
aim is to recon stru ct them within the fram ew ork of the com putational
neuroscience outlined above. In section 5 w e saw how this could be
done for the case of theoretical simplicity. W e there saw a new w ay of
conceiving of this feature, and found a new perspective on w hy it is a
genuine epistem ic virtue. The hope in w h at follows is that w e m ay do
the sam e for other problem atic notions and issues.
A good place to begin is w ith the issue of foundationalism . H ere
the central bone of contention is w heth er our observation judgm ents
m u st alw ays be theory laden. The traditional discussion endures
largely for the good reason that a great deal hangs on the outcom e,
but also for the less m om entous reason that there is am biguity in
w h at one m ight wish to count as an "ob servation ju d g m en t" (an ex
plicitly u ttered sentence? a covert assertion? a propositional attitude?
a conscious experience? a sensation?), and a slightly different issue
em erges depending on w here the debate is located.
But from the perspective of this essay, it m akes no difference at
w h at level the issue m ight be located. If our cognitive activities arise
from a w eave of netw orks of the kind d iscussed above, and if w e
con strue a global theory as a global configuration of synaptic w eights,
as outlined in section 5, then it is clear that no cognitive activity w h at
ever takes place in the absence of v ectors being processed by som e
specific configuration of w eights. That is, no cognitive activity w h at
ever takes place in the absence of som e theory or other.
This perspective bids us see even the sim plest of anim als and the
y ou n gest of infants as possessing theories, since they too process
their activation vectors w ith som e configuration of w eights or other.
The difference betw een us and them is not that they lack theories.
R ather, their theories are just a good deal sim pler than ours, in the
case of anim als. A nd their theories are m u ch less coherent, less orga
nized, and less inform ed than ours, in the case of hum an infants.
W hich is to say, they have yet to achieve points in overall w eight
space that partition their activation-vector spaces into useful and
w ell-structured subdivisions. But insofar as there is cognitive activity

at all, it exploits w h atever theory the creature em bodies, h ow ever
useless or incoherent it m ight be.
The only place in the netw ork w here the w eights need play no role
is at the absolute sensory periphery of the system , w here the external
stim ulus is transduced into a coded input vector for subsequent deliv
ery to the transform ing layers of w eights. H ow ever, at the first occa
sion on w hich these p reconceptual states have any effect at all on the
d ow n stream cognitive system , it is throu gh a changeable configura
tion of synaptic w eights, a configuration that p rodu ces one set of
partitions on the activation-vector space of the relevant layer of
n eu ron s, one set out of millions of alternative possible sets. In other
w ord s, the very first thing that happens to the input signal is that it
gets conceptualized in one of m any different possible w ays. A t subse
quent layers of processing, the sam e process is rep eated , and the
m essage that finally arrives at the linguistic centers, for exam p le, has
been shap ed at least as m u ch by the partitional constraints of the
em bedded conceptual system (s) throu gh w hich it has passed as by
the distant sensory input that started things off.
F rom the perspective of com putational n eu roscience, therefore,
cognition is constitutionally theory laden. Presum ptive processing is
not a blight on w hat w ould otherw ise be an unblem ished activity; it is
just the natural signature of a cognitive system doing w h at it is sup
posed to be doing. It is just possible that som e theories are en d o
genously specified, of course, but this will change the p resen t issue
not at all. Innateness prom ises no escape from being theory laden, for
an en d ogen ou s theory is still a theory.
In an y case, the idea is not in general a plausible one. The visual
system , for exam ple, consists of som ething in the neighborhood of
1010 n eu ron s, each of w hich enjoys better than 103 synaptic con n ec
tions, for a total of at least 1013 w eights each w anting specific genetic
determ ination. That is an implausibly heavy load to place on the
coding capacity of our D N A m olecules. (The entire h u m an genom e
contains only about 109 nucleotides.) It w ould be m uch m ore efficient
to specify endogenously only the general structural principles of a
type of learning netw ork that is then likely to learn in certain standard
directions, given the standard sorts of inputs and error m essages that
a typical h u m an upbringing provides. This places the b u rd en of steer
ing o u r conceptual developm ent w here it belongs: on the external
w orld, an inform ation source far larger and m ore reliable than the
genes.
It is a com m onplace that w e can co n stru ct endlessly different
theories w ith w hich to explain the familiar facts of the observable
w orld. But it is an im m ediate consequence of the perspective here

adopted that that w e can also apprehend the "observable w o rld " it
self in a similarly endless variety of w ays. For there is no preferred set
of partitions into w hich our sensory spaces m u st inevitably fall. It all
d ep en ds on how the relevant netw orks are taught. If w e system atical
ly change the p attern of the error m essages delivered to the develop
ing netw ork, then even the very sam e history of sensory stim ulations
will produ ce a quite differently w eighted netw ork, one that partitions
the w orld into classes that cross-classify those of current com m on
sense, one that finds perceptual similarities along dim ensions quite
alien to the ones w e currently recognize, one that feeds its outp u ts
into a very differently configured netw ork at the higher cognitive
levels as well.
In relatively small w ays this p h en om en on is already familiar to us.
Specialists in various fields, people required to spend years m astering
the intricacies of som e dom ain of percep tion and m anipulation, reg
ularly end up being able to perceive facts and to anticipate behaviors
th at are w holly opaque to the rest of us. But there is no reason w hy
such variation should be confined to isolated skills and specialized
un derstan din g. In principle, the h um an cognitive system should be
capable of sustaining any one of an enorm ou s variety of decidedly
global theories concerning the character of its co m m on sen se Lebens-
ivelt as a w hole. (This possibility, defended in Feyerabend 1965, is
explored at som e length via exam ples in C hurchland 1979. For
extend ed criticism of this general suggestion, see Fod or 1984. For a
rebuttal and a counterrebuttal see chap ter 12 below and Fod or 1988.)
To appreciate just how great is the conceptual variety that aw aits
u s, consider the following num bers. W ith a total of p erhaps 1011
n eu ron s w ith an average of at least 103 connections each, the h u m an
brain has som ething like 1014 w eights to play w ith. If w e co n serv a
tively supp ose that each w eight adm its of only 10 possible values, the
total num ber of distinct possible configurations of synaptic w eights
(that is, distinct possible positions in w eight space) is 10 for the first
w eight, tim es 10 for the second w eight, tim es 10 for the third w eight,
e tc., for a total of 101014, or lO100-000-000-000'000! This is the total num ber
of (just barely) distinguishable theories em braceable by a hu m an ,
given the cognitive resources w e currently com m and. To put this
num ber into som e rem otely adequate perspective, recall that the total
num ber of elem entary particles in the entire universe is only about
1087.
In this w ay does a neurocom putational ap proach to perception
allow us to recon stru ct an old issue, and to provide novel reason s for
the view that our perceptual know ledge is both theory laden and
highly plastic. A nd it will do m ore. N otice that the activation-vector
spaces that a m atured brain has generated, and the p rototyp es they
em body, can en com p ass far m ore than the sim ple sensory types such
as p h on em es, colors, sm ells, tastes, faces, and so forth. G iven high
dim ensional spaces, w hich the brain h as in abundance, those spaces
and the p rototypes they em body can en com p ass categories of great
com plexity, generality, and abstraction, including those w ith a tem
poral dim ension, such as harm onic oscillator, projectile, traveling
w ave, Sam ba, tw elve-bar blues, dem ocratic election, six-course din
ner, courtship, elephant h u n t, civil disobedience, and stellar collapse.
It m ay be th at the input dim ensions that feed into su ch abstract
spaces will them selves often have to be the expression of som e earlier
level of processing, but that is no problem . The netw orks und er dis
cussion are hierarchically arranged to do precisely this as a m atter of
course. In principle, then, it is no h ard er for such a system to rep re
sent typ es of processes, procedures, and techniques than to rep resen t the
"sim p le" sensory qualities. From the point of view of the brain, these
are just m ore high-dim ensional vectors.
This offers us a possible m eans for explicating the notion of a p ara
digm , as used by T. S. Kuhn in his arresting characterization of the
nature of scientific understanding and d evelopm ent (K uhn 1962). A
p aradigm , for Kuhn, is a prototypical application of som e set of m ath e
m atical, conceptual, or instrum ental resou rces, an application ex
pected to have distinct but similar instances w hich it is the job of
norm al science to discover or construct. Becom ing a scientist is less a
m atter of learning a set of laws than it is a m atter of m asterin g the
details of the prototypical applications of the relevant resou rces in
such a w ay that one can recognize and generate further applications
of a relevantly similar kind.
K uhn w as criticized for the vagu en ess of the notion of a paradigm ,
and for the unexplicated criterion of similarity that clustered further
applications around it. But from the perspective of the n eu rocom pu -
tational ap p roach at issue, he can be vindicated on both cou n ts. For a
brain to com m an d a paradigm is for it to have settled into a w eight
configuration that p rodu ces som e w ell-structured similarity space
w hose central hypervolum e locates the prototypical application(s).
A n d it is only to be expected that even the m ost reflective p erson will
be incom pletely articulate on w hat dim ensions constitute this highly
com plex and abstract space, and even less articulate on w h at m etric
distributes exam ples along each dim ension. A com plete an sw er to
these questions w ould require a m icroscopic exam ination of the p er
son 's brain. That is one reason w hy exp osu re to a w ealth of exam ples
is so m uch m ore effective in teaching the techniques of any science

than is exp osu re to any attem pt at listing all the relevant factors. W e
are seldom able to articulate them all, and even if w e w ere able, listing
th em is n ot the best w ay to help a brain con stru ct the relevant internal
similarity space.
K uhn m akes m uch of the resistance typically show n by scientific
com m unities to change or displacem ent of the current paradigm . This
stubbornness here em erges as a natural expression of the w ay in
w hich netw orks learn, or occasionally fail to learn. The p rocess of
learning by gradient descent is alw ays threatened by the p rosp ect of a
purely local m inim um in the global error gradient. This is a position
w here the error m essages are not yet zero, but w here every sm all
ch an ge in the system produces even larger errors than those currently
en co u n tered . W ith a very high-dim ensional space the probability of
there being a sim ultaneous local m inim um in every dim ension of the
w eight space is small: there is usually som e narrow cleft in the
can yon out w hich the configuration point can eventually trickle,
thence to continue its w andering slide d ow n the error gradient and
tow ard som e truly global m inim um . But genuine local m inim a do
occu r, and the only w ay to escape them , once cau gh t, is to introduce
som e sort of ran d om noise into the system in hopes of bouncing the
sy stem 's configuration point out of such tem pting cul-de-sacs. Fu rth
erm ore, even if a local quasi m inim um does have an escape path
along one or m ore dim ensions, the error gradient along them m ay
there be quite shallow , and the system m ay take a very long time to
find its w ay out of the local im passe.
Finally, and just as im portant, the system can be victim ized by a
highly biased "train in g s e t." Suppose the system has reached a
w eight configuration that allows it to respon d successfully to all of the
exam ples in the (narrow and biased) set it has so far encou n tered .
Subsequent exposure to the larger dom ain of m ore diverse exam ples
will not necessarily result in the system 's m oving any significant dis
tance aw ay from its earlier configuration, unless the relative frequen
cy w ith w hich it encou n ters those new and anom alous exam ples is
quite high. For if the encou n ter frequency is low, the im pact of those
exam ples will be insufficient to overcom e the gravity of the false m ini
m u m that captured the initial training set. The system m ay require
"b litzin g " by new exam ples if their collective lesson is ever to "sin k
in ."
E ven if w e do p resent an abundance of the new and diverse exam
ples, it is quite likely that the delta rule d iscussed earlier will force the
system through a sequence of new configurations that perform very
poorly indeed w hen re-fed exam ples from the original training set.
This tem p orary loss of perform ance on certain previously "u n d e r

sto o d " cases is the price the system pays for the chance at achieving a
b roader payoff later, w hen the system finds a new and deep er error
m inim um . In the case of an artificial system chugging coolly aw ay at
the b eh est of the delta rule, such tem p orary losses need not im pede
the learning p rocess, at least if the frequency of new exam ples is suffi
ciently high. But w ith hum ans the im pact of such a loss is often m ore
keenly felt. The new exam ples that confound the old configuration
m ay sim ply be ignored or rejected in som e fashion, or they m ay
be quarantined and m ad e the target of a distinct and disconnected
learning p rocess in som e adjacent netw ork. Recall the exam ple of
sublunary and superlunary physics.
This raises the issue of explanatory unity. A creature throw n u n
p rep ared into a com plex and unforgiving w orld m u st take its un d er
standing w herever it can find it, even if this m eans generating a
d isconnected set of distinct similarity spaces each providing the
creature w ith a roughly appropriate respon se to som e of the m ore
p ressing types of situation it typically en cou n ters. But far b etter if it
then m an ag es to generate a single similarity space that unifies and
replaces the variation that used to reside in tw o entirely distinct
and sm aller spaces. This provides the creature w ith an effective grasp
on the p h en om ena that lay betw een the tw o classes already dealt w ith,
but w hich w ere successfully com prehended by neither of the tw o old
sp aces. These are p hen om en a that the creature had to ignore, or
avoid, or simply end u re. W ith a new and m ore com prehensive simi
larity space now generating system atic respon ses to a w ider range of
p h en om en a, the creature has succeeded in a small piece of conceptual
unification.
The payoff here recalls the virtue earlier discovered for simplicity.
Indeed, it is the sam e virtue, nam ely, superior generalization to cases
b eyon d those already encou n tered . This result w as achieved, in the
case described in section 5, by reducing the num ber of hidden units,
w hich forced the system to m ake m ore efficient use of the rep resen ta
tional resou rces rem aining. This m ore efficient use is realized w hen
the sy stem partitions its activation-vector space into the m inimal
num ber of distinct similarity subspaces consistent w ith red u cin g the
error m essag es to a m inim um . W h en com pleted, this p rocess also
p rodu ces the m axim al organization w ithin and am on g those sub
sp aces, for the system has found those enduring dim ensions of varia
tion th at successfully unite the diversity confronting it.
Tradition speaks of developing a single "th e o ry " to explain every
thing. K uhn (1962) speaks of extending and articulating a "p a ra d ig m "
into novel dom ains. Kitcher (1981, 1989) speaks of expanding the
range of application of a given "p attern of a rg u m e n t." It seem s to m e

that w e m ight unify and illuminate all of these notions by thinking in
term s of the evolving structure of a hidden-unit activation-vector
space, and its developm ent in the direction of representing all input
v ectors som ew here w ithin a single similarity space.
This m ight seem to offer som e hope for a C on vergen t Realist posi
tion w ithin the philosophy of science, but I fear that exactly the op p o
site is the case. For one thing, nothing guarantees that w e h u m an s
will avoid getting perm anently stuck in som e v ery deep but relatively
local error m inim um . For an other, nothing guarantees that there
exists a possible configuration of w eights that w ould reduce the error
m essages to zero. A unique global error m inim um relative to the h u
m an neural netw ork there m ay be, but for us and for any oth er finite
system interacting w ith the real w orld, it m ay alw ays be n onzero.
A nd for a third thing, nothing guarantees that there is only one global
m inim um . P erhaps there will in general be m any quite different mini
m a, all of them equally low in error, all of them carving up the w orld
in quite different w ays. W hich one a given thinker reaches m ay be a
function of the idiosyncratic details of its learning history. These con
siderations seem to rem ove the goal itself— a unique tru th — as well
as an y sure m ean s of getting there. W hich suggests that the p roper
cou rse to pursu e in epistem ology lies in the direction of a highly
naturalistic and pluralistic form of pragm atism . For a running start on
precisely these them es, see M unevar 1981 and Stich 1989.
8 C oncluding Rem arks
This essay opened w ith a survey of the problem s plaguing the classi
cal or "sen ten tial" approach to epistem ology and the philosophy of
science. I have tried to sketch an alternative approach that is free of all
or m ost of those problem s, and h as som e novel virtues of its ow n.
The following points are w orth noting. Simple and relatively small
netw orks of the sort described above have already dem onstrated the
capacity to learn a wide range of quite rem arkable cognitive skills and
capacities, som e of w hich lie beyond the reach of the older ap proach
to the nature of cognition (e .g ., the instantan eou s discrim ination of
subtle p erceptual qualities, the effective recognition of similarities,
and the real-tim e adm inistration of com plex m otor activity). W hile
the specific learning algorithm currently used to achieve these results
is unlikely to be the brain's algorithm , it does provide an existence
proof: by procedu res of this general sort, netw orks can indeed learn
w ith fierce efficiency. A nd there are m any oth er p rocedures aw aiting
our exploration.
The picture of learning and cognitive activity here painted en co m

passes the entire anim al kingdom . Cognition in h um an brains is fun
dam entally the sam e as cognition in brains generally. W e are all of us
processing activation vectors through artfully w eighted netw orks.
This b road conception of cognition pu ts cognitive theory firmly in
con tact w ith neurobiology, w hich adds a very strong set of co n
straints on the form er, to its substantial long-term advantage.
C on ceptu al change is no longer a problem . It h appens con tin u ou s
ly in th e norm al course of all cognitive developm ent. It is sustained by
m any sm all changes in the underlying h ard w are of synaptic w eights,
w hich chan ges gradually repartition the activation-vector spaces of
the affected population of cells. C onceptual sim plicity is also clearer
w hen view ed from a neurocom putational perspective, both in its
nature and in its epistem ological significance.
The old problem of how to retrieve relevant inform ation is tran s
form ed by the realization that it does n ot need to be "re trie v e d ." In
form ation is stored in brainlike netw orks in the global p attern of their
synaptic w eights. A n incom ing vector activates the relevant portions,
dim ensions, and subspaces of the trained netw ork by virtue of its
ow n vectorial m akeup. E ven an incom plete version of a given vector
(that is, one w ith several elem ents m issing) will often provoke essen
tially the sam e response as the com plete vector by reason of its re
levant similarity. For exam ple, the badly w histled first few bars of a
familiar tune will generally evoke both its nam e and the rest of the
entire piece. A nd it can do this in a m atter of m illiseconds, because
even if the subject know s thou san d s of tun es, there are still n o lists to
be searched.
It rem ains for this approach to com p reh en d the highly discursive
and linguistic dim ensions of h um an cognition, those that m otivated
the classical view of cognition. W e need n ot pretend that this will be
easy, b u t w e can see how to start. W e can start by exploring the
capacity of netw orks to m anipulate the structure of existing language,
its syntax, its sem antics, its pragm atics, and so forth. But w e m ight
also try som e novel ap p roach es, such as allowing each of tw o distinct
netw ork s, w h ose principal concerns and activities are nonlinguistic,
to try to learn from scratch som e system atic m ean s of m anipulating,
throu gh a p roprietary dim ension of input, the cognitive activities of
the oth er netw ork. W h at system of m utual m anipulation— w hat
language— m ight they develop?
The p receding pages illustrate som e of the system atic insights that
aw ait us if w e adopt a m ore naturalistic ap p roach to traditional issues
in epistem ology, an approach that is grounded in com putational
n eu roscience. H ow ever, a recurring them e in con tem porary philoso
p h y is that norm ative epistem ology cann ot be "n atu ralized " or recon
stru cted w ithin the fram ew ork of any purely descriptive scientific
theory. N otions such as "justified belief" and "ratio n ality ," it is said,
can n ot be adequately defined in term s of the nonnorm ative cate
gories to w hich any natural science is restricted, since "o u g h t" cannot
be derived from " i s ." Conclusions are then draw n from this to the
principled au ton om y of epistem ology from any natural science.
W hile it m ay be true that norm ative discourse cannot be replaced
w ithout rem ainder by descriptive discourse, it w ould be a distortion
to rep resent this as the aim of those w ho w ould naturalize epistem ol
ogy. The aim is rath er to enlighten our norm ative en d eavors by
recon stru ctin g them w ithin a m ore adequate conception of w hat
cognitive activity consists in, and thus to free ourselves from the
b u rd en of factual m isconceptions and tunnel vision. It is only the
autonom y of epistem ology that m ust be denied.
A u ton om y m ust be denied because norm ative issues are never in
d ep en den t of factual m atters. This is easily seen for our judgm ents of
instrum ental value, as these alw ays d epend on factual prem ises
about causal sufficiencies and dependencies. But it is also true of our
m ost basic norm ative con cepts and our jud gm en ts of intrinsic value,
for these have factual presuppositions as well. W e speak o f ju stifica
tion, but w e think of it as a feature of belief, and w heth er or not there
are any beliefs and w hat properties th ey have is a robustly factual
m atter. W e speak of rationality, but w e think of it as a feature of think
ers, and it is a substantive factual m atter w h at thinkers are and w hat
cognitive kinem atics they harbor. N orm ative concepts and norm ative
convictions are thus alw ays hostage to som e background factual p re
suppositions, and these can alw ays prove to be superficial, confused,
or just plain w rong. If they are, then w e m ay have to rethink w h at
ever norm ative fram ew ork has b een erected u pon them . The lesson
of the preceding pages is that the time for this has already com e.
Chapter 10
On the Nature of Explanation: A PDP Approach
N eu ral n etw ork m odels o f sensory processing and associative m em ory provide
resources that allow us to state a new theory o f w hat explanatory u n d er
standing consists in. That theory fin d s the theoretically im portant fa ctors to
reside not at the level o f propositions and the relations betw een them , but at
the level o f the activation patterns across large populations o f neurons. This
theory portrays explanatory understanding and perceptual recognition
as being d ifferent instances o f the sam e m ore gen eral sort o f cognitive achieve
m ent: p ro to ty p e activation. It thus purports to effect a unification o f the
theory o f explanation and the theory o f perception. It also fin d s a system atic
and revealin g unity am ong the w ide diversity o f types o f explanation (causal,
fu n ctio n a l, m athem atical, intentional, reductive, etc.), a diversity that has
been a chronic problem f o r older theories o f explanation.
1 Introduction
The notion of explanation has figured centrally in m ost con tem porary
accoun ts of scientific know ledge and rational belief. Explanation is
usually cited, along w ith prediction, as one of the tw o principal func
tions of our factual beliefs. A nd the rationality of such beliefs is com
m on ly said to be m easured, at least in part, by the relative range or
quality of the explanations they m ake possible. If som ething like this
is co rrect, then it is im portant for us to try to u n d erstan d w hat
explanation is, and w hat distinguishes a good explanation from a
p oor one.
Several existing accounts attem pt to m eet this challenge. They will
be ad d ressed below . The present p aper proposes a new accou n t of
the m atter— the prototype activation m odel— an account distinguished,
for starters, by its being grounded in a novel and u n orth od ox con cep
tion of w hat cognition consists in. That conception derives from cur
rent research in cognitive neurobiology and from parallel distributed
processing (PD P) m odels of brain function. (See R um elhart et al.
1986a, 1986b; C hurchland 1986a, 1989.) These PDP m odels are n ote
w orth y for m any reason s, but first am ong them in the p resen t context
is their alm ost com plete dissociation from the sentential or propositional
conception of w hat know ledge consists in, and from the conception
of h u m an inform ation processing as rule-governed inference. Those
venerable conceptions play a central role in all of the older accounts of
explanation, and in orth od ox accounts of cognition generally. They
will play alm ost no role in the accou n t to be p roposed .
This is n ot to say that the older accounts of explanation are entirely
w ithout integrity. O n the contrary, from the PDP perspective w e can
see w h y each of them has the appeal it does, and also w hy that appeal
is to som e d egree genuine. That is to say, the prototype activation
m odel successfully reduces/explains the m ajor features of its com peti
tors. It is also free from m ost of their celebrated defects. I will try to
illustrate these claims as w e proceed.
The p rototyp e activation m odel is focused first and forem ost on
w h at it is to have explanatory understanding of a problem atic thing,
event, or state of affairs. The linguistic exp ression , exch an ge, or p ro
duction of such understanding, should there be an y, is an entirely
secon d ary m atter. I shall approach the topic w ith the aims of an
em pirical scientist rather than w ith the aim s of a logician or con cep
tual analyst. The goal is to outline a substantive em pirical theory of
w hat explanatory understanding really is, rath er than to provide an
analysis of the concept of explanation as it currently is used or ideally
should be used. N orm ative issues will certainly be add ressed , and
som e u n exp ected insights will em erge, but norm ative con cern s are
not the prim e focus of the present chapter. R ather, w hat con cern s us
is the nature of the cognitive process that takes place inside the brain
of the creature for w hom explanatory u n d erstanding suddenly
d aw n s, and in w hom it is occasionally reactivated.
O n the prototype-activation (PA) m odel, a close approxim ation to
this p rocess is the process of perceptual recognition, as w h en one
suddenly recognizes an indistinct outline as the face of a close friend,
or as w h en one finally recognizes the faint m otion un d er the hed ge as
a foraging m ouse. O n the PA m odel, essentially the sam e kind of
com putational achievem ent underlies both perceptual recognition
and explanatory understanding. The latter is distinguished prim arily
by being a response to a w ider variety of cognitive situations: it is not
limited to sen sory inputs.
A close connection betw een perception and explanation is by now a
familiar them e in both psychology and philosophy. O ne's perceptual
jud gm en ts, and p erhaps even o n e's perceptual experiences th em
selves, have often been portrayed as the p erceiver's best explanatory
accou nt of the peripheral stimuli (G regory 1966, 1970a; Rock 1983). In
Explanation: A PDP Approach 199
this tradition the notion of explanation is used in hopes of explicating

the p h en om en o n of perception. The strategy of the p resen t chapter
will reverse the order of things som ew hat: I shall exploit a novel PDP
accou n t of perceptual recognition in hopes of explicating the
p h en om en o n of explanatory understanding. I rem ain faithful to
the earlier tradition, h ow ever, since m y basic aim will be to show that
both p h en om en a are fundam entally the sam e.
Let m e open the discussion by trying to m otivate the search for a
new accou n t of explanation, and for trying to launch it in the specific
directions indicated. W e m ay begin by recalling the covering-law or
deductive-nom ological (D-N) m odel (H em pel 1965), since so m any of
the cu rren t m odels are just artful restrictions or generalizations of that
basic and very elegant idea. W hile m uch attention has b een paid to
the logical virtues and vices of this m odel, relatively little h as been
paid to its shortcom ings w h en evaluated from a psychological point of
view .
In fact, the D -N m odel is psychologically unrealistic in several im
p ortan t w ays. If som eone has just com e to und erstan d w h y a is F, the
D -N m od el typically requires that w e ascribe to that p erson know l
edge of som e universally quantified conditional statem en t having
Fx as its consequent, plus know ledge of a series of initial conditions
adequate to discharge the conjuncts in the an teced en t of that co n
ditional, plus the successful deduction of Fa from this assem bled
inform ation, or at least the appreciation that the deductive relation
exists.
H ow ever, while people have an explanatory understan din g of
m u ch of w hat goes on around them on a m inute-by-m inute and even
a secon d -b y-secon d basis, people are decidedly and regularly inar
ticulate w h en asked to voice either the general law on w hich their
u n d erstandin g is presum ably based, or the set of initial conditions
that tie that law to the explanandum then at issue. W h at prem ises
people are typically able to supply, w hen queried, often falls dram ati
cally sh ort of the full requirem ents of the D -N m odel. M oreover, the
logical acu m en ascribed to people on this accou n t is often substantial
ly in excess of w hat university students w ith form al training in logic
can display.
Fu rth erm o re, the assem bly of relevant factual prem ises and the
search for relevant deductive relations is a process that takes time,
often a good deal of tim e. All of this sits poorly w ith the great speed
w ith w hich explanatory understanding is com m only achieved. It is
often achieved alm ost instantaneously, as w hen one u n d erstan d s at a
glance w h y one end of the kitchen is filled w ith sm oke: the toast is
burning! Such sw iftness is n ot confined to m un d an e cases. If one has
the relevant conceptual skills, the sam e speed is also displayed in

m ore esoteric cases, as w hen one appreciates at a glance w hy Jupiter
is an oblate spheroid: it is a plastic object spinning rapidly; or as w hen
one appreciates at a glance w hy som e red giant star in a close binary
system has the shape of an egg pointed at its m ore com p act blue
com p an ion : it is a very large object free-falling in a gravitational field.
A t the other end of the sp ectrum , n on h u m an anim als provide a
further illustration of these difficulties. A nim als too display behavior
that indicates the achievem ent of explanatory u n d erstanding, as
w h en a frustrated coyote bites and p aw s at the leg trap w hose jaw s
h ave captured its m ate. The coyote u n derstan ds w hy its m ate cannot
leave. A nim als too can anticipate elem ents of the future and u n d er
stand elem ents of the p resent and past, often in som e detail. But the
assem bly of discursive prem ises and the execution of formal infer
en ces is presum ably beyond their capacities, especially at the speeds
that faithfulness to their insight and behavior requires.
These particular criticisms of the D -N m odel are unusual in being
em pirical and psychological rather than logical in ch aracter. E ven so,
th ey are highly general. They will apply to all of the accounts of ex
p lanation that require, as the original D -N m odel requires, extensive
propositional know ledge, relevant retrieval of sam e, and keen d e
ductive insight. For it is precisely these features that give rise to
the difficulties. Is there som e alternative w ay of characterizing the
w ay know ledge is rep resented in living creatu res, and the w ay it is
deployed or accessed in specific cases to provide local explanatory
u nd erstanding? Y es, there is.
2 Conceptual O rganization in PD P N etw orks
R ecent y ears have seen an upsw ell of research into the functional
p roperties of "n e u ra l" netw orks. These are artificial netw orks that
sim ulate or m odel certain salient features of the neuronal organiza
tion of the brain. W hat is interesting is that even simple versions of
these netw orks have show n them selves capable of som e very striking
com putational achievem ents, and they perform these com putations
in a fashion that recalls im portant features of anim al cognition.
A prim ary feature of neuronal organization is abstractly depicted in
the "n e u ra l" netw ork of figure 10.1a. The circles in the bottom row of
the n etw ork rep resent a population of sen sory n eurons. Each of these
units p rojects a proprietary axonal fiber tow ard a second population
of neuronlike units. The axon there divides into a fan of term inal
b ran ch es, so as to m ake a synaptic connection w ith every unit in the
secon d population. Real cells are not quite so exhaustive in their
Figure 10.1
a) A simple network, b) Learning: gradient descent in weight/error space. Axes are
shown for only 2 of 24 synaptic weights
connectivity, but a typical axon can m ake m an y thou san d s or even

h u n d red s of thousan d s of connections.
This arran gem en t allows any unit at the input layer to have an
im pact on the activation levels of all of the units at the second or
"h id d e n " layer. A n input stim ulus p rodu ces som e activation level
in a given input unit, w hich then conveys a signal of proportional
strength along its axon and out the end branches to the m an y syn ap
tic connections onto the hidden units. These connections stim ulate or
inhibit the hidden units as a function of the strength of the signal, the
size or "w e ig h t" of each synaptic connection, and its polarity. A
given h id den unit simply sum s the effects incident from its m any
input syn ap ses. The global effect is that a pattern o f activations across
the set of input units produ ces a distinct pattern o f activations across
the set of hidden units. W hich pattern gets p rodu ced , for a given
input, is strictly determ ined by the configuration of synaptic w eights
m eetin g the hidden units.
The units in the second layer project in turn to a third population of

units, there to m ake another set of synaptic connections. In real
brains this pattern is typically iterated through m any layers (roughly,
5 < n < 50) before the chain concludes in a population of m otor or
other outpu t neurons, but for p u rp oses of illustration, a netw ork of
just three layers will suffice.
In this upp er half too, the global effect is that an activation pattern
across the hidden units p roduces a distinct activation p attern across
the outp u t units. A s before, exactly w hat p attern -to-p attern tran s
form ation takes place is fixed by the configuration of synaptic w eights
m eeting the outpu t units. All told, this netw ork is a device for tran s
form ing any one of a great m any possible input vectors (i.e ., activa
tion patterns) into a uniquely corresponding outp u t vector. It is a
device for com puting a specific function, and w hich function it
co m p utes is fixed by the global configuration of its synaptic w eights.
N ow for the payoff. There are various p roced u res for adjusting the
w eights so as to yield a netw ork that com p u tes alm ost any function—
that is, any vector-to-vector transform ation— that w e m ight desire. In
fact, w e can even im pose on it a function w e are unable to specify, so
long as w e can supply a m odestly large set of exam ples of the desired
in p ut-ou tp u t pairs. This process is called "train in g up the n e tw o rk ."
It typically proceeds by entering a sam ple input vector at the low est
layer, noting the erron eous vector this p rodu ces at the top m ost layer,
calculating the difference betw een this actual outp u t and the desired
outp u t, and then feeding this error m easure into a special rule, the
generalized delta rule (Rum elhart et al. 1986a). That rule then m akes a
small adjustm ent in the an tecedent configuration of all of the synaptic
w eights in the netw ork. Repeating this proced u re m any tim es, over
the m an y exam ples, forces the netw ork to slide d ow n an error
gradient in the abstract space that represents its possible synaptic
w eights (see figure 10.1b). The adjustm ents continue until the n et
w ork has finally assum ed a configuration of w eights that does yield
the appropriate outputs for all of the inputs in the training set.
To give a real exam ple, suppose w e w ant the netw ork to discrim i
nate sonar echoes return ed from large metallic objects, such as explo
sive m ines, from sonar echoes retu rn ed from large subm arine rocks.
The discrim ination of such echoes p oses a serious problem because
they are effectively indistinguishable by the h um an ear, and they
v ary w idely in character even w ithin each class. W e begin by record
ing fifty different mine echoes and fifty different rock echoes, a fair
sam ple of each. W e then digitize each echo w ith a frequency analyz
er, and feed the resulting vector into the bank of input units (figure
10.2a). W e w ant the output units to respond w ith appropriate activa-
Figure 10.2
a) Perceptual recognition with a large network, b) Learned partition on hidden-unit
activation-vector space. Axes are shown for only three of seven hidden-unit activation
levels
tion levels (specifically, <1, 0> for a m ine, < 0, 1 > for a rock) w h en fed
an echo of either kind.
The n etw ork 's initial verdicts are chaotic, since its synaptic w eights
w ere set at rand om values. But und er the p ressure of the w eight-
nudging algorithm , it gradually learns to m ake the desired distinc
tion. A nd after it has m astered the echoes in the training set, it will
generalize: it will reliably identify mine and rock echoes from outside
its training set, echoes it has never h eard before. M ine echoes, it turns
out, are indeed united by som e subtle w eave of features, to w hich
w eave the netw ork has becom e tuned during the training process.
The sam e is true for rock ech oes. (See G orm an and Sejnowski 1988.)
H ere w e have a binary discrim ination betw een a pair of diffuse and
very hard-to-define acoustic properties. Indeed, zve never did define
them ! It is the netw ork that has generated an appropriate internal
characterization of each type of sound, fueled only by exam ples. If w e
now exam ine the behavior of the hidden units during discrim inatory
acts in the trained netw ork, w e discover that the training process has
partitioned the space of possible activation v ectors across the hidden
units. (See figure 10.2b. N ote that this space is not the space of figure
10.1b. Figure 10.1b depicts the space of possible synaptic w eights.
Figure 10.2b depicts the space of possible activation vectors across the
m iddle layer.) The training process has g enerated a sim ilarity gradient
th at culm inates in tw o "h o t sp o ts "— tw o rou gh regions that rep re
sent the range of hidden-unit vector codings for a prototypical m ine
and a prototypical rock. The job of the top half of the netw ork is then
just the relatively simple one of discrim inating these tw o subvolum es
of th at vector space.
Several features of such netw orks beg em phasis. First, the output
verdict for any input is produced very quickly, for the com putation
occu rs in parallel. The global com putation at each layer of units is
distributed am on g m any sim ultaneously active processing elem ents:
the w eighted synapses and the sum m ative cell bodies. H ence the ex
p ression 'parallel distributed p rocessing'. M ost striking, the speed
of processing is entirely independent of b oth the num ber of units
involved and the com plexity of the function execu ted . Speed is d eter
m ined solely by the num ber of distinct layers in the netw ork. This
m akes for very swift processing indeed. In a living brain, w here a
typical inform ation-processing p ath w ay has som ething betw een five
and fifty layers, and each pass throu gh that hierarchy takes som e
thing betw een ten and tw enty m illiseconds p er layer, w e are looking
at overall processing tim es, even for com plex recognitional problem s,
of betw een one-tw entieth of a second and one second. Em pirically,
this is the right range for living creatures.
Second, such netw orks are functionally persistent. They degrade

gracefully u n d er the scattered destruction of synapses or units. Since
each syn ap se supplies such a small part of any com putation, its
dem ise leaves the netw ork essentially u n ch an ged . Third, and very
im p ortan t for our p u rp oses, the netw ork will regularly render
correct verdicts given only a d egraded version or a sm allish p art of
a familiar input vector. This is because the d egraded or partial vector
is relevantly sim ilar to a prototypical input, and the internal coding
strategy generated in the course of training is exquisitely sensitive to
such similarities am ong possible inputs.
A nd exactly w hich similarities are those? They are w hichever simi
larities m eet the joint condition that they unite som e significant p or
tion of th e exam ples in the training set, and the netw ork m an aged to
becom e tuned to them in the course of training. The point is that
there are often m an y overlapping dim ensions of similarity being indi
vidually m onitored by the trained netw ork: individually th ey m ay be
m o d est in their effects, but if several are detected togeth er their im
p act can be decisive. H ere w e m ay recall W ittgenstein's description of
how h u m an s can learn, by ostension, to detect "fam ily resem
b lan ces" that defy easy definition. PDP netw orks recreate exactly this
p h en om en on.
Finally, such netw orks can learn functions far m ore com plex than
the one illustrated, and m ake discrim inations far beyond the binary
exam ple p ortrayed . In the course of learning to p ronou n ce English
text, R osenberg and Sejnowski's N ETtalk (1987) partitioned its
hidden-unit vector space into fully 79 subspaces, one for each of the
79 letter-to-p h onem e transform ations that characterize th e phonetic
significance of English spelling. Since there are 79 distinct ph on em es
in English speech , but only 26 letters in the alphabet, each letter clear
ly adm its of several different phonetic interpretations, the correct one
being d eterm ined by context. D espite this am biguity, the netw ork
also learned to d etect w hich of several possible transform s is the
appropriate one, by being sensitive to the various letters that flank
the target letter inside the w ord. All of this is a notoriously irregular
m atter for English spelling, b ut the "ru le s" w ere learned by the n et
w ork even so.
O th er netw orks have learned to recognize the com plex configura
tion an d orientation of curved surfaces, given only gray-scale pictures
of those surfaces as input. That is, they solve a version of the classic
sh ape-from -shading problem in visual psychology. Still oth ers learn
to divine the gram m atical elem ents of sen ten ces fed as inp u t, or to
predict the m olecular folding of proteins given am ino-acid sequences
as input. These netw orks perform their surprising feats of learned
categorization and perceptual recognition w ith only the sm allest of

"n e u ro n a l" resou rces— usually m u ch less than 103 units. This is
only one hundred millionth of the resou rces available in the h um an
brain. W ith such pow erful cognitive effects being displayed in such
m o d est artificial m odels, it is plausible th at they rep resent a m ajor
insight into the functional significance of our ow n brain's m icrostru c
ture. That, in any case, is the assum ption on w hich the following
discussion will proceed.
3 R ecognition and U nderstanding
L et m e now try to highlight those functional features of PDP n et

w orks that will lead us back tow ard the topic of explanation. The first
feature I w ant to em phasize is the partitioning, in a suitably trained
netw ork , of its hidden-unit activation-vector space into a system of
p rototyp e representations, one for each of the general categories to
w hich the netw ork has been trained (see again figure 10.2b for the
sim plest sort of case). A n y prototype rep resentation is in fact a speci
fic vector (that is, a p attern of activations) across the netw ork's hid
d en units, but w e m ay conceive of it m ore graphically as a specific
point or small volum e in an abstract state space of possible activation
v ectors, since that portrayal highlights its geom etrical relations with
rep resentations of distinct p rototyp es, and with activation v ectors
th at are variously close to (that is, similar to) the prototype vector.
The second point to em phasize is that a single prototypical point or
activation vector across the hidden units rep resents a wide range of
quite different possible sensory activation p attern s at the input layer:
it rep resents the extended family of relevant (but individually
p erh ap s nonn ecessary) features that collectively unite the relevant
class of stimuli into a single kind. A n y m em ber of that diverse class of
stimuli will activate the entire p rototyp e v ector at the hidden units.
Also, any input-layer stim ulus that is relevantly sim ilar to the m em
bers of that class, in part or in w hole, will activate a vector at the
hidden units that is fairly close, in state sp ace, to the prototype
vector.
In dynam ical term s, the prototype position is called an "a ttra c to r."
W e m ay think here of a w ide-m outhed funnel that will draw a broad
but delicately related range of cases into a single narrow path. This
p rocess is instanced in your ability to recognize a friend's face in any
of a w ide variety of expressions, positions, and conditions of view ing.
O r in y ou r ability to recognize a horse in alm ost any p osture and from
alm ost any perspective. These are exactly the sorts of capabilities dis
played by suitably trained PDP netw orks.
Explanation: A PDP A pproach 207
A third point is to em phasize again that PDP netw orks are

extraordinarily fast. O nce trained, they achieve the "reco g n itio n s" at
issue in a m atter of m illiseconds. A nd they will m ake distinct recogni
tions, one after an other, as fast as you can feed them appropriately
distinct stimuli.
Turn now to the units at the output layer. In the stick-figure
accou n t of cognition I am trying to outline, these are to be conceived
as driving or initiating som e specific m otor activity: p erh ap s som e
thing relatively sim ple, as in N ETtalk, w here the output vector codes
a p h on em e and actually produ ces, via a speech synthesizer, an audi
ble sou n d . In a living creature, h ow ever, the output will typically be
m ore com plex, as w h en a dog's sudden olfactory recognition of a
gopher initiates a routine of rooting and digging at the favored loca
tion; or as w hen a bird's sudden visual recognition of a stalking cat
p rom p ts it to initiate a sequence of w ing m otions that launch it into
the air.
The picture I am trying to evoke, of the cognitive lives of simple
creatu res, ascribes to them an organized "lib rary" of internal repre
sentations of various prototypical perceptual situations, situations to
w hich prototypical behaviors are the com p u ted outp u t of the well-
trained netw ork . The prototypical situations include feeding opp ortu
nities, groom ing dem an d s, territorial defense, pred ator avoidance,
m ating opportunities, offspring d em ands, and oth er similarly basic
situations, to each of w hich a certain broad class of behaviors is
ap propriate. A nd within the various generic prototype rep resen ta
tions at the appropriate level of hidden units, there will be subdivi
sions into m ore specific subprototypes w hose activation p rom pts
highly specific versions of the generic form of behavior. (Is the m ouse
eating sesam e seeds? O r hickory nuts? Is it avoiding a cat? O r a
haw k?) These various p rototypes are both united and distinguished
by their relative positions in the hidden-unit vector space. They are all
close tog eth er, but they differ slightly in their positions along one
or m ore of the relevant axes. These differences evoke relevantly
different respon ses at the output layer.
A s just outlined, this picture will inevitably recall m em ories of be
haviorism , for the perceptual environm ent is here portrayed as the
fundam ental control of m otor behavior, and the link b etw een the two
will ap p ear very stim ulus-responsish to m any eyes. But th at construal
of the basic character of the m echanism s at issue is oversim ple and
deeply m isleading, as I shall try briefly to explain.
The sim ple netw orks of figures 10.1a and 10.2a fail to p ortray som e
obvious augm entations that living brains clearly use. Real brains
b oast m an y successive layers of hidden units rather than just o ne—
p erh ap s as m any as a hundred layers along som e pathw ays. Furth er,
real brains divide into m any distinct processing hierarchies w orking
side by side on different problem s. A brain is not a single netw ork,
but a com m ittee of m any cooperating netw orks— p erhaps over a
thou san d of them in a typical m am m alian brain. A nd m ost im portant
for the p resen t issue, the input to a given bank of hidden units com es
not ju s t from the sensory periphery, but from elsew here in the brain
itself. The brain is a recurrent netw ork. The all-up input to any layer
will alm ost alw ays include som e "cu rren t co n te x t" inform ation that
derives from earlier processing elsew here in the brain.
Unlike the m odels in the figures, alm ost any layer of units in the
brain will receive som e "d escen d in g " axonal projections returning
from the next layer of units u p w ard in its hierarchy, and it will often
receive "h o rizo n tal" projections from layers w ithin the m any adja
cen t netw orks w orking on related inform ational tasks. A ccordingly,
w hich p rototyp e vectors get activated w ithin a given layer is generally
a function of diverse inputs, som e of w hich arrive unvarnished from
the sen sory periphery, but m any of w hich reflect con cu rren t activity
elsew here in the brain.
This frees the brain from the knee-jerk style of operation that
w orried us a few m om en ts ago. Its ultim ate behavior is a function
of factors so m any and so subtle, factors that interact in such highly
volatile w ays, that the brain's behavior has becom e predictable only
in its broadest outlines and only for very sh ort periods into the future.
M oreover, the factors controlling behavior reside w ithin the brain
itself as m uch as in the external environm ent.
To retu rn to the basic issue, w e can now see how the brain can
com m an d a large and sophisticated repertoire of prototype activation
vectors, each one representing som e com plex prototypical situation
in the external w orld. W e have seen how such vectors can be
activated by the perceptual apprehension of even a small portion of
the relevant external situation, and how those vectors can activate in
turn behaviors appropriate to the entire external situation, and n ot to
just the small part that w as initially coded in perception.
I w ish to suggest that those p rototype vectors, w hen activated, con
stitute the creatu re's recognition and con cu rren t understanding of its
objective situation, an understanding that is reflected in the creatu re's
subsequent behavior. Of course, a creature m ay fa il to recognize/
u n d erstan d its current perceptual or cognitive situation. The vector
activated at the relevant layer of hidden units m ay fall well outside
any of the prototypical volum es of the relevant state space, and the
behavior subsequently p rodu ced will therefore not be draw n from its
w ell-honed repertoire. The resulting behavior m ay be just confused.
O r it m ay be a default routine of flight from the unknow n. O r p erhaps

it will be a default routine of stumbling exploration, one that m ay final
ly find either a physical or a cognitive perspective from w hich the
situation suddenly does activate one of the creatu re's m any prototype
vectors. It m ay find, that is, a perspective from w hich the situation
suddenly does m ake sense.
By w ay of w h atever learning algorithm governs synaptic adjust
m en ts, such failures and subsequent su ccesses, especially in quanti
ty, will m odify the character and state-space location of the creatu re's
internal p rototyp e representations, so that situations of the puzzling
kind just solved will successfully activate a prototype v ector m ore
readily in future.
4 P rototype A ctivation: A U nified Theory o f Explanation
The aim of the preceding sections w as to illustrate the initial plausibil

ity of a novel conception of cognitive activity, a conception in w hich
vector coding and vector-to-vector transform ation constitute the basic
form s of rep resentation and com putation, rath er than sentential
stru ctures and inferences m ade according to structure-sensitive rules.
Let us assu m e, for the sake of argum ent, that this conception is basi
cally accu rate even for h um an brains. If so, then w e m ust im m ediate
ly be im pressed by the range of conceptual resou rces such system s
can com m an d , given the neuronal m achinery available.
W ith roughly 1011 non sensory n eurons, the h um an brain com
m ands a global state space of fully 1011 dim ensions. E ach brain
subsystem will typically be operating w ith som ething like one-
thou sand th of that num ber, w hich gives a typical specialized state
space approxim ately 108 proprietary dim ensions to play w ith. This
will allow for som e stunningly com plex and fine-grained rep resen ta
tions, since a single vector w ith 108 elem ents can code the co n ten ts of
an entire book. A state space of 108 dim ensions will also allow for a
similarly stunning variety of coding vectors. If w e assum e that each
neu ron adm its of only 10 distinct levels of activation (a serious un d er
estim ation), then that typical specialized state space m u st have at
least 10108 or Ю100-000-000 functionally distinct positions w ithin it. This
is the num ber of distinct possible activation vectors. To appreciate the
m agnitude of this num ber, recall that the total num ber of elem entary
particles in the entire physical universe, photons included, is only
about 1087. A nd recall that, on the above assum ptions, y o u r brain
com m and s som ething like a thousand of these specialized state
spaces.
Of cou rse, the vectors them selves rep resen t nothing, save in the
context of the global configuration of synaptic w eights that p rodu ced
them , and w hich dictate their effects on subsequent layers of
n eu ron s. H ere again w e have a superastronom ical range of possibili
ties. In a given subsystem of 108 n eurons, a typical neuron will have
at least 103 synaptic connections arriving from other n eurons, for a
total of 1011 distinct synapses w ithin that subsystem . If each synapse
adm its of only 10 distinct w eights, then w e have 101011 or
20ioo-ooo-ooo,ooo distinct possible configurations of w eights for that sub
system alone.
O verall, and crudely speaking, this m ean s that a typical subsystem
will have a lexicon of 10108 possible representation s, each of w hich has
108 elem ents, and each of w hich could have any one of 101011 possible
"m e a n in g s ." W e should not balk, therefore, at the prem ise of the
following discussion, which regards it as unproblem atic that the brain
should com m an d intricate prototype representations of such things
as stellar collapse, cell m eiosis, p ositron-positron collision, redox
reaction, gravitational lens, oceanic rift, harm onic oscillator, inten
tional action, and econom ic depression. Such p h en om en a, intricate
thou gh they are, are not beyond reach of the representational re
sources described.
The discussion to this point has all been pream ble to the following
suggestion: Explanatory understanding consists in the activation of a
specific prototyp e vector in a w ell-trained netw ork. It consists in the
ap prehension of the problem atic case as an instance of a general type,
a type f o r w hich the creature has a detailed and w ell-inform ed representation.
Such a rep resentation allows the creature to anticipate aspects of the
case so far unperceived, and to deploy practical techniques appropri
ate to the case at han d (see figure 10.3, a and b). Given the preceding
discussion, this idea has som e plausibility already. It is m y aim in the
rem aind er of this ch apter to illustrate how m u ch illumination and
unity this suggestion can bring to a w ide range of cognitive p h e
n om ena.
L et m e open m y exposition by responding to a possible objection,
w hich will allow m e to em phasize an im portant feature of the v ecto
rial representations here at issue. "W h a t you have o u tlin ed ," runs the
objection, "m a y be a successful accou n t of sp ontaneous classification,
but explanatory understanding surely involves a great deal m ore than
m ere classification."
The objection evokes a process of m ere "lab elin g ," a process that
pu ts the appreh en ded situation into a featureless pigeonhole, a
process in w hich m ost of the com plex inform ation contained in the
input is lost in its reduction to a canonical neural respon se. Y et this
Figure 10.3
Explanatory understanding as the activation of a prototype vector, a) Ampliative activa
tion of desert rat vector, b) ampliative activation of rotating plastic body vector
is precisely the w ron g w ay to view the p rocess of recognition and the

character of the rep resentation activated.
W hat w e m u st rem em ber is that the p rototyp e vector em bodies an
en orm ou s am ount of inform ation. Its m any elem ents— p erh ap s as
m any as 108 elem ents w as our earlier gu ess— each constitute one
dim ension of a highly intricate portrait of the prototypical situation.
That vector has structure, a great deal of stru ctu re, w hose function
is to rep resent an overall syndrom e of objective features, relations,
sequences, and uniformities. Its activation by a given p erceptual or
other cognitive circum stance does n ot rep resent a loss of inform ation.
O n the contrary, it rep resents a m ajor and speculative gain in in
form ation, since the portrait it em bodies typically goes far b eyond the
local and perspectivally limited inform ation that m ay activate it on
any given occasion. That is w hy the process is useful: it is quite d ra
m atically am pliative. O n each such occasion, the creature ends up
u n derstan din g (or p erhaps m isunderstanding) far m ore about the ex-
planandum situation than w as strictly presen ted in the explanandum
itself. W h at m akes this w elcom e talent of am pliative recognition
possible is the m any and various exam ples the creature has already
en cou n tered , and its successful generation of a unified prototype rep
resentation of them during the course of training.
This view entails that different people m ay have different levels or
degrees of explanatory un d erstanding, even though they classify a
given situation in w hat is extensionally the sam e w ay. The reason is
that the richness of their respective p rototyp e representations m ay
differ substantially. This is a w elcom e consequence, since explanatory
u n derstan din g does indeed com e in d egrees. O n the p resen t view , its
m easure is just the richness and accuracy of the creatu re's prototyp e.
W ith these points in h and, let us now turn to a larger issue. O ne
p rom in ent fact, ill addressed by any existing accou n t of explanation,
is the variety of different types of explanation. W e have causal ex
planations, functional explanations, m oral explanations, derivational
explanations, and so forth. D espite som e p rocru stean analytical
attem p ts, no one of these seem s to be the basic type to w hich all of the
others can be assim ilated. O n the prototype-activation m odel,
h ow ever, w e can unify them all in the following w ay. Explanatory
u n derstan din g is the sam e thing in all of these cases: w hat differs is
the ch aracter of the prototype that is activated.
4.1 Property-cluster prototypes I begin w ith w hat is presum ably the

sim plest, m ost com m on, and m ost superficial kind of explanatory
u n derstanding, and w ith the sim plest and m ost superficial kind of
p rototyp e: the cluster of typically cooccu rrent properties. Think of the
typical cluster of features that constitutes a cat, or a cookie, or a tree,

or a bird. These p rototyp es com p reh end the vast m ajority of on e's
conceptual population, and they are activated on a regular basis in
the cou rse of on e's m undane affairs. Because of their familiarity to
everyon e, cases involving them are seldom puzzling to an yon e. But
the explicit questions of inquiring children reflect the background ex
p lanatory role these p rototyp es continue to play for all of u s. "W h y is
its neck so long, D ad d y ?" "It's a sw an dear; sw ans have v ery long
n e ck s." "W h y is he all spotted, M o m m y?" "H e 's a leopard dear;
leopards are alw ays spotted, except w hen they're y o u n g ."
4 .2 E tiological prototypes These are w hat lie behind causal explana

tions. A n etiological prototype depicts a typical tem poral sequence of
event typ es, such as the cooking of food upon exposure to h eat, the
deform ation of a fragile object during im pact w ith a tou g h er one, the
escape of liquid from a tilted container, and so on. These sequences
contain prototypical elem ents in a prototypical order, and they m ake
possible our explanatory understanding of the tem porally extended
w orld. W e apparently com m and a very large repertoire of such
p rototyp es, m u ch of it organized into an elaborate h ierarch y of
species and subspecies.
N ote that the tem poral inverse of an etiological p rototyp e is gen er
ally not an etiological prototype as well. This m eans th at causal
explanations are generally asym m etric in ch aracter. The height of a
flagpole and the altitude of the sun m ay jointly explain the length
of the p ole's shadow . But the length of the pole's shadow will serve
to explain neither of the other tw o facts. That asym m etry, a m ajor
problem for other accounts of explanation, is a natural consequence
of the p resen t account.
Som e m ay com plain here that I have simply helped m yself to an
unexplicated notion of "cau sal p ro cess" or "etio lo g y " in ord er to sus
tain a particular analysis of explanation, one that takes "c a u s e " as a
prim itive. This charge m isses m y p u rpose. I am not attem pting to
provide an analytic definition of the notion of explanation. I am trying
to provide a revealing and unifying characterization of the sorts
of brain events that constitute explanatory u n d erstanding. Those
events, I suggest, are prototype activations. A n im portant subclass of
activated p rototyp es rep resent typical tem poral sequences or p ro
cesses. These, I observe, appear to underw rite w hat philosophers
have called causal explanations.
N ow just w hat intricacies constitute a genuine etiological p roto
type, and how the brain distinguishes betw een real causal processes
and m ere p seud oprocesses, are secondary m atters I shall leave for
a future occasion. W e have, at a m inim um , a novel set of resou rces

w ith w hich to address that old problem . It is sufficient for p resen t
p u rp oses to observe that, w ithin the fram ew ork of the PA m odel, w e
can im m ediately recon stru ct, at least to a first approxim ation, the
m ajor types of antecedently recognized explanation. To this task I
now return.
4 .3 P ractical prototypes These, I suggest, are w hat lie behind fu n c

tional explanations. O ne thing hum ans u n derstand very well, being
agents in the w orld, is com plex m eans-en d relations betw een possible
situations realizable by us and expectable consequences thereof that
m ay be desirable to us. To portray any tem poral configuration of
actual or potential situations in this m ean s-en d w ay is to m ake
graphic or salient for us certain of the causal relations that unite them .
It is a w ay of portraying a causal structure in the w orld in the guise of
an actual or figurative practical problem , w ith the explanandum event
or feature as its practical solution.
Practical prototyp es, like etiological p rototyp es, also depict se
quences of event types or feature dependencies, but in the case of prac
tical prototypes the explanandum begins the explanatory sequence,
w hereas in etiological p rototypes the explanandum concludes the
explanatory sequence. Thus, a functional explanation does provide
som e entirely objective inform ation, as in "H is peripheral blood v es
sels all con tracted in order to p rotect his central organs from the
gathering co ld ." But som etim es they ring v ery hollow , as in "S u p e r
n ova 1987A collapsed on itself in order to spread photons and n eu
trinos throu ghou t the physical u n iv e rse ." Just w hen functional
explanations are appropriate and w hen they are not is a delicate issue
that I shall bypass on this occasion.
4 .4 Superordinate prototypes Som e explanations, typically those in

science, logic, and m athem atics, con cern n ot singular facts but gen eral
truths. W hy do planets m ove on ellipses? W hy are the theorem s of
the sentential calculus all tautologies? W h y do the interior angles of
any triangle alw ays sum to exactly one straight angle? H ere the ob
jects of p u zzlem ent are not singular situations; they are prototypical
synd rom es them selves. Even so, explanatory understanding seem s
to consist in the apprehension of the subordinate prototype as being
an instance of som e superordinate prototype.
The puzzle about triangles will illustrate the point. W hy is the sum
alw ays the sam e? A nd w h y is it exactly 180°? C onsider the p rototyp i
cal situation depicted in figure 10.4a, nam ely, tw o parallel lines cut by
a third. A prototypical feature of that situation is the equality of the
Figure 10.4
Understanding a subordinate prototype as one instance of a superordinate prototype
angles /3 and /3'. (As it happens, this captures the substance of

Euclid's fam ous parallel p ostulate.) A n oth er depiction of that sam e
p rototyp e appears in figure 10.4b, a feature of w hich is the equality
y = y ' . N ow im agine these tw o figures superim posed, as in figure
10.4c, to form the triangle A B C . The sum a + /3' + y' is evidently a
straight angle. A nd given the tw o equalities noted earlier, it is evident
that a + (3 + y m ust equal the very sam e value. This prototypical fea
ture of triangles can now be seen as just another instance of a yet
m ore general prototype. O nce the latter has been relevantly acti
vated , as in figure 10.4c, the form er is no longer puzzling.
Explanations of this subordinate-superordinate kind are typically
displayed in intertheoretic reductions, w here one theory is subsum ed
in w hole or in p art by a m ore general theory. They are also displayed
w hen o u r scattered understanding in som e dom ain is successfully
axiom atized, w hich on the p resen t view is just an oth er instance of the
sam e process.
A xiom atizations, in this cen tu ry's experience, are ruthlessly syn tac
tic affairs. The D -N m odel w as right at h om e in accounting for ex
p lanations w ithin that general context. But axiom atization w as not
alw ays thu s. I have used an exam ple from geom etry here in order to
suggest that the successful system atization of theoretical u n d erstan d
ing can be pictorial or graphical, as well as occasionally syntactic. A s
w ith m an y others, I acquired m y conception of axiom atization as a
student of formal logic and of the philosophy of science and
m ath em atics in the mid sixties. W ithin that tradition, axiom atization
w as essentially a syntactic affair. W hen subsequently I looked through
Euclid's original Elem ents— acclaim ed as the first great axiom atization
in h istory— I w as therefore struck by how very ««syn tactic and how
vividly diagram m atic the w hole affair w as.
This w as standardly p ut dow n to the sloppiness of Euclid's p re
sentation. I can no longer entirely believe that. Though quite real,
syn tactic system atization now appears as only one of several m odes
of cognitive system atization, p erhaps of relatively m inor im portance.
The basic m ode of system atization is the reconfiguration of the simi
larity m etric across the relevant v ector space, so that previously di
verse and unrelated cases are now rerep resen ted as being relevantly
similar to each other, as being slightly different instances of the sam e
m o re general p rototype. This reconfiguration can happen in tw o
w ay s: w ithin the space that contains the relevant prototyp es, so that
th ey com e to occupy closely proxim ate positions; or in the next layer
u p , so that they are seen as instances of a higher prototype located in
a subsequent layer of hidden units.
This m ore encom passing view of conceptual unification is of course
entirely com patible w ith the idea of syntactic unification through ax
iom atization. It m erely subsum es that m ode of organization und er a
m o re general m ode. The presen ted "a x io m s" of formal system s, you
will recall, are alm ost alw ays axiom schem ata, having a potential infin
ity of distinct but recognizable instances. A nd the rules of inference
are inference schem ata as well, w hose diverse applications m u st be
m astered by the person using them . Thus do w e learn to discrim inate
proofs. W h at unites the class of theorem s is that each is the last line of
som e proof or other. In this w ay, evidently, can prototype-trained
PD P netw orks handle the com binatorial and generative asp ects of
cognition. Their capacity for discrim inating relevant structure and
principled iteration m ay be limited, of course. But then, so is ours (see
Bechtel and A braham son, forthcom ing).
There is a great deal m ore to be said about explanatory unification,
b u t for now let us m ove on.
4 .5 Social-interaction prototypes These underw rite ethical, legal, and

social-etiqn ette explanations, a class of explanations u n ad dressed , to
m y know ledge, by any prior accou n t of explanation. This is faintly
surprising, since they are both real and familiar. "W h y shouldn't
I disperse this bunch of troublem akers?" asks the redneck cop
ap p roach in g a so-far peaceful dem onstration. "B ecau se that w ould be
a case of violating their constitutional right to peaceful assem bly" is
an explanatory reply. "W h y shouldn't I discuss M ary's marital prob
lem s w ith D oris?" asks the gossip. "B ecau se that w ould be violating a
con fiden ce" is an explanatory reply. "A w , w h y can 't I go play outside
in the rain ?" asks the seven-year-old boy. "B ecau se you h ave your
new clothes on, and com pany will be here for Thanksgiving dinner in
half an h o u r" is an explanatory reply. H ere w e appeal to various p ro
totypical m isbehaviors— denying a constitutional right, b etraying a
confidence, being inappropriately turned out for a family fete— of
w hich the contem plated behavior is claimed to be an instance. A c
quiring a legal, m oral, or social sensibility is a m atter of m asterin g a
large system of such social-interaction p rototyp es, and of com ing to
perceive the social w orld from w ithin that fram ew ork. (Since w riting
this p ap er, I have learned that m oral explanations have indeed been
recognized and discussed. Sturgeon [1985], for exam ple, assim ilates
them to the D -N p attern .)
4 .6 M otivational prototypes These underw rite our familiar folk-

psychological explanations of intentional actions. By rights, such ex
planations should not be assigned a sep arate category in the p resent
list, since m otivational p rototypes are properly just a special subclass
of the etiological prototyp es discussed in section 4 .2 above. They are,
I believe, causal explanations (chapters 1 and 6). But I did n ot wish to
break the flow of exposition in section 4 .2 by addressing a disputed
case. O n the other h and, I do not wish to leave this im p ortan t class
entirely unm entioned.
M otivational p rototyp es depict typical configurations of desires,
beliefs, and preferences as the anteced ents of intelligent behavior.
The explanations they underw rite are distinct from the functional ex
planations described in section 4 .3 , since the elem ents in the causal
sequence of a functional explanation are typically external to any
agent's b ody and tem porally later than the action to be explained.
W ith action explanations, by contrast, the initial elem ents in the
explanatory sequence are alw ays internal to the agent— his desires,
beliefs, and deliberative activities— and they are alw ays prior to
the behavior to be explained.
M otivational prototyp es are but a subclass of a large family of affec
tive, deliberative, and cognitive prototyp es, those that collectively
m ake up folk p sychology. A perennial objection to the idea that our
self conception has the character of a theory has been its apparent
requirem ent that each of us com m and m any thou san d s of universally
quantified conditionals— the "la w s" of folk p sychology. From the
PD P persp ective, this is no longer a requirem ent. A s a picture is
w orth a thou sand w ords, so a few hundred com plex p rototyp es can
do the w ork of m an y thousands of law s. A nd unlike law s, a p ro
totype is activated directly by the app rehended situation it m ay serve
to explain. The access problem largely evaporates. To be sure, in
appropriate prototyp es will often be activated, conflicts w ith further

experience will thus ensue, and the puzzling explanandum will then
have to be read dressed from within a m ore inform ed con text. But the
p rototype vectors, w heth er appropriate or inappropriate, will typical
ly be activated w ithin a second or so.
The up sh ot for the dispute about folk psychology is that it is still a
theory, just like any other theory. W hat has chan ged , and changed
for the better, is our conception of how theoretical know ledge is both
stored and accessed in cognitive creatures.
T hough I very m uch doubt it is com plete, I shall now bring this
catalog of prom inent kinds of prototyp es to a close. You can see how
they allow the prototype-activation m odel to account in a unified
fashion for the m ost familiar and widely discussed types of exp lan a
tory un d erstanding, and for som e previously undiscussed types as
well.
5 Inference to the Best Explanation
The idea of prototype activation throw s som e m u ch-needed light on

the popular idea of "in feren ce to the best exp lan ation ," a process that
h as often been invoked to accoun t for the fixation of m any of our
beliefs, even our observational beliefs (see, for exam ple, H arm an
1 9 6 5 ,1 9 7 3 ). That idea is appealing, since it does seem to address w hat
distinguishes the beliefs w e do acquire from the m any w e m ight have
acquired: the form er have better explanatory p ow er relative to the
overall circum stance that occasioned them .
But the idea is also problem atic, since it suggests a choice m ade
from a range of considered alternatives. A s a m atter of psychological
fact, alternatives are rarely even present. A nd in any case, our beliefs
are typically fixed so swiftly that there is no time for the com parative
evaluation of com plex m atters such as the relative explanatory p ow er
of each of a range of alternatives.
O n the PD P approach, w e can begin to explicate the crude notion of
"inferen ce to the best explan ation" with the m ore penetrating notion
of "activation of the m ost appropriate p rototyp e v e cto r." A ctivating
the m ost appropriate available prototype is w hat a w ell-trained n et
w ork does as a m atter of course, and it does it directly, in response to
the input, w ithout canvassing a single alternative vector. In the end,
the p rocess is not one of "in feren ce" at all, n or is its outcom e general
ly a sentence. But the process is certainly real. It just needs to be
reconceived w ithin the m ore penetrating fram ew ork of cognitive
neurod yn am ics. W hen it is, both the alternatives problem and the
speed problem disappear.
C. S. Peirce, w ho called the process "ab d u ctio n ," found the form er
p roblem especially puzzling. Peirce, one of the pioneers of m od ern
form al logic, appreciated very early that for any set of observations
there is a literal infinity of possible h ypotheses that m ight be posed in
explanation. But how can w e possibly search a space of infinite size?
Indeed, h ow can w e even begin to search it effectively w h en its ele
m en ts are not well ordered? Peirce m arveled that h u m an scientists
are able so regularly to p roduce, from this infinite sea of m ostly
irrelevant and hopeless possible candidates, hyp oth eses that are both
relevant and stand som e nontrivial chance of being tru e. From the
sentential perspective, Peirce w as right to m arvel. But from the
n eurocom putational perspective, the situation is not so m ysterious.
W e do not search an infinite space of possible explanations. In
general, w e do not search at all: in familiar cases a suitable prototype
is activated directly. A nd if the novelty of the case foils our w aiting
categories and thus forces us into search m ode, then w e search only
the com paratively tiny space com prising the set of our ow n currently
available p rototyp e vectors. Even here the search is m ostly blind and
probably stops at the first su ccess. If on e's initial en cou n ter w ith the
problem atic explanandum fails to activate directly a familiar and sub
sequently successful prototype vector, then one repeatedly reenters
the problem atic input in a variety of different cognitive con texts, in
hopes of finally activating som e prototype vector or oth er, or som e
vector close enough to an existing p rototype to give one at least som e
handle on the problem (see chapter 11).
Since the range of concurrently possible understan din gs is closed
un d er the relation "is at least w ithin hailing distance of an existing
p ro to ty p e ," then o f course any elem ent from that range will appear
both relevan t and potentially true. Peirce, and w e, are the victim s of a
perspectival effect. O ur hyp oth eses will look at least passably hopeful
to us b ecau se they are draw n from a source that collectively defines
w hat will be found plausible by us. W e should thus be w ary of
assum ing, as Peirce seem s to have assu m ed , that w e have any special
nose for truth.
I h ave raised Peirce's w orry here for a further reason : the p rocess
just described illustrates the role of analogy in the d iscovery of new
explanatory h yp oth eses. So m any of our best theories h ave had their
origins in a provocative initial analogy, perhaps strained in som e
dim ensions, that subsequently p roved to be a fruitful vehicle of
u n d erstandin g (see H esse 1966; Kuhn 1962). The p rocess finds a
read y depiction w ithin the p resent fram ew ork. A p rototyp e vector
w hose activation has hitherto been confined to one em pirical dom ain
subsequently com es to be activated w ith profit in a new dom ain.
M ore accurately, the new dom ain activates a vector that is close to the
old p rototyp e, close enough to evoke at least som e of its cognitive and
behavioral consequences. A n analogy betw een tw o dom ains has been
discovered and exploited. (O n this, see also ch apter 11, section 3.)
U nfortunately, talk of analogy has alw ays been hobbled by our
inability to say anything very specific about w hat constitutes the
relevant kind of similarity. W e are now in a position to be entirely
specific. A nalogy, or similarity, consists in the close proxim ity of the
respective prototype representations, activated by the tw o empirical
dom ains, in the relevant hidden-unit activation-vector space. A n d the
constituting dim ensions of that very high-dim ensional similarity are
given by the response properties of each n eu ron in the relevant hid
den layer. This m eans that any analogy is a com plex and highly subtle
m atter, easily recognized, perh ap s, even though it will typically
tran scend effective verbal description. M ost im portant, it is no longer
a m ystery w hy both of these things should be so. The m any dim en
sions of the relevant vector space are used by us, but they are not
generally know n to us.
The title of this section refers to the "b e s t" explanation. But how ,
on this PD P approach to explanation, are com peting m odes of u n d er
standing to be evaluated? W hat m akes one explanation better than
an other? H ere w e m ust answ er carefully, since w e are denied the
usual sem antic vocabulary of reference, truth, consistency, entail-
m ent, and so forth. The cognitive kinem atics here being explored
does n ot have sen ten ces or propositions as its basic elem ents; the
basic elem ents are activation vectors. The various dim ensions of epis
tem ic virtue will therefore have to be reconceived in term s that are
grou n ded in this new conception of w hat cognitive activity consists
in.
That will be no small task, and I cannot pretend seriously to u n d er
take it here. W hat I shall do is illustrate how som e aspects of the
problem can be add ressed , and how the netw ork approach supplies
un exp ected insight into at least som e perennial problem s.
A t the sim plest level, it is easy enough to describe the differential
virtue that tw o distinct prototype vectors A and B can have as a re
sp onse to a problem atic situation. If w e assum e that w e are dealing
w ith a w ell-trained netw ork, and that the integrity of the tw o p ro
totypes is not in question, then it m ay be a just criticism to say that A
is sim ply the w rong p rototyp e for the problem atic situation at h and. It
m ay be w ron g because the situation confronted is not a m em ber of
the class of situations that will reliably activate A from alm ost any
perspective, even though it happ ened to activate A on this occasion.
This can occur if the agent appreh en ded only a m isleading part of
the problem atic situation, a part that led to the activation of A because
that unusual part w as relevantly similar to A 's typical activators. U n
fortunately for the agent, h ow ever, the problem atic situation is such
that, if ad dressed from any one of m any slightly different perceptual
or cognitive points of view and then reen tered into the netw ork, it
w ould reliably activate B in alm ost every case. In this statistical sense,
A m isrepresents the situation, w hereas B does not.
C orrelatively, the behavioral consequences typically activated in
turn by A m ay be highly, even lethally, inappropriate to the prob
lem atic situation in question, in contrast to those activated by B. A
real exam ple of both failures w ould be a co y ote's faulty u n d erstan d
ing of the natu re of a small tapered appendage disappearing into a
tuft of long grass. The hungry coyote und erstan ds it as the tail of a
retreating desert rat and acts accordingly. In fact, the problem atic
item is the tail of a poisonous snake, w hich coyotes generally avoid
(see again figure 1 0.3a).
P ragm atic and statistical considerations can thus provide a fairly
robust distinction betw een appropriate and inappropriate p ro
totypes, at least for simple cases. W e can even recon stru ct the distinc
tion betw een the correctness of a prototype on a given occasion, and its
w arrant on that occasion. H igh w arrant is a m atter of low am biguity in
the input. W e need to ask, Is the input vector closely similar to any
other possible input vector that w ould activate a different prototype?
If so, then the ambiguity of the current input is high and the w arrant
of the p rototype vector activated is correspondingly low. If the input
is not rem otely similar to any activationally diverse input, then its
am biguity is low and the w arran t of its activated p rototyp e is high.
Similarity am ong input vectors is m easured as usual: by their geom et
rical proxim ity w ithin the activational state space for the relevan t in
put layer.
The evaluation of occasion-specific explanations is one thing, but
w hat of evaluating the propriety of an entire class of explanations?
H ow do w e evaluate, that is, the integrity of the p rototyp es, and sys
tem s of prototypes, them selves? It is difficult to provide m u ch of an
an sw er to this question, since the answ er m u st draw on som e general
story of how to evaluate the global cognitive configurations of neural
netw orks. It m ust draw , that is, on som e analog of a general account
of "th eoretical v irtu e."
W e are as yet in no position even to sketch such an accou n t. But it is
clear that there m ay be gold in these hills. The cognitive behavior of
quite sim ple netw orks displays an u n expected connection betw een
representational econom y, conceptual unification, and the capacity
for successfully generalizing past experience to novel cases. Very
briefly, the story is this. Suppose w e wish a netw ork to discrim inate
F s. A nd suppose w e train that netw ork to a high level of success on
som e training set (the rock/m ine netw ork, for exam ple). D espite its
success at recognizing Fs d raw n from the training set, the netw ork
will do very poorly at recognizing new exam ples of F draw n from
outside the training set if the num ber of hidden units in the netw ork
is v ery m u ch larger than a certain optim al num ber. The reason is that,
w ith so m any hidden units to exploit, the lazy netw ork will tend to
learn a distinct and unrelated p rototyp e for each of the distinct and
slightly different exam ples of F in its training set. These unrelated
p rototyp es will sustain appropriate respon ses at the output layer, and
w ith the error m essages thus reduced to nothing, the system will
cease any further learning. In particular, nothing forces it to generate
a sin gle p rototype (at the cen ter of a unitary similarity gradient) as its
respon se to the entire range of relevantly similar exam ples of F in the
training set.
The cost of such laziness com es due w h en the netw ork is p resented
w ith a new exam ple from the relevant class of stimuli, an exam ple
draw n from outside its training set. That exam ple will be similar to all
of the oth ers, but the netw ork is in a relatively poor position to see
th at similarity, since it has not learned to see w hat is com m on to all of
the relevant exam ples in the original training set. To be sure, it will do
b etter than chance at recognizing new exam ples, since it does have
som e scattered grasp of the class at issue. But still, it perform s poorly.
By contrast, a netw ork that is denied a surplus of hidden units, a
n etw ork given too few resou rces to perm it such ad hoc and un p ro-
jectible learning, is a netw ork that is forced to continue learning until
it finds a single p rototyp e region, a region at the cen ter of a unified
similarity gradient, a region near w hich to code every exam ple of F in
its training set. A unified solution to the recognition problem is the
only solution for w hich it has adequate resou rces. After being forced
b y circum stance to find such a solution, the netw ork will subsequent
ly recognize a w ide range of novel exam ples of F alm ost as reliably as
those it w as trained on, since it possesses a similarity gradient ad e
quate to catch alm ost all of them . (For a m ore detailed discussion,
see P. M. C hurchland 1989b or chap ter 9.)
C onceptual unification, evidently, is a cognitive virtue of enorm ous
im portan ce, at least as conceived on the p resen t m odel of cognition. It
is im p ortan t for the very good reason that cognitive configurations
having that virtue do m u ch b etter at generalizing their past experi
ence to new cases. It is therefore reasonable to regard explanations
that involve p rototypes em bedded in a unified cognitive configura
tion as superior to those that involve p rototyp es n ot so em bedded.

H ere is a further dim ension in w hich explanations can be evaluated.
All told, then , w e can evaluate them as follows. A v irtuous m ode of
explanatory understanding (that is, an activated p rototyp e vector)
should be a rich portrait of the general type at issue; it should be
strongly w arranted (that is, have low am biguity in the input that o cca
sions it); it should be correct (relative to the library of currently avail
able alternative p rototypes); and it should be part of the m ost unified
cognitive configuration possible. Evidently, norm ative epistem ology
is not essentially tied to a propositional-attitude conception of cogni
tive activity. It is robustly possible w ithin the fram ew ork here being
explored. A nd on the specific m atter of conceptual u n ity— w hat it
is, and w hy it is a virtue— the new fram ew ork can already claim a
p roprietary success, a success w here the old fram ew ork has been a
chronic failure.
6 Com parison with Earlier M odels
Let us begin w ith the venerable D -N m odel. It w as correct in insisting

that explanatory understanding requires the deploym ent of som e
inform ation that is general in character. Beyond this insight, alm ost
nothing is correct. The m odel's com m itm ent to a sentential or p ro p
ositional m ode of know ledge representation renders it unable to
accou n t for explanatory understanding outside of that n arrow con
text, and it generates a host of problem s even within th at context.
Slow access, inarticulation of law s, and deductive inappreciation
w ere d iscussed at the beginning of this essay. To these w e m ay add
the problem s of explanatory asym m etry (Brom berger 1966; Teller
1974), irrelevant explanation (Salmon 1970), and accidental universals
(E. N agel 1961).
N one of these difficulties attend the prototype-activation m odel.
C on cern ing the m atter of access, relevant un d erstan din g is usually
accessed in m illiseconds. C oncerning our inability to articulate
law s, the PA m odel does not even suggest that w e should be able to
articulate them . For w hat gets accessed is not a stored universal
conditional, but a com plex prototyp e. Similarly, while our ded u c
tive incom peten ce is a problem for the D -N view , on the PA m odel
deductive inference typically plays no role at all in the process of
p rototyp e activation. M oreover, as noted in section 4 , etiological
p rototyp es are in general tem porally asym m etric. Exp lan atory asym
m etries are thus only to be expected. Irrelevant explanations (such as
appealing to "h exed salt" in order to explain a sam ple's dissolving in
w ater) appeal to strictly nonexistent p rototyp es. A nd finally, w hat

m oves us to reject an im potent explanatory prem ise as a m erely
accidental universal is that fact that, despite its truth, it fails to ex
press any of our learned prototypes.
W h at a w ell-turned deductive-nom ological argum ent certainly
can do is successfully evoke explanatory understan din g in the h earer
by provoking activation of the relevant p rototyp e with som e well-
stru ctured and highly salient verbal stimuli. D-N argum ents are
therefore entirely appropriate things to exchange in a great m any
explanatory contexts, especially in com plex contexts. But they do not
em body or accoun t for the understanding itself.
Som e of the attem pts m ade to patch the D -N m odel can be seen as
w ell-m otivated but opaque attem pts to bridge the gap betw een the
cognitive w eakness of universally quantified conditionals and the
cognitive m uscle of genuine learned p rototyp es. In response to som e
of the standard difficulties, Brody (1972) suggested im posing a
"cau sal relation" condition an d /o r an "essen tial featu re" condition in
addition to the usual D -N requirem ents. O n the prototype activation
m odel, the relevant difficulties are handled by som ething rath er simi
lar: the existence within the creature of learned etiological prototypes
and p roperty-clu ster p rototypes. This m ay explain the appeal of
B rody's proposals. H ow ever, these tw o kinds of prototypes consti
tute only tw o of five or six different kinds com m only deployed in
explanation. It is a shortcom ing of B rod y's accou n t that it en co m
passes only these tw o kinds of explanation. Also, his account rem ains
a restricted version of the D -N m odel. It is fundam entally sentential
in ch aracter, and it does not escape the em pirical objections voiced in
section 1.
A n im portant alternative to the D-N m odel is the statistical relevance
or SR m odel (Salm on 1971). P erhaps the m ost salient and appealing
feature of this m odel is its rejection of the requirem ent that the expla
n an d u m be inevitable or even probable in light of the explanans.
R ather, understanding consists in grasping the statistics of the expla
n an d um situation correctly, w h atever those probabilities happen to
be. Thus, it is said, one m ay properly explain m y having the flu by
pointing to m y w eek of nursing a pair of flu-infected children, even
though the probability of the form er on the latter is less than 10 p er
cent in an adult.
Though m any have contrary intuitions, I think this is a w elcom e
m odification. The explanans in such a case does throw light on the
explan an du m : one's understanding of the situation is significantly
im proved, despite the low probability at issue. In m any dom ains,
probabilities far below unity is all that the w orld ever provides. E x
p lan atory understan din g then appears on the thin side, but it does
not disapp ear entirely.
For better or for w orse, the p rototyp e activation m odel seem s com
m itted to the sam e view . A prototype is by definition "g e n e ra l," in
the sam e sense in w hich a property is general: it has m an y instances.
But th at is the only sense in w hich it need be general. A prototype
v ector is not just a cluster of universally quantified conditionals in
neural disguise. A prototype, you will recall, can rep resen t a wide
range of diverse exam ples. There need be no feature that is universal,
or even nearly universal, to all exam ples in the class. The statistical
distribution of relevant features, across the class com p reh en d ed by
the p rototyp e, m ay have alm ost any profile. But w h atev er that dis
tribution is, it will be rep resented in the salience accorded each rel
evant feature in the similarity space configured at the relevant layer
of hidden units. A nd it will likely be reflected in the subsequent
cognitive and m otor behavior that has been learned by the netw ork.
A ctivation of that v ector, nonuniform though its portrait of the class
m ay b e, still rep resents the netw ork's understan din g of the input.
T hough all of this m ay be true, I do not think it accou n ts for m ore
than a small p art of the explanatory appeal of the rem ark "I nursed
tw o flu-infected children for a w eek ," relative to the question "W h y
do you have the flu?" R ather, that explanans has the clout it does
b ecause it activates an etiological p rototyp e concerning the typical
transm ission of viral infections, a prototype w hose w arran t in this
case m ay be extrem ely high, despite the general low probability of
adult illness given exposure to infected children. The w arran t m ay be
high b ecause infection from those tw o children m ay be the only
serious opportunity for infection you encountered, and w e do demand
som e infective etiology for every case of flu. The im portance of causal
stru ctures is also a m ajor them e of Salm on's m ore recent w ork within
the statistical perspective (1978, 1984). From the PD P persp ective, this
rep resents a w elcom e recognition of the im portance of etiological
p rototyp es for explanatory understanding.
O verall, h ow ever, it also rep resents a vision of explanation that is
confined to only the second of the five or six im portant kinds of p ro
totyp es outlined in section 4. C ausal stru ctu res are relevant to som e
kinds of explanations, but certainly not to all: think of explaining
m athem atical facts. A nd despite its sharing som e im portant com m it
m en ts w ith the p rototype activation m odel, the SR m odel rem ains
firmly lodged w ithin the sentential fram ew ork of folk p sychology,
and thu s is heir to all of the empirical and psychological problem s
d iscussed in section 1.
A different and increasingly popular approach to explana
tion abandons hope of finding anything very interesting,
well defined, and unitary about w hat cou n ts as an explanation. Van

Fraassen (1977, 1980) has urged that w h atever red u ces som eon e's
p u zzlem ent can count, in that context, as an explanation, and it is
evident that a considerable diversity of things can be expected to m eet
this vagu e and ultim ately pragm atic condition. H ere our attention is
shifted aw ay from understanding itself and is focused instead on the
large variety of verbal or other acts that m ight, in con text, produ ce it.
It is a consequence of the prototype-activation m odel that the varie
ty of things that can finally prom p t a relevant prototype in a situation
of p u zzlem ent is endlessly various. There is nothing canonical about
inputs that fail to activate a prototype, and there is nothing canonical
about w hat subtle m odifications or augm entations of those inputs
will finally m ake them do so. A one-w ord hint, a salient gesture, an
accidental analogy, alm ost anything, in con text, m ight augm ent the
available input so as finally to trip the activation of som e appropriate
prototyp e. "W h y is everyone so tense around here this m o rn in g ?"
asks the office m an ager of a ran d om em ployee. A rolling of the eyes
and a flicker of a glance tow ard the desk of the office troublem aker
tells him all he n eeds to know .
W ell and good. But endless variety in the im m ediate causal an tece
dents of explanatory understanding is wholly consistent w ith unity in
the basic nature of understanding itself. The case for seeing such uni
ty is detailed throu ghou t this essay. The variety van Fraassen finds in
exp lan atory contexts is entirely real, but I su ggest it properly belongs
only to the pragm atics of explanation.
Of cou rse, van Fraassen's accou n t is first and forem ost an account
of the pragm atics of explanation. But it directly suggests a correlative
accou n t of w hat explanatory understanding consists in. If w e take his
story at face value, explanatory understan din g w ould presum ably
consist in the possession of an acceptable an sw er to a topic question,
"W h y P ? " w here the answ er is acceptable b ecause it entails or oth er
w ise satisfactorily singles out the explanandum " P " from its local
"co n tra st class" of relevant alternatives.
I think this has to be w rong. Explanatory understanding is an ubi
quitous p h enom en on, an alm ost continuously unfolding feature of
the bulk of any cognitive creature's ongoing conscious experience. It
is n ot a specific an sw er that singles out a unique m em ber from a sp e
cific contrast class of possible alternatives, contextually p osed. N or is
it generally an an sw er to any question, explicit or implicit, at all. It is
nothing rem otely so linguistic, so stylized, or so occasional. In gen er
al, explicit questions arise only w hen explanatory understanding, for
som e reason or other, tem porarily fails or eludes us. If w e confine
attention to such atypical cases, then van F raassen 's account of the
pragm atics of such occasional "cognitive reco v ery " m ay begin to look
plausible. Y et even in these atypical cases it is plausible only for
language-using hu m an s. Puzzlem ent and its resolution in n on h u m an
anim als (that is, in m ost of the cognitive creatures on the planet) is
n ot plausibly p ortrayed in van F raassen 's term s. A n d n ot surprising
ly. They are the sam e term s that got the D -N m odel in trouble. They
p ortray know ledge as essentially propositional in ch aracter.
A ccordingly, w hat van Fraassen has given us is, at m ost, an
accou n t of the structure and rules of a certain gam e, an essentially
linguistic gam e, a gam e that is occasionally played by the older m em
bers of a single species of anim al on those exceptional occasions w hen
exp lan atory understan din g for som e reason eludes them . It does not
p rovide, nor does it im ply, a satisfactory accou n t of explanatory
u n d erstan ding itself.
P erh ap s because of his general com m itm ent to constructive
em piricism — w hich sees bare descriptive empirical ad eq u acy as the
only genuine epistem ic virtue— van Fraassen despaired too quickly of
finding anything unitary, interesting, and epistem ologically significant
in the general topic of explanation. O n the prototype-activation
m odel, h ow ever, explanatory understanding em erges as being on a
cognitive continuum w ith perceptual recognition, a genuinely episte-
m ological m atter if ever there w as. A dditionally, as w e saw at the end
of the p receding section, explanatory unification is an im p ortan t and
understandable determ inant of how successfully one is able to g en er
alize p ast experience to new cases, w hich is also an epistem ological
rath er than a m erely pragm atic m atter. Let us look m ore deeply into
this topic.
I have m ade conceptual unification an im p ortan t elem ent of the
accou nt p roposed. But the presen t accoun t is not the only one that
accords a central im portance to explanatory unification. Friedm an
(1974) and Kitcher (1981, 1989) have both urged that the virtue of an
explanation is a function of w hether it is, or prom ises to be, an ele
m ent or instance of a unified account of a broad range of p h en om en a.
Explan atory virtue, according to them , is not just a m atter of the rela
tion b etw een the local explanans and the local explanandum . It is also
a m atter of the global virtues of the fram ew ork of w hich the local
exp lanans is but one application. While each offers a different account
of w hat the explanatory unification of our know ledge consists in,
they are agreed in m aking participation in such unity the dom inant
virtue of any explanation.
O n the view of explanation here being defended, th ey are alm ost
certainly right. Som e will resist the elevation of unity to this level of
im p ortan ce, arguing that explanatory unity is a purely aesthetic or
pragm atic virtue, valued by h u m ans, p erh ap s, but strictly irrelevant

to truth (van Fraassen 1980). Y et w e have already seen in section 5
h ow a "su p erem p irical" virtue like conceptual unity has inevitable
consequences for the “empirical ad eq u acy'' of a cognitive configura
tion, as displayed in its ability to discrim inate correctly novel in
stances of a learned category. The lesson here repeats an earlier lesson
(urged in ch apter 8): w e should be very reluctant to assent, as van
Fraassen does, to any distinction draw n betw een the "gen u in ely
em p irical" and hence "tru th -relev an t" virtues of a theory, and its
"m erely superem pirical" and hence "tru th -irrelevan t" virtues. The
distinction is w ithout integrity.
W e can push the general lesson of section 5 even farther. If p ercep
tual recognition and explanatory understanding are really instances
of the sam e form of cognitive achievem ent, as I have su ggested re
peated ly in this essay, then it is proper to regard perceptual recog
nition itself as being just a case of explanatory understanding at the
sensory periphery. O n this view , the factors that m ake for w orth y
m od es of explanatory understanding are the very sam e factors that
m ake for w orth y m odes of perceptual apprehension of the w orld in
the first place.
This vital point is the m otivating them e of m y 1979 book. The in
sight derives ultim ately from C. S. Peirce, w ho explicitly voices the
position on perceptual recognition here being defended: "P ercep tu al
jud gm en t is the limiting case of abductive ju d g m en t" (H artshorne
and W eiss 1935, vol. 5, para. 186). W e can now sustain Peirce's
insight (and G regory's and Rock's) w ith a revealing and neurally
grou n ded accou n t of w hat "abd u ctive ju d g m en t" really consists in. It
consists in p rototype activation. W ith choices at the empirical
p eriphery thus hopelessly hostage to explanatory values, w e can now
see that the virtue of explanatory unity m u st play a dom inant role in
any adequate epistem ology. It cann ot be d em oted to a m erely p rag
m atic statu s, as it is in the austere ideology of constructive em piricism .
The PD P perspective also allows us to say som ething useful about
the conflict betw een the Gibsonian "ecolog ical" approach to p ercep
tion and the "n e w look" approach to the sam e topic. The first point to
m ake is that the G ibsonians appear to be right on the following m at
ter. Perceptual processing is im pressively fast and direct: inference
app ears to play no role at all. O n the other hand, the new -look th eo r
ists appear to be right on the following point. O ur perceptual process
ing is profoundly theory laden, in that it is dictated largely by the
accum ulated know ledge— the configuration of w eigh ts— acquired in
the course of p ast experience. There is no longer any conflict b etw een
these tw o convictions. They are both a p art of the PDP perspective.
A notable view that does suffer, h ow ever, is Fod or's view (1984),
w hich attem p ts to em brace the w orst of both positions. Fod or and
Pylyshyn (1981) have argued against the G ibsonians that the p ercep
tual p ro cess is extensively inferential, and Fodor (1984) has argued
against m e (1979), and other N ew Look sym pathizers, th at n everth e
less the process is basically theory neutral. For a b etter insight on
these issues, see Hatfield 1989.
Let u s retu rn to K itcher's account of explanation. It does m ore than
com pellingly p ortray unity as the preem inent cognitive and explana
tory virtue that it is. It also discovers and profitably deploys the idea
of p ro to ty p es. Kitcher takes an im portant step aw ay from the tradi
tional D -N picture by introducing the notion of an abstract argum ent
pattern . These are highly abstract logical structures characteristic of
the particular scientific theory being exploited, stru ctu res that will
yield a great variety of relevantly specific argum ents w h en instanti
ated to suit a specific explanatory problem . O ne is rem inded here of
Kuhn 1962. But unlike K uhn, Kitcher explains the m atter in great
detail and show s how an im pressive am ount of light can be throw n
on m any chronic problem s if w e adopt his point of view .
From the PD P perspective, K itcher's achievem ent m u st be judged
g enuine. The deploym ent of prototypes is w h at is central to explana
tory un d erstan din g, and for m axim al virtue it is vital that they be part
of a unified cognitive configuration. The problem with K itcher's story
is that it is still m uch too narrow ly linguistic in its conception of know l
edge rep resentation , and it encom p asses only a small subset of the
full range of different types of explanation. These are serious defects
in the story. E ven so, I w ould m uch rath er highlight its equally p rom
inent virtues, since I w ould like to claim that the P A account
successfully red u ces it. The incorporation of the general features
of K itcher's story further illustrates the conceptual unity that the
prototype-activation m odel brings to the topic of explanation.
If the conceptual unity it brings is genuine, then it ou gh t to perm it
the recognition and understanding of hitherto u n recognized types of
explanation. There is som e sign of this. The belated recognition of
legal, m oral, and social-etiquette explanations w as a novel experience
for m e, I am som ew h at em barrassed to say, and one m ight suggest
aesth etic explanation as a further category. Perhaps oth ers will see
m ore.
Finally, let m e close this essay by adverting to a highly general vir
tue of the prototype-activation m odel. I am com pelled to cite it, since,
as I look back over this essay, I am distressed at how fragile is the
accou n t p rop osed , and how sketchy are the few details provided.
W hat m akes m e hopeful in the face of this is the fact that the
prototype-activation m odel of explanation m eets its ow n prim ary

condition of virtue, a condition of ind ep en d en t and an tecedent
plausibility. Specifically, the m odel brings a w elcom e and revealing
unity into a stubborn diversity of explanation types, and the m odel is
itself an integral part of a highly unified background conception of
cognitive activity, one that encom p asses with som e success the
general structure and activity of biological brains, and the structure
and cognitive behavior of a new class of artificial com puting system s.
For this reason , if no other, w e should be m oved to explore it further.
Chapter 11
Learning and Conceptual Change
1 Introduction
To cognitive creatures, the w orld is a highly am biguous place. N ot

just in the am biguity it presents to our sensory system s, w here the
initial coding is typically consistent w ith a diversity of external cir
cu m stan ces, but m ore profoundly in the am biguity it p resen ts to our
conceptual system s. A n y conceptual fram ew ork, no m atter how robust
or natural its categories m ay seem to us, is but a single point in a
practically infinite space of alternative possible fram ew orks, each
w ith a com parable a priori claim on our com m itm ent. Som e of the
fram ew orks in this vast and alm ost entirely unexplored volum e will
be closely similar to our current schem e, b u t countless others will be
so distant and alien as to escape intelligibility to us, sh ort of a long
period of reeducation.
This talk of a vast space of alternatives is not m erely rom antic. Each
of us h as a history of conceptual diversity already. For you w ere not
born w ith you r adult fram ew ork. You cam e to it slowly, throu gh a
long p eriod of developm ent. There is indeed a space, through w hich
each of us h as a com plex journey already com pleted.
A n individual's conceptual history is represented by a specific
trajectory throu gh this vast space of conceptual alternatives. That
trajectory is traced by a point that changes its position swiftly and
dram atically in the early stages of life, m ore slowly in later childhood,
and only v ery slowly throughout the adult years. Fortunately, sheer
determ ination can extend the capacity for continued exploration of
the vast space available, at least in som e individuals, and this m akes
possible an ongoing tradition of institutionalized scientific research.
Talk of conceptual "s p a ce " m ay seem m etaphorical still, but as out
lined in the tw o preceding chapters, recent research h as show n us
how to m ake literal and very useful sense of it. If w e assum e that the
h u m an brain is a m ultilayered netw ork of intercon n ected units, w e
can uniquely specify its current position in conceptual space by spe
cifying the individual strengths or w eights of its m yriad synaptic con
nections (figure 11 .lb). That configuration of w eights can be directly

rep resented by a specific point in a m ultidim ensional space, a space
w ith a distinct axis for each of the brain's 1014 synaptic connections
(figure 1 1 .la). For а h um an brain, therefore, this "w eigh t sp a c e ," as
it is called, will have fully 1014 dim ensions with at least 10 possible
positions along each. Its volum e is alm ost unim aginably v ast— at
least Ю1014 functionally distinct positions— as our guiding m etap hor
suggested.
A nd there is a second space to consider h ere, com parable in its
vastness: the space of possible activation patterns across the brain's
1011 neurons (figure 11.1c). This "activation -vector sp a c e ," as it is
called, h as a distinct axis for each neuron, an axis that m easures the
level of that n eu ron 's activity. A s outlined in the earlier ch ap ters, a
specific configuration of synaptic w eights will partition the activation
space of a given neuronal layer into a taxon om y of distinct prototypes
or " universals" (figure 11.1c depicts a simple binary partition). To sp e
cify that global configuration of w eights is thus to specify the global
conceptual fram ew ork currently in use by the relevant individual. To
change any of those w eights is to ch an ge, h ow ev er slightly, the con
ceptual fram ew ork they dictate. To trace a creatu re's actual path
throu gh the space of possible synaptic configurations w ould be to
trace its conceptual history (figure 11.1a). A nd to un d erstan d w hat
factors induce changes in those w eights w ould be to u n d erstand w hat
drives conceptual change.
The p resen t ch ap ter continues the exploration of this view . I wish
to address four problem s in particular. The first concerns the p h e
n om enon of multiple conceptual com p etence. The second con cern s
an im portant distinction betw een genuine conceptual change and
m ere conceptual redeploym ent. The third problem concerns the fac
tors that drive these tw o kinds of change. A nd the fourth con cern s
the recently ap p rehended vastness of cognitive space, and the need
to autom ate not just the experim ental b ut also the theoretical as
p ects of the scientific enterprise, if w e are ever to explore that space
effectively.
I begin by reraising a problem addressed in ch apter 9. Should w e
identify on e's conceptual fram ew ork w ith the configuration of syn ap
tic w eights in o n e's brain? O r w ith the partitions they effect across the
activation vector space of the assem bled neurons to w hich they con
n ect? O r p erh ap s w ith the overall input-output fu n ction that the net
w ork com es to instantiate? The w eights uniquely dictate both the
partitions and the function, but despite the functional prim acy of the
w eights, there are good reasons for identifying the partitions, and the
Learning and Conceptual Change
Figure 11.1
233
a) Weight space: the space of all possible configurations of the network's synaptic weights. Axes are
shown for only 3 of 27 synaptic weights, b) A schematic network, c) Hidden-unit activation-vector space,
the space of all possible activation vectors across the population of hidden units
function they serve, as reflecting m ost directly the an tecedent notion

of a "con cep tu al fram ew ork ."
W hile the w eights are of essential im portance for understanding
long-term learning and fundam ental conceptual change, the parti
tions across the activation space, and the prototypical hot-spots they
harbor, are m uch m ore useful in reckoning the cognitive and b e
havioral similarities across individuals in the sh ort term . People react
to the w orld in similar w ays not because their underlying w eight co n
figurations are closely similar on a synapse-by-synapse com parison,
but because their activation spaces are similarly partitioned. Like trees
similar in their gross physical profiles, brains can be similar in their
gross functional profiles, while being highly idiosyncratic in the
m yriad details of their fine-grained arborization (com pare Quine
1960, p. 8).
Thus, a perfect identity of w eight configurations will indeed p ro
duce a perfect identity of partitions on the activation space, but one
can also achieve alm ost identical partitions w ith a large variety of
quite different w eight configurations. Synaptic contrasts in one place
m ay com p en sate for further synaptic contrasts in another place, so
that the functional profile of tw o brains m ay end up practically the
sam e. A t least for now , therefore, let us adopt the partitions and the
functions they serve as the closest available neural analog of w hat the
philosophical tradition conceives as our "con cep tu al fram ew o rk ."
(W e could insist on the point in w eight-space as the m ore profound
analog, as I earlier recom m end ed [p. 177], and w e m ay yet find
reason to do so. But if w e do, a vital similarity m easure across the
w eight configurations will still advert to the partitional and functional
factors at issue.)
2 M u ltiple C onceptual Com petence
H um ans have the occasional capacity to ap p rehend the sam e thing in

one of tw o or m ore quite different w ays, often at will. Exam ples range
from the simple case of being able to see a familiar curve as now a
duck and now a rabbit, to the m ore unusual and global case of
ap prehending natural ph enom ena now in an A ristotelian fashion,
n ow in a N ew tonian fashion, and now in an Einsteinian fashion.
H ow is it possible, on the netw ork m odels at issue, for a single indi
vidual to do this? H ow is it possible for one to bring distinct con cep
tual resou rces alternatively to bear on one and the sam e problem ?
H ow could one train a netw ork to have this capacity?
The problem is that to train a netw ork to any sort of com petence is
to im pose a function on it. A function delivers a unique output for
Learning and Conceptual Change 235
any given input. But the situation at issue seem s to require that the
netw ork som etim es deliver a different output given the sam e input.
These d em an ds are incom patible, and the solution is to recast the
problem so that the relevant inputs are not strictly identical after all.
W ith the inputs distinguished, w e can then train the n etw ork to joint
com p etence on tw o functions with nonoverlapping dom ains.
A n exam ple will illustrate. Suppose w e w ish to train a n etw ork to
give as o u tp u t the sum of any tw o single-digit num bers en tered as
input. A typical input w ould be <2, 5>, and the desired outp u t <7>.
The relevan t function is thus a smallish set of ord ered pairs, each pair
being an input and its p roper output. N ow suppose w e also w ish to
train the netw ork to give as output the product of any tw o single-digit
num bers entered as input. A typical input w ould be <2, 5 > , and the
desired outp u t <10>. W e are now asking the im possible, unless w e
m ake it possible for the netw ork som ehow to discrim inate inputs on
w hich addition is to be perform ed from inputs on w hich m ultiplica
tion is to be perform ed.
This is easily done. A dd to each input a "con text-fixin g " elem ent,
so that w e en ter <2, 5, + > w hen w e w ant addition perform ed, and w e
enter <2, 5, X > w hen w e w ant multiplication performed. W ith the sets
of input elem ents from each function now com pletely disjoint, the
union of the tw o functions desired will be a legitim ate function also.
W e then train the netw ork on that larger function (i.e ., on that larger
set of in p u t/o u tp u t exam ples), and the result is that it learns both to
add and to multiply. Intriguingly, the typical result of such training is
the partitioning of the hidden-unit activation space into tw o distinct
subspaces, one in w hich the various m ultiplicative transform ations
are all cod ed , and the oth er in w hich the various additive tran sform a
tions are all coded. W hat trips these different cognitive behaviors is
the ap p earan ce of one of the context-fixing elem ents, " + " or " X " , as
part of each input.
This sim ple exam ple illustrates a general strategy. A n etw ork can
learn to deal w ith the "s a m e " class of inputs in a variety of different
w ays according to w hich context-indicating elem ents accom pan y
those inputs. Those context fixers can and often will be supplied by
the external environm ent. But that is n ot their only possible source.
The only firm requirem ent is that the context-fixing inform ation
som ehow reaches the hidden units along w ith the rest of the input
inform ation. This requirem ent is consistent w ith the con text fixers
having their origins som ew here else w ithin the netw ork itself, esp e
cially if w e are dealing w ith a com plex netw ork like the brain, w hich
has m an y layers and m any distinct processing path w ays. The idea is
that the total input to the relevant layer of hidden units includes
either “ descen d in g" or "h o rizo n tal" projections from one or m ore
distinct layers elsew here in the larger netw ork. In figure 11.1b, this
could be visualized by im agining som e additional axons com ing in
from the side to form synaptic connections onto the hidden units over
and above the connections already arriving there from the sensory
periphery.
This yields a netw ork w here the m an n er in w hich the sensory input
is processed is both variable and under the control of the netw ork
itself. Such a netw ork can p rocess a given p erceptual input in any one
of several different w ays. It can " s e e " a figure as either a duck or a
rabbit, as it w ere, depending on w hich of its activational subspaces
(conceptual resources) it has kicked into gear with the relevant
context-fixing inputs. This is true not only at the level of perceptual
p rocessing, b ut at higher levels as well. A situation draw n from
m em ory, or contem plated in im agination, or app reh en ded from a
printed description can also be u nd erstood in a variety of different
w ays, depending on w hat contextual inform ation accom panies its
ap prehension at the relevant population of p rocessing units. It is not
m ysteriou s, then, on the m odel of cognition here being explored, that
one can learn to perceive/understand the w orld in an Aristotelian
fashion, and in a N ew tonian fashion, and in an Einsteinian fashion,
and then use each fram ew ork (each subvolum e of the activation
space) by turns, to suit either aesthetic w him or practical occasion.
3 Conceptual Change versus C onceptual R edeploym ent
A possible problem for the v ector-processin g m odel of cognition co n

cerns the time course of conceptual ch an ge. O n the face of it, the
unique determ inant of a netw ork 's conceptual resou rces (activation-
space partitions) is the netw ork's configuration of synaptic w eights,
and the only w ay to change the form er is to change the latter. But
ch an ges in the w eight of any given synaptic connection h appen only
in small increm ents. A ccordingly, the learning netw ork 's global
trajectory through w eight space (figure 11.1a) m u st always approxi
m ate a continuous path. But conceptual change is a p rocess that at
least occasionally displays dram atic discontinuities. Scales fall from
the eyes, the light daw ns, the structure is suddenly app aren t, and so
forth, som etim es on a time scale of secon d s. H ow to reconcile the
ap p aren t conflict?
There are several w ays w e m ight approach this problem . First, w e
m ight highlight the recent evidence indicating that changes in the
num ber an d /o r surface area of synaptic connections m eeting a given
cell can, at least in som e circum stances, take place on a time scale of
m inutes (D esm ond and Levy 1983). A related line of evidence con
cerns th e long-term potentiation (LTP) of neuronal resp on se in som e
cells, a p ro cess also occurring in the one-m inute range (D esm ond and
Levy 1986). W hile this evidence is encouraging, it rem ains unclear
w h eth er such changes are either large enough or fast en ou gh to solve
the problem at issue.
A secon d approach points out that the functional relation betw een
the w eight configurations on the one h an d , and the resulting parti
tions on the activation space on the other, is a highly nonlinear and
occasionally volatile relation (see p. 172). Relatively small ch an ges in
the w eigh ts can occasionally produce large and su dden ch an ges in
the partitions across the conceptual space. This fact could m itigate the
problem atic lethargy show n by the point in w eight space.
A third approach m ight attem pt to play dow n the discontinuities
claim ed for our conceptual behavior, by arguing that historical and
autobiographical descriptions of the relevant shifts have b een chroni
cally exag gerated . H ere w e w ould attem pt to deny, or play d ow n , the
p h en om en a w e had originally hoped to explain. Som e com bination of
these poin ts, perh ap s, will constitute a solution to the problem .
P erh ap s they m ay. Fortunately, the discussion of the preceding
section m akes available to us im m ediately a quite different and rather
less strained solution. There is a w ay to accou n t even for large-scale
conceptual shifts, on a time scale as short as m illiseconds, that re
quires n o m otion from the w eight-space point w h atsoever. The cru
cial idea is the idea of conceptual redeploym ent, a process in w hich a
conceptual fram ew ork that is already fully developed, and in regular
use in som e oth er dom ain of experience or com prehension, com es to
be used for the first time in a new dom ain.
E xam p les are m any and familiar. C onsider H u ygen 's seven teen th -
cen tu ry realization that optical phen om en a, previously grasp ed via
the ray traces of geom etrical optics, could be m ore com prehensively
u n d erstood as instances of w ave p h enom en a. H ere the theory of
w aves in m echanical m edia— a theory already w ell-form ed in
H u y g en 's m ind in connection w ith sound w aves and w ater w aves—
w as applied in a dom ain hitherto un ad dressed by that fram ew ork,
and w ith system atic success. There w as no need for H uygens to effect
a global reconfiguration of his synaptic w eights to achieve this con
ceptual shift. H e had only to apprehend a familiar class of ph e
nom ena in a new cognitive context, one supplied largely b y him self,
in order to have the inputs activate vectors in an area of his con cep
tual space quite different from the areas they had previously acti
vated. The difference w as the context fixers b rough t to the problem .
A nd the result w as a radically new u n d erstanding of optical p h e
n om ena. The novelty, h ow ever, consisted in the unusual red ep loy

m en t of old resou rces, not the creation from scratch of new ones. No
new resou rces w ere created; nor w ere any old resou rces destroyed.
A second exam ple is provided in the various seven teen th -cen tu ry
attem p ts to apply the conceptual resou rces of terrestrial m echanics to
the case of m otions in the superlunary h eaven s, a dom ain long
thou gh t to be governed by distinct and divine principles. From the
rectilinear perspective of the recently developed m echanics, the circu
lar m otion of the planets around the Sun— w hich constitutes a cen
tripetal acceleration tow ard the Sun— clearly asked for а fo r c e on the
planets directed tow ard the Sun. It could be a push from the outside
or a pull from the inside, but from the new perspective it had som e
how to be there. D escartes' vortex theory tried to fill out the story in
the first w ay; N ew ton 's gravitational theory tried, with m ore striking
su ccess, to fill it out in the second. But in both cases, existing co n cep
tual resou rces w ere being reapplied in a new dom ain. Contem plation
of the heaven s w as now activating, on a regular basis, prototype
vectors that w ere initially provoked only in respon se to terrestrial
situations.
A third and very striking exam ple is the system atic reconception of
optical p h enom en a as electrom agnetic ph en om en a, a shift that spread
quickly throu ghou t the scientific com m unity of the late 1800s. Jam es
M axw ell's beautiful sum m ary of the relations betw een electric and
m agn etic fields entailed the existence of a w avelike electrom agnetic
disturbance, spreading out from any oscillating ch arge, w ith a veloc
ity of (ju, X e)~1/2, w here ц and e rep resent the m agnetic perm eability
and the electric perm ittivity of the surrounding m edium . (These fea
tures are a m easure of how m uch an electric or m agnetic field is "d i
lu ted " w h en it p asses through the m edium in question.) For the
atm osphere these tw o values w ere well know n. A quick calculation
yielded a velocity of roughly 3.0 X 108 m /s for such spreading electro
m agnetic disturbances, a velocity indistinguishable from the m eas
ured velocity of light. This extraordinary coincidence invited an
attem p t to see further optical ph en om en a as facets of oscillatory elec
trom agnetic phenom en a.
A s it developed, this electrom agnetic reincarnation of H u ygen 's
m u ch sim pler vision im m ediately displayed all of the virtues of its
an teced en t, plus an u n expected cornucopia of further virtues. Elec
trom agnetic (EM) w aves w ere tran sverse, and thus w ere polarizable,
just as light had p roved to be. Unfamiliar features of tran sp aren t sub
stan ces, such as their perm ittivity (e) and perm eability ( /i.), suddenly
becam e salient, since it is they that dictate the differing velocities of
EM w aves in the relevant substances, and it is those relative velocities
that dictate the refractive index of any substance. The refractive in
dexes for transp aren t substances w ere already well know n, and the
system atic agreem en t w ith the predictions of the new theory w as
striking. These various indexes suddenly em erged as tran sp aren t
instances of the electric and m agnetic properties of m atter.
H ere again a familiar dom ain w as am biguous and p roved to be
understandable in m ore than one w ay. W h en addressed w ith the
appropriate context-fixing inputs (perhaps no m ore than the adm oni
tion "A n y ray of light is a train of EM w av es!"), optical p h en om en a
b egan system atically to activate vectors in an un exp ected subvolum e
of conceptual sp ace, a subvolum e that w as initially partitioned by its
extensive training on entirely nonoptical p h en om en a. M oreover,
after extended practice at approaching the old p h en om en a w ith the
new subvolum e in gear, one clearly did better at understanding
things than one did w ith the old fram ew ork. A nd finally, a m ajor
virtue of this shift, a virtue displayed in both of preceding exam ples
as well, is that one now had a unified understanding of w h at initially
ap p eared as disjoint empirical dom ains.
It is clear from these three exam ples that conceptual change is
regularly a m atter of conceptual redeploym ent, as opp osed to fun
dam ental conceptual novelty. It is also clear that such shifts can
initially take place, in a given individual, on a time scale of secon d s or
less, although the full exploration of the novel use of old resources
m ay well take years. Indeed, so m any of the historical exam ples fit
this red ep loym en t m old that one m ay begin to w on d er if history con
tains any exam ples of real conceptual novelty. I believe th at it d oes—
F arad ay 's conception of a "field of fo rce" com es quickly to m in d — but
I also su sp ect that such cases are relatively rare. The bulk of the con
ceptual discontinuities displayed in the history of science are clearly
cases of con cep tu al redeploym ent.
These can often be cases of learning, h ow ever, in the d eep er sense
and b eyond the m aking of the shift itself. The redeployed resou rces
seldom survive extended contact w ith the new dom ain entirely u n
ch anged, and the process of shaping and refocusing those resources
is a p rocess in w hich the relevant subvolum e of on e's activation space
is now subject to a new regim e of training exam ples. This will often
lead to a y et m ore subtle articulation of the an tecedent partitions,
a p rocess of learning that is com paratively slow and thu s easily
explained in term s of the gradual m otion of on e's w eight-space
position.
W hat w e have to acknow ledge is that the notion of "le a rn in g " is
starting to fragm ent in interesting w ays. Beyond the basic b ut com p ar
atively slow process of synaptic adjustm ent, there is the m ore sh ort
term p rocess w hereby one learns how to deal w ith a puzzling new
situation by repeatedly reapprehending it in conjunction w ith various
context-fixing auxiliary inputs, in h op es of eventually activating som e
robust p rototyp e vector w ithin a subvolum e that is already well
trained. The ten-year old takes apart the old alarm clock and after a
half-hour's pondering sees how it all w orks. The m ath student p u z
zles over a hom ew ork problem and after several false starts suddenly
sees the p ath through it. The physician confronts a confusing set of
sym p tom s and, several failed tests and incoherent diagnoses later,
finally lights on a successful one. These are all paradigm cases of
learning, in the sense of "com in g to u n d e rsta n d ," but the underlying
process here is quite different from the slow process that partitions
on e's activation spaces in the first place.
The frequency and im portance of conceptual redeploym ent re
quires us to acknow ledge a further divergence, w hich I have been
suppressing to this point, b etw een the partitions across on e's activa
tion space(s) and the input-output function one instantiates. Plainly it
is possible for tw o people to have closely similar partitions, but w ide
ly divergent deploym ents: they m ay com m an d essentially the sam e
conceptual resources, but apply them to quite different dom ains. Tw o
physicists (N ew ton and H uygens, say) m ay have a com parable com
m and of both projectile m echanics and w ave m echanics, and yet one
chronically un d erstan ds light as the high-speed ballistic m otion of
tiny corpuscles, while the oth er chronically u n d erstan ds light as a
train of com pression w aves in the ether. The sam e sensory inputs
produ ce different conceptual respon ses in each, and thus different
behavior from each , since particles and w aves often call for different
techniques of m anipulation and behave differently in m any circum
stances. Because of their different applications, the tw o physicists will
h ave a different input-output function, despite com m anding identical
con cep tu al resou rces.
A m ore familiar case concerns the dom ain of application of the
resou rces of folk psychology. W e all share a m ore or less com m on
conception of intentional agents, and yet som e of us, the devoutly
religious, chronically find intentional significance in a wide range of
ph en om en a that are perfectly natural to the rest of u s. The interp re
tation of natural disasters and natural blessings, both large and small,
as the deliberate punishm ents, rew ards, and m essages of a hidden
intentional agent (God) w ho has specific expectations of us is an inter
pretation still em braced by a large portion of the con tem porary p op u
lation. N ot surprisingly, their behavior is often quite different from
those w ho deploy the relevant resou rces m ore narrow ly: they dis
play p rayer, sacrifice, p enance, resignation, and so forth. H ere again
w e have com m on resou rces differently applied, with a quite different

global inp u t-ou tpu t function as the result.
Such cases give us reason to regard a p erso n 's trajectory through
w eight space as capturing only a part of w hat w e w ould norm ally
regard as one's conceptual evolution. The repeated red ep loym en t of
existing conceptual resources can produ ce som e profound ch an ges in
on e's cognitive and practical life, w ith only m inimal ch an ges in the
configuration of o n e's synaptic w eights and in the activation-space
partitions that they p roduce. If w e w ant to know w hat drives con cep
tual ch an ge, then , w e m u st address both the dynam ics of the m oving
point in w eight space and the m ore superficial but still vitally im por
tant dynam ics of conceptual redeploym ent.
4 W hat D rives C onceptual Change?
W e m ay begin by asking the narrow er question of w h at drives

conceptual redeploym ent. H ere there are several factors, the first of
w hich is blind luck. That M axw ell's EM theory should have yielded a
velocity for EM w aves exactly equal to the know n velocity of light w as
the sh eerest serendipity. That N ew ton 's analysis of bodies falling
u n d er an inverse-square law should have yielded K epler's well-
k now n ellipses w as an oth er stroke of sheer good fortune. O ther
theories (H u ygen 's) and other analyses (D escartes') w ere not nearly
so fortunate. But p erhaps these are factors that pull rath er than push.
W hat factors positively drive conceptual redeploym ent?
Frustration with the poor perform ance of older fram ew orks figures
prom inently. Recall K uhn's analysis of gathering anom alies, crisis
science, and the resulting radiation in conceptual ap p roaches to old
problem s. Individuals show an increasing willingness to explore, and
the scientific com m unity show s an increasing willingness to tolerate,
u n orth od o x conceptions of recalcitrant p h enom en a. Old inputs are
rep eatedly reen tered into one's already trained netw ork, w ith a varie
ty of increasingly unusual context fixers, in hopes of activating som e
an tecedently-developed prototype vector in a subvolum e of activa
tion space hitherto devoted to other ph enom en a entirely. Should
success be achieved, the (hyper)distance betw een the old and new
p rototyp e v ectors is a m easure of how great the conceptual change
effected.
T hough gathering anom alies are perhaps the m ost com m on force
behind su ch explorations, they are clearly n ot necessary for conceptual
exploration of this kind. The simple desire for theoretical unity can
drive a system atic search for new w ays of com p reh en d in g old p h e
n om ena, even w hen the old w ays are functioning quite nicely. H ere
the only defect that need be felt in on e's cu rrent conceptual resources
is the fact that they are still diverse rather than unitary.
Exam ples are com m on enough. Classical, or "p h en om en o log ical,"
therm od yn am ics w as enorm ously successful (it helped to produce
the industrial revolution), but this did not dissuade the tradition of
Bernoulli, Joule, Kelvin, and Boltzm ann from repeatedly trying to re-
conceive therm al ph enom ena w ithin the broader fram ew ork of kine
tic and corpuscular theories. N ew tonian m echanics had conquered
m otion at both the astronom ical and the h u m an scales. O ne had to
w on d er if it also held true at the subm icroscopic scale. The possibility
of ap prehending heat as m echanical en ergy at the m olecular level w as
therefore v ery inviting.
A n unusual sensitivity to failures of unity seem s to h ave driven the
g reater p art of Einstein's theoretical w ork. Special relativity w as an
attem pt to bring m echanics and electrodynam ics togeth er un d er a
com m on and internally coh eren t roof. G eneral relativity w as an
attem pt to unify the physics of both accelerated and unaccelerated
reference fram es. In both cases the new conceptual perspective w as
provided by four-dim ensional geom etry w ith n on stan d ard m etrics.
But in neither case w as the search for this m ore unitary perspective
driven by any prom inent experim ental failing in any of the older
view s. His later search for a unified field theory is a further instance
of the sam e general yearning.
This im pulse tow ard unity is vitally im portant in any cognitive cre
ature, an im pulse coequal with sensitivity to the d ata, for reasons w e
can now u n derstand (see chapter 9, section 5). It is curious that the
relative strengths of these tw o im pulses seem so variable across indi
viduals. A m ajor im balance in either direction yields a familiar pathol
ogy. Valuing unity at the radical expense of local empirical success
yields a castle-in-the-air fantasy w orld for its victim to live in. A nd
valuing local empirical success at the radical expense of synoptic
unity traps its victim in a disconnected set of small and w indow -
less room s: it yields a hidebound and narrow vision that will not
generalize successfully to unfamiliar cases.
Finally, conceptual redeploym ent is occasionally prom p ted by
som e fortunate novelty in o n e's experience. A m p ere's observation
that a cylindrical coil of current-carrying w ire produ ced a bipolar
m agnetic field m oved him to reconceive the long-familiar case of
m agnetized iron bars as having circular cu rrents som ehow flowing
inside of them as well, just as in the coil. F resn el's striking d em on
stration of concentric circles of light at the cen ter of a tiny circular
sh ad ow m oved m any thinkers to foresake N ew ton ian corpuscles and
to set about reconceiving light on a w ave m odel. A nd Einstein's m uch
later observation of the curious photoelectric effect m oved him to re-

conceive light w aves as quantized after all. In such cases, the striking
new p h en om en o n discovered in a familiar dom ain is capable, all by
itself and w ith ou t arduous context fixing, of activating specific vectors
in som e h eterod o x portion of one's activation space. H ere the ph e
n om ena them selves have the salient character n ecessary to activate
an unusual interpretation directly.
C onceptual shifts of the kind un der discussion constitute p erh ap s
the g reater portion of our scientific developm ent, but they seem su
perficial relative to the prior and deeper learning process by w hich
our activation spaces are partitioned in the first place, and by w hich
they m u st be readjusted. W h at factors drive change at this m o st fun
dam ental level? W h at forces the w eight-space point to m ove? This
rem ains an open question. The best I can do on this occasion is briefly
to sum m arize the dynam ical alternatives salient both in netw ork
m odeling and in physiological research . (A m ore com prehensive
survey of the form er is available in H inton, forthcom ing.)
Supervised Learning
This class of learning algorithm s requires a suitable n um ber of
ord ered pairs, draw n from the function to be learned, for p resen ta
tion to the netw ork. Of crucial im portance is the availability of the
correct ou tp u t for any of the given inputs. The m ost p opular algor
ithm of this kind is currently the back-propagation p rocedure described
in ch apter 9, section 5. H ere the desired outpu t vector is com p ared
elem ent-by-elem ent with the actual outp u t vector p rodu ced by the
student n etw ork in response to a training input. The difference be
tw een the tw o outputs is then used to com pute a small and propri
etary adju stm ent for each w eight in the entire netw ork. That change
is such as to nudge the netw ork's global point in w eight space a small
distance d ow n the steepest slope of the local "e rro r su rfa ce ." The
path of rep eated nudgings, produced by repeated presen tation s of
the inp u t-ou tp u t pairs in the training set, is a path that eventually
leads to a position of low error or accurate perform ance on all of the
training exam ples.
This learning procedu re is highly effective, but it does have several
severe sh ortcom in gs, both as an account of how biological brains
learn, and even as a technology for training artifical nets. First, the
requirem ent that the correct output be available to the learning n et
w ork in every case is clearly unrealistic. After failing to solve a prob
lem , real brains do not generally get to look at the correct an sw ers at
the back of the book. Second, the brain show s no plausible m ech an
ism s for com puting and distributing su ch globally inform ed adjust

m en ts to its m yriad w eights. A nd third, the back-propagation p ro ce
dure scales upw ard to large netw orks only very poorly. The num ber
of w eights increases exponentially w ith the num ber of neuronal units
in the netw ork, and thus so does the task of com puting a proprietary
adjustm ent for each, especially w h en the p rocedure requires
thou san d s of repetitions. Even for artificial nets, this will im pose a
ceiling on the size of the nets w e can train in acceptable tim e w ith the
back-propagation algorithm . The training of netw orks approaching
103 units already poses a serious strain on the best existing m achines.
Boltzm ann learning p roced u res offer som e im provem ents over back
propagation , although they bring defects of their ow n. Let us address
the sam e kind of three-layer netw ork d iscussed above, and the sam e
problem of training it up on a given set of inp u t-ou tp u t vector pairs.
H ere w e take an inp ut-ou tp u t pair and "cla m p ” each onto the input
and outp u t units respectively. Suppressing nicities, w e m ay describe
the Boltzm ann procedu re for synaptic adjustm ent as follows. Each
synap se is sensitive to the level of activity it receives from its ow n
axon, and to the level it finds in the unit to w hich it connects. U n d er
Boltzmann learning, its weight will be driven up or down to conform
w ith these local constraints; that is, they will tend to assum e local
values that w ould tend to produ ce the local outp u t given the local
input.
Given that the vectors across the input and outp u t layers are tem
porarily fix e d , and that the input vector is repeatedly propagated u p
w ard through a netw ork w hose synaptic w eights are set at ran d om
v alues, there will inevitably be failures of fit betw een the activity
levels of the fixed outpu t vector and the sum m ed synaptic activity
that reaches it from below . These clashes produ ce changes in the
w eights of the synapses m eeting the final, or outp u t, layer of units.
Similarly, a w hole hidden unit w hose ou tp u t synapses are u n d er uni
form pressu re to shrink (or expand) is a unit w hose internal activation
level is put un d er pressu re to shrink (or expan d ). This pressu re is
released by appropriate reductions (increases) in the w eights of its
input synapses.
O verall, and in respon se to these m ultiple constraints, the n et
w ork 's w eights have a tenden cy to settle gradually into a final config
u ration that is consistent w ith the overall constraints, and especially
w ith the constraints im posed by the fixed outp u t vector. This is a
configuration that w ould yield a real o u tp u t vector that is similar to
the training output v ector, if the netw ork w ere suddenly unclam ped
and presented again w ith the relevant input. (W ith back propagation,
by con trast, the presentation of each training pair p rodu ces only a
tiny ch ange in the w eight-space position and only a tiny im prove
m en t in the netw ork 's p erform ance.)
In ord er to lubricate this p rocess, and to free it from getting cau gh t
in purely local error m inim a of poor final p erform ance, each w eight is
rep eatedly subjected to ran d om increases and decreases during the
course of settling. These ran d om changes are of significant size at the
beginning of the settling p rocess, but are slowly reduced to zero as it
p roceed s. It is as if the netw ork 's critical elem ents w ere h eated up at
the beginning and slowly allowed to cool in the p ro cess of finding
their m inim ally stressful m utual accom odation of w eights. This p ro
cess is called "sim ulated an n ealin g." In all, the system is evolving
tow ard the m ore probable of its possible configurations relative to the
clam ped constraints. That is w h y the p rocess is called "B oltzm an n
learn in g ."
The p ro cess described constitutes only one training cycle. The
m odified net m ust now be clam ped with the next in p u t-ou tp u t pair,
and the p rocess rep eated. A nd so with all of the pairs in the training
set. The secon d cycle m ay well obliterate m any or m ost of the gains
m ade in the first, but the overall p rocess eventually drives the w eight-
space p oin t to an error m inim um that is com m on to all of the training
pairs. The w eight-space point thus follows a m ore lengthy and
ad ven tu rou s p ath than it does under back propagation, b ut the result
is m u ch the sam e.
There are ad van tages to this proced ure. For one, it does not require
a globally com p u ted error, nor any global system to distribute it.
Global inform ation about error filters across the w hole netw ork , but
by m ean s of m any purely local effects com pu ted locally, in d ep en d en t
ly, and sim ultaneously. For this reason, Boltzm ann learning scales up
to large netw orks rather b etter than does back propagation. The local
ch aracter of the teaching m ay also be m ore plausible from a biological
point of view .
O n th e oth er h an d , Boltzm ann learning is still slow for very large
netw ork s, since they can take a long time to settle to equilibrium, and
a distinct settling cycle m u st be done for each training pair. This diffi
culty can be circum vented w ith artifical netw orks, since the settling
p rocess can be au tom ated to exceptionally high speeds in electronic
analog system s. But it rem ains a prim a facie problem for the p ro ce
dure as an accou n t of learning in biological brains. Fu rth er, learning
can n ot take place unless the correct answ ers (outputs) are som ehow
supplied. Boltzm ann learning, as described above, still d ep en ds on
having the answ ers at the back of the book.
R einforced learning
W e can escape the unreality of an om niscient teach er by exploring
learning proced u res in w hich the error m essages are less well in
form ed. Indeed, w e can use either of the tw o p rocedures just d e
scribed, w ith the difference that the output v ectors to w hich they are
trained are variously degraded versions of the desired output vectors:
they are partial, or they are distorted by ran d om influences. Surpris
ingly, this can actually im prove the quality of learning in som e cases,
since adding a very small am ount of noise to the teaching signals can
sm ooth out the m isleading idiosyncracies of the exam ples in a given
training set (Zipser and Elm an 1987). O n the w hole, h ow ever, d e
grading the teaching signals produces a proportional (or w orse) d e
gradation in the learning process. For the m ost part, the w eight-space
point is still being nudged dow n an error gradient, but it is no longer
taking the steepest path available at each step. It thus takes longer to
find the relevant error m inim um . W e buy a greater realism at the
expense of slow er learning.
N either the back-propagation nor the Boltzm ann p rocedure is
n ecessary for reinforced learning, how ever. A n y p rocedure that sets
the w eight-space point in m otion, p erh ap s even in ran d om m otion,
and then modifies that m otion in response to som e m easure of su c
cess or failure, counts as an instance of reinforced learning. This will
include a w ide range of possibilities.
U nsupervised learning (Hebb rules)

It is a surprising fact that som e learning algorithm s will allow a n et
w ork to learn a good deal about its environm ent w ith no error m es
sages concerning its output perform ance w h atever: all they need is a
large sam ple of inputs. W hat netw orks do w hen developing under
such algorithm s is to evolve processing strategies that (a) m axim ize
their capacity for identifying salient inform ation in the set of input
vectors, (b) convey such inform ation from layer to layer in efficiently
coded form s, and (c) find similarities am ong the inputs so that they
are taxonom ized into potentially useful groupings. Such algorithm s
partition the relevant spaces so that diversity is recognized, inform a
tion is usefully com pressed and tran sm itted , and p rototyp es are
developed.
The learning rules that perform these surprising feats are called
"H eb b ru le s," after D. O. Hebb (1949). The basic form of synaptic
ad justm ent is as follows. If a given synapse is the site of both a strong
presyn ap tic signal and a highly activated postsyn ap tic cell, then the
w eight of that synapse is increased. This proced u re is thus sensitive
to the correlations am ong the diverse elem ents of the input signals
arriving at a given cell, and it modifies the netw ork's processing

behavior so as to m agnify them .
The qualitative description just given adm its of a wide variety of
quantitative realizations, and the m any possible "flav o rs" of Hebbian
learning yield an un expected variety of cognitive results. F o r som e
settings of the relevant param eters, a Hebbian learning p rocedure
will p rodu ce units that show the sam e respon se properties found in
biological neuron s. In one striking experim ent (Linsker 1986), the in
p ut units are arranged in a tw o-dim ensional layer so as to sim ulate a
retina. Even if this netw ork is fed a series of inputs that rep resent
sheer noise, a Hebbian algorithm will modify the response properties
of the hidden units so that their "recep tiv e fields" (i.e ., that p attern of
activations across the input units w hich p roduces a m axim al response
in the h idden unit at issue) show the on-center/off-surround pattern ,
or the oriented bar pattern , found in real cortical cells (Hubei and
W eisel 1962).
A different flavor of the Hebb rule produ ces a partitioning of the
hidden-unit vector space such that the m any inputs are grou p ed into
distinct sim ilarity classes. A further flavor yields a netw ork that p er
form s a "prin cip al-com p on en t analysis" on the set of its inputs. That
is, it spontaneously finds a set of axes of representation such that
variation across the set of input vectors is m axim ized: the netw ork
b ecom es tuned to the im portant differences am ong the input vectors.
(See Linsker 1986, 1988.) This variety of results is exciting because
Hebb rules hold real prom ise of biological reality. It is already know n
that variation in synaptic w eights is driven by a Hebbian dynam ic for
a central class of cells in the hippocam pus (Kelso, G anon g, and
Brow n 1986). A nd there is evidence for a negative Hebb rule— one
that decreases the w eights of previously potentiated syn ap ses— in
oth er cells in the sam e area (Stanton and Sejnowski, in prep aration ).
F u rth erm ore, Hebbian adjustm ents are locally driven and occu r inde
pendently throu ghou t the netw ork. This allows global adjustm ents
in the w eigh t-sp ace position to happen quickly and w ithout the
guidance of any global "te a c h e r."
W ith u n supervised learning, how ever, w e are not training the n et
w ork to instantiate a specific input-output function. W h at function it
does com e to instantiate is dictated solely by the statistical properties
of the set of inputs, the connectivity of the netw ork, and the flavor of
the Hebb rule that governs synaptic adjustm ents. This limitation is
not as severe as one m ight have guessed, p erh ap s, but a creature
m ust have som e w ay of subjecting its outp u t vectors to external criti
cism if it is to learn the m any intricacies of the w orld in general. The
problem here is w ith com pletely unsupervised learning, h ow ever,
rath er than w ith Hebb rules them selves. For Hebb rules can perfectly
well subserve learning in supervised as well as in unsupervised n et
w orks. The Boltzm ann procedure described earlier, for exam ple, uses
principles of local synaptic adjustm ent that are already one instance
of a Hebb rule. Such rules also scale very well to large netw orks. All
told, Hebb rules appear to buy us speed, biological reality, and func
tional fecundity.
That functional fecundity rem ains largely unexp lored , h ow ever,
and few lessons can yet be draw n concerning the character of learn
ing and rational m ethodology in science. E ven so, one or tw o very
general lessons em erge from the preceding discussion. They are
w orth noting.
The first lesson concerns the character of the factors that drive
synaptic adjustm ent and conceptual change. A re they exhausted by
considerations of m ere “ professional in te re st," as som e theorists
have argued (see Pickering 1 9 8 1 ,1 9 8 4 ), or does the w orld itself exert a
robust influence on the process? In the m any netw ork m odels that
have ap p eared in the research p rogram under discussion, it is clear
that it is the w orld itself that is driving the learning process, w heth er
by m eans of back propagation of m easured error, by m eans of the
progressive reduction of cognitive dissonance with im posed output
vectors, or by m eans of a progressive accom odation to the objective
statistical distribution am ong inputs.
It m u st im m ediately be adm itted, of course, that these netw orks are
not functioning in a com plex social w orld, as is a real scientist. A nd it
m u st also be adm itted that for a netw ork as com plex and sensitive as
a h u m an brain, the pressure to instantiate socially acceptable func
tions can often be overw helm ing. But while the character of social
p ressu res will have a vital role to play in any adequate accou n t of
learning in scientific com m unities, there is no reason w h atever to
regard them as exhausting the dynam ical pressu res. W e know that in
nonsocial cases of learning (artificial netw orks, simple anim als), it is
the nonsocial w orld itself that is the instructor, a relentless and often
highly successful instructor. A n d unless institutionalized science
som eh ow rep resents a total corruption of a process that show s
system atic integrity elsew here, there is no reason to em brace the
extrem ely skeptical, antirealist social determ inism suggested above.
O n the contrary, science has outperform ed those p u rer but simpler
creatures.
O n the oth er hand, from the perspective of the p resen t chapter, the
naturalism of the “ strong p ro g ram " in the sociology of know ledge
(Bloor 1976) appears entirely justified. T h roughout this chap ter w e
have b een exploring causal accounts of the learning p ro cess—
accounts, m oreover, that are uniform for successful and unsuccessful

cognitive configurations alike. A nd w e have found it neither n eces
sary nor useful to fall back on the language of observation statem en ts,
logical inferences, rational beliefs, or truth. H ow ever, the alternative
to these antiquarian notions need not be a skeptical account of know l
edge, as so m an y fear. R ather, one hopes, it will simply be a better
accou n t of know ledge.
The second m ajor lesson I w ish to draw , from the picture of cogni
tion explored in this chapter, concerns the appalling vastn ess of the
conceptual space in w hich w e find ourselves. I am rem inded here of
the shock that m u st have confronted Greek thinkers w hen A ristar-
chos of Sam os first p ut a realistic m etric on the dim ensions of the
physical u niverse. W hile scholars of the period w here quite prep ared
to believe that the heavens w ere very large, A ristarchos' crude but
w ell-conceived calculations caught them quite u n p rep ared for just
how large. To a com m unity used to thinking of the scale of the
heaven s in term s of thou san d s of miles at m ost, A ristarchos b rough t
com pelling geom etrical reasons for extending the yardstick to h u n
dreds of th ou san d s of miles in the case of the M oon, millions of miles
in the case of the Sun, and m any thou sand s of millions of m iles in
the case of the stars. A n expansion of this m agnitude ch an ges one's
perspective on things.
I believe w e are now confronting a similar lesson. C on tem p orary
scholars, w ith a few notable exceptions (e .g ., D avidson 1973), are p re
pared to con ced e the possibility of alternative conceptual sch em es,
p erhaps a great m any of them . But the reckoning of the true extent of
the space of alternatives that arises from the point-in-w eight-space
m odel of h u m an know ledge is one that catches us u n p rep ared ,
h ow ever liberal our prior sentim ents, ю 100'000-000-000'000 (just-barely
discriminable) alternatives is not a num ber one w ould have picked,
and yet th at is a m inim um reckoning (see again ch apter 9, section 7).
The significance of this num ber will be clearer if w e note the follow
ing points. Simple netw orks have already been trained up on a wide
range of dram atically different problem s: ph on em e recognition,
shape discrim ination, multiplication tables, m usic com position, loan-
application evaluation, h and-eye coordination, text reading, and so
on. Each of these trained skills represents w hat w e w ould call a "d ra
matically different conceptual configuration" of the netw ork. The
num ber of such dram atically distinct skills already p rodu ced by v ar
ious research ers, m any of w hich are now available for sale, is in the
thou san d s and is still climbing. But nearly all of them are achieved in
netw orks of less than 103 units. This m eans that any one of th em is a
skill that a standard 103-unit netw ork could h ave; it needs only to be
w eighted properly.
A s w e noted, such a netw ork com m ands, at a bare m inim um , a
thou san d dram atically different possible conceptual configurations,
and it has this capacity with only 103 units and som ething like (103)2
or 106 synaptic connections. A ssum ing, as before, 10 distinct possible
values for each synaptic w eight, such a system has 10106 distinct
possible w eight configurations. This is a very large num ber indeed,
but it is a paltry fraction of the figure for a h u m an brain. In principle,
our com binatorial options are greater than the small artificial n et
w ork's by a factor of 1010I4/1 0 106 = 10<1014- 106) = 999,000,000. T h e
num ber of "dram atically different conceptual configurations" open to
us should therefore be greater than the small netw ork 's by the sam e
factor.
O n e's excitem ent at the extent of the opportunities available here is
quickly joined by a dism ay at the problem of how to explore that
space effectively. If w e suppose that one could m ake an arbitrary
change in each one of one's synaptic w eights 10 tim es every second
and did so for every second of o n e's life (1 0 10 u pdates), one w ould
still have visited a total of only l/lO^100'000'000'000'000-10Hh portion of
the functional positions available. Figures like this, and those in the
p reced in g paragrap hs, change o n e's perspective on things, for they
begin to p ut a recognizable m etric on the space. A m axim al reckoning
of any possible hum an cognitive excursion com p reh en d s but an
infinitesim al part of a minimally reckoned cognitive space.
A s w ith astronom ical space, it is clear that the effective exploration
of cognitive space will require m ajor instrum ental help. W e cannot
run fast en ou gh , jum p high enou gh , or see far enough to explore the
h eaven s w ithout technological augm entation of our native resou rces.
W e n eed m anned spacecraft, u n m anned probes, and optical and
radio telescopes. The sam e is true for the exploration of cognitive
space. O ur native resources are inadequate to the task, by m any
ord ers of m agnitude. But w e need not be limited by our native re
sources. Let us discuss the possibilities.
5 A utom ated Science
The ad ven t of artificial neural nets, and of autom ated p rocedures for
teaching them , opens the possibility of autom ating aspects of the sci
entific enterprise itself. C om p uters, of course, have been helping us
to assem ble, organize, and filter data for decad es. But teachable n et
w orks prom ise returns far in excess of these hum ble duties. For they
prom ise to do som ething conceptual w ith the data, som ething similar
to w hat intelligent creatures do w ith it. They prom ise the possibility
of effectively autom ating, for the first tim e, the theoretical p art of the
scientific enterprise.
The p ro sp ects here cover a wide range of possible achievem ents.
Let us begin w ith som e of the simplest. "E xp ert sy stem s" are now a
part of the m arketplace. These are carefully w ritten p rogram s, typi
cally diagnostic in their practical applications, that attem p t to en
capsulate and to exploit the expert know ledge available in som e
dom ain. They are regularly good enough to be useful, b ut chronically
they are p oo r enough to be frustrating. Part of the problem is that
such p ro gram s inevitably rep resent som eon e's attem p t to articulate
the available w isdom in the relevant dom ain. Such recon stru ction s
typically fall well short of the detailed expertise of a skilled profes
sional, thou gh in p rogram m ed form they do display the virtues of
tirelessness, speed, and uniform ity of treatm ent. They fall short
because m u ch of an exp ert's w isdom is inarticulate: it consists of
know ledge that is not stored in linguistic form and is difficult both
to recov er from the exp ert and to recast in the idiom of a p ro gram
m ing language (see D reyfus 1979; D reyfus and D reyfus 1986).
From the perspective of know ledge representation in neural nets,
these difficulties are not surprising. O ne's capacities for discrim inat
ing subtle and com plex patterns typically resides in a v ery high
dim ensional representation space, a space w hose individual dim en
sions each codify som e intricate feature of the input space (rem em ber
that each hidden unit receives w eighted inputs from thou san d s of
sensory units). Recovering all of this inform ation from a living expert
is effectively im possible, and exploiting all of it effectively in a serial
m achine m ight take too m u ch time even if it could be recovered .
The solution is to forget the task of trying to articulate the desired
know ledge within a set of explicit rules. Instead, train up an artificial
neural n etw ork on the sam e data set that trained the h u m an expert.
For exam p le, if m edical diagnosis is the expertise being m odeled,
then w h at is needed is a large num ber of pairs in w hich the input is
the profile of metabolic param eters and pathological sym p tom s of a
real patient, and the output is the correct diagnosis as to his disease.
H ere no attem p t is m ade to articulate rules that will con n ect com plex
sym p tom s to specific diseases. The netw ork is left to generate its ow n
"ru le s" in response to the patterns implicit in the large d ata set. W e
do, of co u rse, exploit the h um an exp ert's know ledge to provide
the diagnoses on w hich the netw ork is trained, but it is the m any
exam ples that do the w ork: no attem pt is m ade to articulate that
know ledge. W e w ish only to re-create it— and p erhaps to exceed it in
speed, ran g e, and reliability— in the trained artificial netw ork.
O nce that expertise has been achieved, w e can read out the con
figuration of w eights that sustains it and then fix those values im
m ediately into any num ber of new netw orks. The expertise can thus
be m ass-p rod u ced w ithout further training. M ore im portantly, w e
can also read out the partitions effected across the various activation
sp aces of the various layers of hidden units, in order to discover w hat
taxon om ic strategies w ere found by the netw ork as its solution to the
general problem set it. P erhaps its taxonom ies will parallel our ow n,
and p erh ap s they will not. The netw ork m ay find new groupings of
old cases, and it m ay identify, as diagnostically im portant, features
that w ent unnoticed by h um an exp erts. In this w ay m ight artificial
netw orks provide us with new insights into the taxon om y and causal
structure of the w orld, even in dom ains w e already com m and.
This approach is repeatable in a wide variety of contexts: psycholog
ical, chem ical, geological, econom ic, m eteorological, and industrial.
A nd no doubt it will be, since PD P expert system s will offer real
ad van tages over the conventional program m ed exp ert system s avail
able today. Instead of struggling to equal h u m an expertise in a speci
fic dom ain, they prom ise to exceed it in alm ost every respect.
Y et these m inor prognostications are not w h at m otivate this closing
section. They serve only to introduce the shape of the larger project.
W hat w e need to address is the problem of training netw orks to a
useful understan din g of dom ains w here h u m an experts have no
u n d erstanding, or none that is satisfactory. This will require that our
artificial netw orks use learning strategies that place negligible reliance
on an teced en t know ledge and expert teach ers, beyond w hat instru c
tion the w orld itself can provide. If w e can co n stru ct genuinely paral
lel hardw are realizations of the large netw orks that will be needed,
and if w e can autom ate such learning p roced u res so that they will
take place m any orders of m agnitude faster than they do now , either
in serial m achines or in hum an brains, then w e can turn such system s
loose on existing data sets like stellar and galactic surveys, the be
havior of national econom ies, the properties of millions of chem ical
com p oun d s, and the varieties of psychological dysfunction. P resen ta
tion of the data m u st be autom ated as well, to exploit the netw ork's
great speed . W e can then exam ine w ith interest w hat order our artifi
cial netw orks m an age to find in such com plex and teem ing dom ains.
For they will be able to explore the space of cognitive possibilities—
large subspaces of it, an yw ay— far m ore swiftly and extensively than
w e can ever hope to explore it w ithout their help.
W h at sort of symbiotic relationships m ay em erge here, betw een
existing hum an brains on the one h an d , and very large and fast n eu
ral nets on the oth er, is an engaging question that invites the im agina
tion for a ride. M aking a netw ork equal in all respects to the hum an
brain, but just faster, seem s still m uch too hard a job to be com pleted
in the n ear future. So w e m ust n ot expect to have som ething to w hich
w e can sim ply talk. If w e are willing to settle for less familiar kinds of
interactions, h ow ever, then netw orks large and fast en ou gh to be
useful seem designable and buildable right here and now . The diffi
culty will lie in m aking accessible to us the cognitive achievem ents we
m ay exp ect them to m ake. The goods will alw ays lie in the structured
partitions that em erge, in the course of learning, w ithin the hidden-
unit activation spaces. But as the netw orks get larger and the dim en
sionality of those spaces goes up, it will be progressively h ard er to
display in accessible w ays, and to m ake sense of, the stru ctu res that
develop w ithin them . For w e can expect them , after all, to develop
conceptual resources that are alien to us. Internalizing a penetrating
new fram ew ork m ay thus take som e tim e, even if it is han d ed to us
on a platter.
Even if w e can develop such turbocharged versions of, or adjuncts
to, our native cognitive capacities, the space confronting us rem ains
abyssal. A n electronic or optical realization of the neural organization
of the h u m an brain will have transm ission velocities 107 tim es faster
than axonal velocities, and this m ay allow the artificial system to learn
107 tim es faster than a hum an brain. This w ould be an im pressive
gain. But the conceptual space it could explore in a lifetime w ould still
com prise but a miniscule portion of the space available: it w ould dis
charge less than a single zero in the exp on en t of the den om in ator of
the tiny fraction discussed earlier. There is little p rosp ect, therefore,
that a "fin al, true theory of the co sm o s" is som ething w e can ever
expect to discover. N othing guarantees that any point in hum an
cognitive space is such as to yield a netw ork w ith zero error on every
p erform an ce. A n d should such a point exist, w hich I v ery m uch
doubt, it w ould still be a needle in a m onum ental haystack.
A nd y e t, if there is no real prosp ect of an end to our cognitive jou r
n ey, there is every p rosp ect that our conceptual fram ew orks can con
tinue to get better and better, ad infinitum, especially since w e can
alw ays artificially expand the num ber of neuronal units and synaptic
connections available to a given thinker, and thus expand the con
ceptual sp ace to be searched. In the long ru n , this m ay be a m ore
effective incentive to intellectual progress than the prosp ect of a final
resting place could ever be.
Chapter 12
Perceptual Plasticity and Theoretical Neutrality:
A Reply to Jerry Fodor
The doctrine that the character o f ou r perceptual know ledge is plastic, and can
vary substantially w ith the theories em braced by the perceiver, has been criti
cized in a recent paper by Fodor. H is argum ents are based on certain ex
perim ental fa cts and theoretical approaches in cognitive psychology. M y aim
in this p aper is threefold: (1) to show that Fodor's view s on the im penetrabil
ity o f perceptu al processing do not secure a theory-neutral fo u n d ation fo r
kn ow ledge; (2) to show that his view s on im penetrability are alm ost certainly
fa ls e ; an d (3) to provide som e additional argum ents fo r , and illustrations of,
the theoretical character o f all observation ju dgm en ts.
The idea that observational know ledge alw ays and inevitably in
volves som e theoretical presuppositions or prejudicial p rocessing is
an idea that has provoked m uch discussion in recent years, for its
consequences are profound. If observation cannot provide a theory-
n eutral access to at least som e aspects of reality, then our overall epis-
tem ic adven tu re contains both greater peril, and g reater prom ise,
than w e m ight have thought. The first and p erhaps the m o st im p or
tant consequence is that w e m ust direct our attention aw ay from
foundational epistem ologies, and tow ard epistem ologies that tell a
m ore global story of the natu re of theoretical justification and rational
belief. A second consequence is that our current observational ontol
ogy is just one such ontology out of an indefinitely large num ber of
alternative observational ontologies equally com patible w ith our n a
tive sen sory ap p aratus. A nd a third consequence is that, since som e
theoretical fram ew orks are m arkedly superior to oth ers, the quality of
our observational know ledge is in principle im provable. If the con
ceptual fram ew ork in w hich our perceptual respon ses to the w orld
This paper first appeared in Philosophy o f Science 55, no. 2 (Ju n e 1988). It is
here expanded by some new material on neuroscience added to section 1.2
and som e material on m eaning added to section 2.2.
are habitually fram ed w ere to be replaced by a m ore accurate and

pen etrating conception of physical reality, then our new ly-fram ed
p erceptual judgm ents could be significantly m ore revealing of the
structural properties and the dynam ical details of our perceptual
environm ent.
The m otivation for such a view is not purely philosophical. P er
ceptual psychology provides supporting evidence in the form of
experim ents designed to illustrate both the inevitable am biguity of
p ercep tu al situations and the cunning resolution of those ambiguities
at the han d s of general assum ptions im posed by "h ig h e r" cognitive
cen ters (G regory 1970a, 1974; Bruner 1973; Rock 1983). These "N e w
L o o k " ideas, how ever, have recently com e under interesting attack
from w ithin cognitive and com putational p sychology itself. The com
plaint is that these ideas have exaggerated the extent to w hich p ercep
tual processing is un d er the control of the higher cognitive centers.
A nd the counterclaim is that the job of reducing ambiguity is co n
ducted largely or entirely by peripheral "m o d u le s" w hose activities
are insulated from , and quite insensitive to, the fickle con ten t of h u
m an belief.
It is here that Jerry Fodor enters the debate. In a recent p aper
(Fodor 1984), he m arshalls the alleged m odularity of our perceptual
system s in criticism of various claims m ade by H anson (1961), Kuhn
(1962), C hurchland (1979), and others concerning the theory-laden
character of perceptual know ledge and the holistic nature of the h u
m an epistem ic enterprise. M y principal aim in this ch apter is to show
that Fod or's specific claims about the psychology of h u m an p ercep
tion are m ostly irrelevant to the epistem ological issues at stake here.
His discussion serves m ore to m uddy the w aters than to clarify them ,
for even if the m odularity/encapsulation thesis is correct— w hich
alm ost certainly it is n ot— it contains no significant m essage co n cern
ing the traditional epistem ological issues. It is, in short, a red herring.
In w h at follows, I shall try to defend and expand on the specific
claim s, listed in m y opening p aragrap h, against the several criticisms
d irected at them in Fod or's paper.
There are three principal w ays in w hich any perceptual belief m ay
fail of theoretical neutrality: in its causal history or etiology, in its
sem antics, and in the purely extensional structure of the ontology it p re
su pp oses. In his 1984, Fodor has m u ch to say on the first topic, a little
on the secon d , and he does not discuss the third. Since he does not
address w hat I have called "exten sional b ias" (C hurchland 1975a),
and space does not perm it its exploration h ere, I shall m erely em p h a
size its existence and m ove on. W h at follows will be focused on the
first tw o loci of epistem ic prejudice.
Perceptual Plasticity: A Reply to Fodor 257
1 The E tiology o f Perceptual B elief
1.1 D oes encapsulated processing buy us theory-neutral perceptions?

I shall p ass over Fod or's opening discussion in order to address
im m ediately w h at he describes as his m ain point (p. 35). Fodor,
of cou rse, is quite aw are that early perceptual p rocessing very
likely does involve m any elem ents that resem ble or corresp on d to
general em pirical "assu m p tio n s" about the w orld (e .g ., th e three-
dim ensionality of space, the spatial and tem poral continuity of com
m on objects, the sharp change of lum inance at a body's boundaries,
color co n stan cy through changing environm ents, the occlusion of dis
tant bodies by proxim ate ones, etc.), and to "in feren ces" d raw n or
"h y p oth esis selections" m ade in accordance with a system of such
default assu m p tions. O n this view , the etiology of percep tu al beliefs
looks highly, even dram atically, theoretical in ch aracter, as Fodor
him self rem arks (p. 34).
But F o d o r's view , to a first approxim ation at least, is th at (a) the
assum ptions involved in early processing are endogenously fixed in
all of u s, and (b) the processing in w hich they play a role is insulated
from any con trary assum ptions or theories— indeed, from any addi
tional assum ptions w h atever— that the perceiver m ay subsequently
com e to believe. O ur perceptual processing is thus encap sulated ; it
delivers outp uts to the higher cognitive cen ters, but it is im penetrable
to any inputs from them . The result, according to Fod or, is that all
h u m an s are fated to share a com m on perceptual experience, an ex
p erience w hose character is not subject to change as a function of any
theories w e m ay com e to em brace. There is therefore an im portant
sense, h e concludes, in w hich hum an perception is neutral vis-a-vis
the rou gh and tum ble of com peting theories. There is an unchanging
persp ective, on at least som e parts of reality, that all h u m an theorists
m u st share in com m on.
The evidence in su pp ort of these claims is tw ofold. First, Fodor
cites a n u m ber of experim ental facts that illustrate, n ot the plasticity
of percep tion, but rath er the occasional rigidity of our perceptual
d eliverances (e .g ., the persistence of certain illusions, such as the
M uller-Lyer illusion) even in cases w here w e know them to be m is
taken. S econd, he claims that if perception is to be theory d ependent
in any epistem ologically interesting sense, then the p erceptual m o d
ules m u st have “access to a l l (or anyhow , arbitrarily much) o f the back
groun d inform ation at the perceiver's disposal" (p. 35). Given the rigidity
just cited, h ow ever, he concludes that the m odules at issue lack such
access, and h ence that perception is not theory d ep en d en t in any
interesting sense.
Let us suppose, for the m om en t, that our perceptual m odules are

indeed inform ationally isolated in the fashion claim ed. That is, they
em body a system atic set of endogen ou s or genetically im planted
assum ptions about the w orld, w hose influence on perceptual p ro
cessing is unaffected by any additional or contrary inform ation.
N ow this m ay be a recipe for a certain limited consensus am ong h u
m an perceivers, but it is hardly a recipe for theoretical neutrality and it
is plain m isleading to use this latter term to describe w hat encap sula
tion m ight secure. A s conceived w ithin the relevant dialectical tradi
tion, an observation judgm ent is theory neutral just in case its truth is
not contingent u pon the truth of any general empirical assum ptions,
that is, just in case it is free of potentially problem atic presu p p osi
tions. If an observation judgm ent does have such presuppositions, its
theory-laden character will in no w ay be reduced by hardw iring those
presuppositions into the process by w hich the judgm ent is p rodu ced ,
and by closing the process to all contrary inform ation.
If everyone is a hopeless slave of the sam e hardw ired theory, then
w hat w e have is a universal dogm atism , n ot an innocent Eden of
objectivity and neutrality. The alleged cognitive im penetrability of
our perceptual p rocessing does nothing to red u ce the extent to w hich
the truth of our perceptual beliefs is contingent up on the truth of
those background empirical assum ptions or theories in w hich they
are sem antically em bedded. Encapsulation does nothing to ensure
the truth of our perceptual beliefs, not even their "tru th in g en eral" or
their "tru th und er norm al circu m stan ces." N or does it ensure their
epistem ological integrity relative to com peting interpretations of our
sen sory input. It m erely doom s us to a single point of view , a point of
view that is epistem ologically just as problem atic as any of the infinity
of oth er sets of empirical assum ptions that m ight have been h ard
w ired into us instead.
F od or's prem ises, therefore, do not buy him anything like the
theoretical neutrality of our perceptual jud gm en ts. A n unchangeable
set of prejudicial empirical assum ptions is still a set of prejudicial
em pirical assum ptions.
F od or's prem ises m ay seem to solve, at least, the problem of incom
m ensurability, by guaranteeing som e effective com m unication, at the
observation level, betw een ideologically diverse h um an theorists. But
as w e shall see at the end of this section, they fail to guarantee this
also, since rigidity in our early perceptual processing is entirely con
sistent w ith plasticity at the level of conceptual apprehension and
discursive jud gm en t. A nd despite a popular m isconception on this
point, com m unication w as never the real problem anyw ay. The epis
tem ological problem of incom m ensurable alternatives arises m ost
dearly and forcefully within a single individual, one w ho is '^ th e o re t

ica l." P utting Fod or aside for a m om en t, consider som eone w ho has
internalized tw o competing theories, and has learned two correspond
ingly different w ays of perceiving the relevant aspects of the w orld,
but is torn over w hich of these tw o global packages to ch oose. It is not
com m unication that is the problem here (he can perfectly well un d er
stand him self); the problem is theoretical evaluation and rational
choice in the absence of a neutral touchstone.
I am n ot arguing at this point that Fod or's encapsulation thesis is
false, only that it w ould not secure for us any theory-neutral founda
tion for know ledge, even w ere it true. Fod or's hardw ired con sen sus
is a sh am neutrality: it m istakes the p resu m ed universality of our
prejudice for the absence of any prejudice. A nd hardw ired or no,
that con sen sus w ould last only until the first m u tant or alien com es
along, to confront us with a different p erceptual point of view .
In fact, w e begin to becom e such m utants or aliens ourselves w hen
w e change our sensory modalities by augm enting them w ith unusual
instru m en ts, such as ph ase-contrast m icroscopes, deep-sky tele
scopes, long-baseline stereoscopes, infrared scop es, and so forth.
A nd the m etam orphosis is com pleted w hen , after years of profession
al or am ateur practice, w e learn to see the w orld appropriately and
efficiently w ith these new senses. This learning requires both that w e
suppress certain habits of processing "n a tu ra l" to the naked eye and
to the familiar w orld of m iddle-sized m aterial objects, and that w e
learn to p rocess the retinal data in novel w ays, w ays that are
appropriate to the unfamiliar features one perceives by these novel
m eans (e .g ., interference patterns, diffraction rings, dark nebulae, fu
sion planes, tem p eratu re gradients, etc.). Reflections such as these do
begin to challenge Fod or's factual claim of encapsulation o r im pene
trability. Let us therefore focus on the evidence he cites in support
of that claim.
1.2 Is the im penetrability thesis correct?

Visual illusions are good illustrations of the assum ptions involved in
early p rocessing, since the illusion is often the result of the persistent
operation of som e assum ption that is appropriate for m ost situations,
but w hich is inappropriate for the particular situation at issue. Fodor
cites the stubborn persistence of various visual illusions, even w hen
w e know th at w e are being misled, and even w here w e have the
inform ation about the inappropriate assum ptions responsible for the
illusory experience. W h y, Fod or asks, d o esn 't this inform ation affect
the w ay w e see the w orld, and thereby undo the illusion? H is answ er
is that our perceptual processing is guided by m echan ism s or
Figure 12.1
я) Schematic preparation for illusory color contrast, b) Illusory contours and brightness
contrast
assum ptions that cannot be successfully overridden by contrary

assum ptions im posed from the outside.
A first response is just to point out the great m any illusions and
visual effects w hose character show s that our visual m odules are in
deed penetrable by higher cognitive assum ptions. C onsider the wide
range of am biguous figures, such as the duck/rabbit, the o ld /you n g
w om an , the N ecker cube, and the vase/faces. Such exam ples are
am biguous w ith respect to orientation, or scale, or perspective, or
figure/ground, or any of a variety of oth er dim ensions. But in all of
these cases one learns very quickly to m ake the figure flip back and
forth at will b etw een the tw o or m ore alternatives, by changing on e's
assum ptions about the nature of the object or about the conditions of
view ing. A t least som e aspects of visual processing, evidently, are
quite easily controlled by the higher cognitive centers.
O ne such reversible illusion is striking in that it extends even to
ch an ges in perceived color. Take a m onochrom atic birthday card or
similar folded rectangle. Place it upright and oriented to the light so
that one of the inside faces is in a very slight shadow relative to the
oth er inside face. (Figure 12.1a illustrates the relevant configuration,
b u t only a real card will support the illusion.) D espite this slight sh a
d ow , the tw o faces of the card will be perceived as having the sam e
objective color. N ow , closing one eye to defeat stereoscopic orienta
tion cu es, treat the object as a N ecker cube and deliberately invert its
orientation— in thou ght— so that the m iddle fold appears closer to
y ou than the tw o outside edges. This will p rodu ce an obvious distor
tion in the perceived shape of the card: it will no longer look like a
folded rectangle. A nd it will also produce a change in the perceived
color of the shadow ed and unsh adow ed areas of the card. In its ori
ginal ap p earan ce, the slight contrast in lum inance is su p pressed by

the visual system as a m ere shadow effect. But in the card 's inverted
configuration, the slight contrast in lum inance is no longer consistent
w ith a sh ad ow hypothesis, and the contrast betw een the tw o areas is
robustly interpreted as a sharp difference in their intrinsic colors. (I
ow e this exam ple to Richard G regory.)
Illusory contours provide a similar but contrasting exam ple. The
w hite background in figure 12.1b is, of cou rse, entirely uniform . But
m ost of us can see a slightly brighter triangular figure interp osed be
tw een us and the three black circles, a figure w ith distinct rectilinear
con tou rs m arked by a sharp change in lum inance, even in the gap
b etw een the black circles. H ere the eye-brain conjures up lum inance
differences w here in reality there are none. A nd again, the illusion is
penetrable and reversible. Tell yourself that the circles have w edges
cu t out of them ; see the elem ents of the diagram as six ind ependent
objects artfully arranged against a uniform background; cen ter you r
attention on the tw o p rongs of any V; and the illusory contours
disappear.
These assem bled exam ples com pile a w ide range of elem ents
central to visual percep tion — contour, con trast, color, orientation,
d istance, size, shape, figure versus grou n d — all of w hich are cogni
tively penetrable. Collectively, they constitute a strong case against
F o d o r's claims of im penetrability for our p erceptual processing.
But p erh ap s I am gathering evidence selectively or aim ing it at an
exaggerated version of Fod or's view . P erhaps m any other elem ents of
p erceptual processing, even the dom inant share, are im penetrable,
despite these exam ples of a contrary cast. W h at exam ples does Fodor
cite, th en , in sup p ort of such a claim?
O nly one, the M uller-Lyer illusion (figure 1 2 .2a), th ou gh the class
he has in m ind is clear enough (it will include the Ponzo illusion, the
H ering illusion, and similarly persistent illusions). The M uller-Lyer,
h ow ev er, is an odd exam ple for Fod or to be using, b ecause the "te x t
book sto ry " on how it w orks (a story apparently en d orsed by Fodor,
p. 33) explains it as the effect of our having learned, in judging abso
lute size, to m ake autom atic corrections for the variation of an object's
angular size w ith distance (fig. 12.2b.) That is, the illusion exists in the
first place only because the relevant processing m odule is the well-
trained victim of som e substantial prior education— that is, of som e
pen etration by cognitive activity. The Ponzo and the H ering illusions
m ay h ave a similar origin. A ccordingly, they are all of them poor
exam ples on w hich to base a general claim of im penetrability.
N ow I will grant that, its cognitive origins aside, the M uller-Lyer
illusion cannot be overridden by any casual, fleeting, "v o lu n ta ry "
a) b)
Figure 12.2
a) The Muller-Lyer illusion, b) The Muller-Lyer illusion in a realistic setting (bold, ver
tical lines)
attem p t to m odify the character of on e's visual experience. By itself,

h ow ever, this m ean s relatively little, for the issue is not w heth er
visual processing is in general very easily or quickly p enetrated by
novel or contrary inform ation; the issue is w h eth er in general it is
penetrable at all, w here the acceptable m ean s of penetration can in
clude long regim es of determ ined training, practice, or conditioning.
If the M uller-Lyer illusion is an incidental consequence of a long
period of perceptual training on certain typical kinds of perceptual
problem s, then presum ably a long period of training in an environ
m en t of a quite different perceptual ch aracter w ould produ ce a
subject free from that particular illusion. Fod or, it seem s to m e, is
in no position to insist otherw ise, especially given exam ples of the
following kind, w hich are not speculative, b ut real.
Recall the effects of chronically w orn inverting lenses on the visual
p ercep tion of norm al hum an s. Such lenses h ave the effect of invert
ing the orientation of all visual inform ation relative to the body's
tactile and m otor system s. In short, they turn the visual w orld
upside d ow n. (Kottenhoff [1957] provides a useful sum m ary of this
research .)
The initial effect is profoundly disorienting, but w ith little m ore
than a w eek's practice, subjects adjust to the new perceptual regim e.
The subjects are not confined to a chair or bed for the duration of the
exp erim en t, but are forced by practical necessity to continue to
interact w ith familiar objects and to en gage in the norm al form s
of m otor behavior. The result is that the subjects slowly m anage to
recoordinate their vision with the rest of their sensory an d m otor

system s, and the illusion of the w orld's being upside d ow n is said to
fade aw ay, all on a time scale of roughly a w eek.
W h en the lenses are first p ut on and the w orld is m ade to appear
upside d ow n , the subjects are of course quite aw are of w h at the
lenses are doing. They m ay even know how they do it. But the illu
sion is n ot banished by the m ere possession of this inform ation. It
w ould clearly be w rong, how ever, to draw from this any conclusion
about the im penetrability of our visual p rocessors. A few w eeks of
steady practice and experience penetrates them quite nicely. A nd the
degree to w hich that penetration is successful is fu rther revealed
w hen the lenses are finally rem oved : for a short time thereafter,
the subjects suffer a disorientation illusion v ery m uch like that en
cou n tered w hen the lenses w ere first put on. Their visual processing,
reconfigured by training to com pensate for the lenses, continues to
"co m p e n s a te " after the lenses are gone.
In similar experim ents on anim als, training produces a reversal in
the ch aracter of w hat one m ight have p resu m ed to be en d ogen ou sly
specified reflexes, such as the vestibulor-ocular reflex, w hich directs
on e's eyes, w hen fixated on a target, to m ove an appropriate am ount
to the left or right in order to com pensate for head m ovem en ts in the
opposite direction. H ere the brain seem s literally to rew ire the rel
evant neural m echanism un der the p ressu res im posed by left-right
inverting lenses, (see G onshor and Jones 1976.)
C ases like these are im portant, for they reflect the plasticity of som e
v ery d eep "assu m p tio n s" implicit in visual processing, su ch as the
specific orientation of the visual w orld relative to o n e's oth er sense
m odalities and to on e's m otor system s. If assum ptions as deep as
these can be reshaped in a w eek or tw o, then our perception begins to
look v ery plastic and very penetrable indeed.
I exp ect F od or to object, how ever, that exam ples such as these,
dram atic though they m ay be, are n ot cases of the cognitive p en etra
tion of our peripheral m odules. These perceptual ch an ges are
w rou gh t n ot by the simple acquisition of certain beliefs, n or by re
flecting on them in the relevant perceptual circum stances. R ather,
they are w rou gh t by som e form of training, practice, or conditioning,
often lengthy.
O ne w ay to turn this objection aside is to attack the integrity of the
highly questionable dichotom y betw een "cogn itive penetrability"
and o th er form s of penetrability (see the com m entaries on Pylyshyn
1980 in th at w ork). But I shall not pursue this path here. There is a
simpler an d m ore direct respon se: W ho ever claim ed that the ch arac
ter of a scientist's perception is changed simply and directly by his
em bracing a novel belief? N one of the theorists cited in Fod or's p aper
have defended such an unrealistic view . A nd all of u s, at som e point
or oth er, have em phasized the im portance of long familiarity w ith the
novel idiom , of repeated practical applications of its principles, and of
socialization w ithin a like-minded group of research ers.
K uhn is quite explicit (1962, chapters 5 and 10) that the enveloping
p aradigm that shapes the scientist's perception is n ot constituted
solely by a set of explicit law s, but by an entire disciplinary m atrix that
includes standard w ays of applying and using the resources of the
p aradigm , skills acquired during a long apprenticeship. A nd m y ow n
d iscussion of the plasticity of perception (1979, ch apter 2) has the rel
evant com m unity learning their nonstan dard observational vocabul
ary from birth, in an ongoing practical setting w here no oth er idiom is
even contem plated.
I confess to having used one exam ple w here a tem p orary shift in
p erception can be m ade fairly swiftly: the exam ple of reperceiving the
organization of the solar system in a heliocentric rath er than a geocen
tric fashion (1979, pp. 3 0 -3 4 ). This case is rath er closer to the N ecker
cube in ch aracter than to the case of the inverting lenses. But even
here it w as em phasized that simply having the relevant C opernican
beliefs is not enou gh ; one m u st learn how to see the changing
h eaven s as an unfolding instance of the C opernican organization, as
view ed from our peculiar perspective within it. H aving the relevant
beliefs is one thing: w e are all of us C opernicans, after all. Reshaping
on e's perception is quite another.
The point is a general one. A physics stud en t does not com e to see
the m otions of com m on objects in a new w ay sim ply by m em orizing
N ew ton 's three laws. M ost freshm an physics students do m em orize
those law s, but relatively few have their perceptions m u ch altered.
The few w ho do are distinguished by having practiced the skills of
applying those laws in a wide variety of circum stances. They do com e
to perceive a com m on pattern in the behavior of m oving bodies that
w as hitherto invisible to them , but m em orizing the law s w as only the
first step in a fairly lengthy p rocess. There are sudden flashes of in
sight, to be sure, as w h en one first grasp s how the p attern is instanti
ated in som e typical case. But on the w hole, the process of reshaping
o n e's perception takes time, and it requires m ore than the m ere ad op
tion of a belief or three.
To sum m arize these points, if Fodor is attacking the view that p er
ceptual processing alw ays (or even usually) respon d s directly and im
m ediately to changes in on e's theoretical com m itm ents, then he is
attacking a straw m an. This is not a view that an yon e has defended.
O n the oth er h and, if Fodor is denying that perceptual processing is
plastic in the face of m ore com prehensive and protracted kinds of

p ressu res, such as the forced practical use of som e novel perspective,
then the em pirical facts are against him . For by these m ean s, even
v ery basic aspects of visual processing can be overturned and recon
figured, as w e saw w ith the visual-inversion experim ents.
Som e d egree of "d iach ro n ic" penetrability is grudgingly conceded
by Fod or (p. 39), since the alternative is to hold that all of o u r adult
p erceptual capacities are endogenously specified. W e know th at they
are not, since the developm ent of so-called "n o rm a l" percep tion itself
plainly involves a great deal of learning on the part of the grow ing
infant. O u r perceptual, practical, and social environm ent sh ap es our
perceptual capacities mightily, especially in their early stages of de
velop m en t, and this suggests that different courses of learning w ould
produce interestingly different perceptual capacities. Fod or attem pts
to play d ow n this concession, h ow ever, by suggesting that the range
of possible variation in perceptual developm ent m ight be quite
n arrow .
W hy h e thinks this is left unexplained. The claim needs arguing,
the facts su ggest otherw ise, and one need not turn to academ ic jour
nals for shining exam ples. To see the nonstan dard p erceptual capaci
ties that o u r native m odalities can acquire, think of the following. In
recent centuries m ost hum ans have learned to perceive sp eech not
just auditorally but visually: w e have learned to read . A n d som e have
learned to perceive speech by touch: they read Braille. A n d som e of
us have learned not just to h ear m usic, but to see it: w e have learned
to sight-read m usical notation. N ow , neither the eyes nor the fingers
w ere evolved for the instantaneous perception of those com plex
stru ctu res and organizations originally found in auditory ph e
n om ena, but their acquired m astery here illustrates the highly
sophisticated and decidedly supernorm al capacities that learning can
produce in them . A nd if these capacities, w hy not others? D iachronic
p en etration, I assert, is not only possible and actual; it is com m on
place.
Finally, there is neurophysiological evidence that su ggests the sys
tem atic penetrability of the peripheral m odules by the h igher cogni
tive cen ters. Cell-staining techniques have allowed us to trace out a
gross "w irin g d iagram " for m any parts of the brain. W h en introduced
into a neuronal body, certain chem ical stains— notably, h orseradish
p eroxid ase— are tran sp orted dow n the entire length of its long fiber
like axon. This m arks the axons visually, and the paths they trace
through successive sections or slices of the brain can then be followed
with an optical m icroscope. In the case of vision, for exam ple, the
dom inant nervous path w ay starts at the retina, and p roceed s via the
optic nerve to the lateral geniculate nucleus (LG N ), and stepw ise
from there by other pathw ays to the prim ary visual cortex, to the
secon d ary visual cortex, and from there to a variety of other areas
even higher in the p rocessing hierarchy.
But these “ ascen d in g" p ath w ays are alm ost invariably m atched by
"d e scen d in g " p ath w ays that lead us stepw ise back through the in
term ediate brain areas and all the w ay out to the earliest processing
sy stem s at the retina. The descending projections from the visual cor
tex back to the LG N , for exam ple, are even greater in num ber than
those in the ascending direction. A nd though the claim is not well
established, there is som e evidence that fully 10 percen t of the axonal
fibers in the h um an optic nerve are descending projections from the
LG N back out to the retinal surface itself, the very first tran sd u cer in
the processing hierarchy (W olter 1965; W olter and Lund 1968; Sacks
and Lindenberg 1969).
T here are similar chains of d escending p ath w ays— from the v ar
ious areas top m ost in the inform ation-processing hierarchy, dow n
throu gh all of the interm ediate processing stages, and all the w ay out
to the periphery— for all of the oth er sen sory m odalities as well. This
organizational pattern is typical in m am m als and also in birds (Living
ston 1978, pp. 4 5 -4 9 ). Prim a facie, the function of these descending
p ath w ays is “ centrifugal co n tro l." They allow for the m odulation of
low er level neural activity as a function of the dem an d s sent dow n
from levels higher in the cognitive h ierarchy. Experim entation on
their functional significance is so far lim ited, but lesions confined to
the d escending optic nerve p ath w ays (from LG N to retina) are know n
to cause perceptual deficits in birds, even though the descending
fibers in their case constitute only 1 percen t of the optic nerve total.
Lesioned birds are less able than intact birds to distinguish edible
seeds from oth er m inute objects in dim light (Rogers and Miles 1972).
If such descending path w ays w ere alw ays sharply confined close to
the sen sory p eriphery, or if they w ere to be found scattered only here
and there in the inform ation-processing hierarchy, then w e m ight
have som e realistic hope of dism issing any backw ard loop as an ele
m en t of w hat is still an "en cap su lated m o d u le" from a functional
point of view . But descending p athw ays are the rule in the processing
hierarchy of the brain, not the exception. They app ear to con n ect the
u p p er levels in the hierarchy to m ost and perh ap s to all of the low er
o nes, in each and every one of the sen sory m odalities. In sum , the
w iring of the brain relative to its sen sory p eriphery certainly does not
su ggest the encapsulation and isolation of p erceptual processing. A s
w ith the psychological data discussed earlier, it strongly suggests
exactly the opposite arrangem ent.
[A dded in 1989: Tw o new pieces of neuroscientific evidence have

recently em erged that b ear on the questions of both plasticity and
theory-lad en ness. The first piece of evidence is theoretical and d e
rives from the new connectionist m odels of inform ation p rocessing in
the brain. Those m odels identify the general know ledge acquired by
any organism w ith the acquired configuration of its m yriad synaptic
w eights. Since on this m odel all cognitive processing, including p er
ceptual p rocessing, consists in vector transform ations at the han d s of
those m odifiable and m u ch instructed w eights, it w ould seem that all
p ercep tu al processing is inescapably laden w ith the legacy of general
know ledge shaped by p ast experience. (For a m ore detailed discus
sion, see ch ap ter 9, section 7.)
The second piece of evidence is experim ental and co n cern s the
functional plasticity of auditory cortex. Sur et al. (1988) induced the
axons in the optic nerve of neonate ferrets to project into the anim als'
auditory p ath w ay (the m edial geniculate nucleus or M GN) instead of
to their norm al visual path w ay (the LG N ). The result is an animal
w hose au ditory cortex is now driven exclusively by inform ation sent
from the eyes. Such anim als do develop significant visual function as
they m atu re, and recordings from cells in the "a u d ito ry " cortex of
adult anim als show the cells to have developed the sam e directional
sensitivity, orientation selectivity, and capacity for edge detection dis
played by cells in the visual cortex of norm al anim als. This striking
result su ggests that the processing characteristics peculiar to our
adult sen sory system s are not endogenously specified, as F od or's pic
ture invites us to suppose, but rath er are developed over tim e in a
highly plastic system that is shaped by the long-term characteristics of
the sen sory input they receive from the periphery. This does n ot sit at
all well w ith a picture of endogenously specified assum ptions unique
to each m odality. This result is also exactly w h at one should exp ect if
the connectionist m odels of learning and inform ation p rocessing just
m en tion ed have any integrity.]
1.3 Is the encapsulation thesis relevant?

Before concluding this section on the etiology of p erceptual jud g
m ents, I w ish to address a further and vitally im portant point. L et us
su p p ose, for the sake of argum ent, that perceptual processing is en
tirely rigid and im penetrable up to the contents of on e's visual (audi
tory, tactile, etc.) m anifold; rigid, that is, up to the ch aracter of on e's
sensations. E ven if, as now seem s very unlikely, visual p rocessing is
thus rigid, the outputs of that system are still capable of driving in
turn an en orm ou s variety of quite different conceptual fram ew orks.
The p oint here is that sensations them selves are not yet truth-
valuable or sem antically-contentful states: they are still a stage,

thou gh p erhaps a late stage, in the processing that leads to specific
p erceptual judgm en ts or beliefs. N ow (and this is a point that I have
m ad e explicitly before in 1979, pp. 3 8 -3 9 ), h ow ever rigid that prior
p rocessing m ight be, there are indefinitely m an y different possible
m appings from the dom ain of sensations to the dom ain of proposi
tions (judgm ents, beliefs), and w hich of these m any m appings com es
to characterize you r ow n perceptual activity is a function of w hich of
the indefinitely m any conceptual fram ew orks you have learned as the
fram ew ork of spontaneous response to the contents of you r sensory
manifold.
A ccordingly, the plasticity of perceptual jud gm en t defended by m e
in earlier w ritings does not require that w e "p e n e tra te " the peripheral
p erceptual m odules in any case. W e need only connect the o utp u ts of
those m odules to w hatever system of conceptual activity governs our
discursive thinking. Furth er, if tw o people have learned radically
different fram ew orks, then they will have a severe com m unication
problem despite the rigidity of their peripheral processing. Thus m y
earlier observation (section 1.1) that encapsulation fails to solve the
problem of incom m ensurability.
That our conceptual system is plastic I regard as obvious. That w e
can successfully connect different conceptual system s to one and the
sam e sen sory system I regard as only slightly less obvious. The
d octrine of the plasticity of hum an perceptual judgm ent requires no
p rem ises beyond these.
A few w ords, then, in support of the less obvious of the tw o p rem
ises. C onsider the conceptual fram ew ork used for describing pitch
in m usical theory. It begins w ith the chrom atic scale: C, C # , D, D # ,
E, F, F # , G, G $ , etc. This is not just a list of n am es. The sequence has
a periodic ch aracter (octaves), an absolute position in auditory space,
and a well-defined m etric of various intervals. In fact, the chrom atic
scale and its various properties form the foundation of m usical
theory. Clearly, h ow ever, this conceptual fram ew ork is not innate to
our auditory processing, nor is it a part of ordinary language. But
people are regularly trained to use it in auditory perception. In tim e,
the better students m aster w hat w e call a sense of absolute pitch
("T h a t's a m iddle C , . . . and th at's the A above m iddle C ").
M ore intricate y et, there is the dom ain of m usical ch ord s, and of
harm onious sequences of chords. C hords are structured sets of simul
taneously sounded n otes, sets that fall into an organized m atrix of
different types (m ajors, m inors, seven th s, ninths, dim inisheds,
au gm en ted s, etc.). These too can be directly recognized, by ear, by
one suitably practised in the relevant theory and vocabulary. Such a
p erson p erceives, in any com position w hether great or m u n d an e, a

stru ctu re, developm ent, and rationale that is lost on the untrained
ear.
W e are contem plating a m usical exam ple n ot because it is the only
em pirical exam ple one can cite, but because it is an unproblem atic
exam ple. E veryone know s that the "e a r can be tra in e d ," as w e say, to
sustain these rem arkable and nonstandard perceptual capabilities.
But the exam ple of trained m usical perception is a straightforw ard
existence proof for the possibility of theoretically-transform ed p ercep
tion in general. W hat w ants appreciating is that this exam ple is re
peatable in a great m any other dom ains. W h at is required is learning
the relevan t theory, and extended practice in using it.
W e m ay begin to see som e of the endless possibilities by noting that
one can just as easily learn to recognize sounds un d er their dom inant
frequ en cy descriptions as un d er their m usic-theoretic descriptions
("T h at is an oscillation of 262 h e r t z ,. . . and th at's one of 415 h e rtz ").
Equally possible, one can learn to recognize them u n d er their
w avelength descriptions ("T h at has a w avelength of 1.19 m eters,
. . .a n d that has a w avelength of 0.75 m e te rs"). The payoff is that
on e's sp on tan eou s perceptual judgm ents then put one in a position
to anticipate, m anipulate, and exploit the details of such auditory
p h en om en a as interference effects, standing w aves, d oppler shifts,
intensities, and so forth. O ne requires only a facility w ith a few
elem en tary law s of w ave propagation. (For exam ple, recall the sound
m ade by a m oving car as heard by the pedestrian it p asses:
Z E E E E E EY ow w w w w w w . Suppose you can hear that the dom inant
frequency of its approaching hu m (Z EEEEEE) is 262 hertz (a m iddle
C ), and th at its receding h u m (Yow w w w w w ) is 220 hertz (a low er A).
In such a case you m ay safely infer that its unshifted, or intrinsic,
frequency m u st be roughly halfw ay betw een the heard extrem es, or
about 241 h ertz. Since the heard frequency is D oppler-shifted from
this value by about 21 hertz, w hich is about 9 p ercent of th e intrinsic
frequency (241 hertz), then the velocity of the car m u st also be about 9
p ercent of the velocity of sound (740 m p h ), or about 65 m p h . So, if
y ou learn to recognize sounds under their frequency descriptions,
then the velocities of u nseen objects are often but a quick inference
aw ay. Exam ples like this can be m ultiplied indefinitely.)
I conclude this section w ith an instructive fable. C onsider the im
aginary com m u n ity, discussed at length in m y 1979, w hose m em bers
all have the unquestioned assum ption that physical objects contain
an observable fluid substance, called "c a lo ric," that is confined in
com m on objects un d er a variety of different p ressu res. A ccord in g to
ev eryo n e's "co m m o n sen se" convictions, caloric is p rodu ced or re
leased in great quantities by fires and by friction; it alw ays flows from
high-pressure bodies to low -pressure bodies; a sufficiently high
pressure causes the boiling of w ater; a sufficiently low p ressure
cau ses w ater to freeze; and so forth.
M ost im portant, all m em bers of this com m unity regard the p res
sure of caloric as an observable feature of the w orld: w here y ou , upon
touching a sim m ering kettle, have the sp on tan eou s perceptual belief
that this kettle is hot, they have the sp ontaneous perceptual belief that
this kettle has a high caloric flu id pressure. In sum , w e have here a com
m unity using the conceptual fram ew ork of early classical th erm o
d ynam ics as a com m onsense observation fram ew ork for that sam e
range of p hen om en a com m only ad dressed by us with "folk
th erm o d y n am ics"— the familiar fram ew ork of hot and cold. The
virtues of this caloric fram ew ork, and its persistence in the face of
criticism , will not be repeated here, but they are considerable.
I cite this exam ple because the people of this com m unity are m ak
ing sp ontaneous "ob servation al" judgm ents that are obviously laden
w ith theory. M oreover, the theory at issue is know n by us to be false,
and so the prejudicial character of all of their p erceptual beliefs in this
area is m ade even m ore dram atic. C onsider now a philosopher of this
society, Jerry Caloric, w ho argues as follows.
"C o n sid er the illusion produced w hen on e's left hand is allowed to
rest in a bucket of w ater at high caloric p ressu re, and on e's right hand
in a bucket at low caloric pressu re, and then both hands are im m ersed
in a bucket at an interm ediate caloric p ressu re. If you judge w ith the
left h an d , this w ater has a low caloric p ressu re; if you judge w ith the
right, it has a high caloric pressure.
"O f cou rse, both w e and G ranny know how this illusion is p ro
duced. The nerves of the tw o han ds becom e differently fatigued by
the extrem e caloric pressures in the first tw o buckets, and thus each
gives a different and false response to the interm ediate pressure of
the third bucket. But notice that possession of this inform ation does
absolutely nothing to dispel the illusion.
"W e m ay conclude, therefore, that our peripheral m odules are cog
nitively im penetrable. A ccordingly, our perceptual judgm ents about
the caloric-fluid pressures of com m on objects are in an im portant sense
theory neutral. The theories w e em brace have no effect on caloric
perception , and all hum ans w ith norm al perceptual system s will thus
perceive the w orld in exactly this sam e w a y ."
I have here recreated the form of F od or's argum ent in a setting
w here the conclusion is clearly false. The point is to highlight som e of
the w ays his argum ent fails in its original setting. The first lesson is
that Fod or, like the philosopher just quoted, fails to appreciate the
highly system atic and speculative character of his ow n observational

idiom s, a character they will have quite independently of any rigidity
in our peripheral m odules. A nd the second lesson is that F od or never
takes seriously the possibility that, even given the rigidity of p ercep
tual p rocessing up to the character of our sensations, on e can still
train oneself to use, in spontaneous "ob serv atio n al" m o d e, con cep
tual fram ew orks radically different from those w e learned at m oth er's
knee.
2 The Sem antics o f O bservation Predicates
Implicit in the preceding rem arks is the view that the m eaning of an
observation term derives not prim arily, nor even p erhaps at all, from
the typical etiology of its observational application, but rath er from
the n etw ork of general beliefs and assum ptions in w hich it is em
b ed d ed. Because the contents of such em bedding netw orks can vary
substantially, so also can the m eaning of our observation term s.
Fod or correctly identifies this approach to m eaning as a m ajor ele
m en t in m y argum ent for the theory-laden character of perceptual
jud gm en t (I call it "th e conceptual-role theory of m ean in g " or "th e
n etw ork ap p ro ach "; he calls it "m ean in g h olism "). The argu m en t is
simple and quickly stated.
(1) A n y jud gm ent consists in the application of concepts (e .g ., a

is F ).
(2) A n y con cep t is a node in a n etw ork of concepts w h ose con
necting threads are sentences, and its m eaning or sem antic
identity is determ ined by its peculiar place in that netw ork.
(This, in stick-figure form , is the theory of m eaning referred
to.)
(3) A n y netw ork of concepts is a theory, m inim ally, a theory as to
som e of the classes into w hich nature divides itself, and som e
of the relations that hold betw een them .
.'. (4) A n y judgm ent presup p oses a theory.
.'. (5) A n y observation judgm ent p resu p p oses a theory.
The theory-lad en ness of observation term s thus em erges as a con

sequence, n ot of their having som e special and regrettable disease,
but sim ply as a consequence of their being m eaningful term s at all.
2.1 O bjections to the netw ork approach: Fodor's reductio

The defect Fodor finds in an unqualified netw ork ap p roach to m ean
ing is th at it allows too m uch leew ay in w hat an observation sentence
m ight m ean. "S o Churchland holds, on holistic grounds, that an

observation sentence m ight m ean anything depending upon theoreti
cal con text. I em phasize that this conclusion is equivalent to the claim
that anything m ight be an observation sentence depending upon theoret
ical con text; or, in m aterial m ode, that anything m ight be observed de
pending upon theoretical co n text" (Fodor 1984, p. 28). This tracing of
p resum ed equivalences overextends itself. I do not hold that, given
norm al hum an senses, anything m ight be observed by us. W e cannot
observe w hat does not exist, and w e cannot observe (w ithout in
strum ental help) w hat is beyond any physical detection by our native
senses. O n the other h and, I do assert that alm ost any predicate could
function as the vehicle of spontaneous perceptual judgm ent for som e
one trained to conceive of things in the relevant w ay. But w heth er
his "o b serv atio n " judgm ents constitute genuine cases of veridical
percep tion will be a function of w h eth er the feature he takes himself
to be observing really exists, and w heth er his sensory system has
som e reliable discrim inatory response to the occurrence of that fea
ture. Failing either of these conditions, his "o b serv atio n " judgm ents
will be system atically m istaken, as in the case of the Friends of C alor
ic. But while m istaken, those judgm ents m ay still be highly su ccess
ful, both from a practical and from a theoretical point of view .
In sum , m y position entails that w e can observe m any features of
the w orld quite different from the features w e are used to observing,
and that w e m ight not really be observing som e of the features
that w e think w e are. But it does n ot entail that w e can observe
everything.
2.2 B elief netw orks versus causal connections

The p receding is n ot Fod or's m ain w orry, h ow ever, and w hat looked
like an attem p t at a reductio m ay be just a rhetorical flourish. W h at he
really seem s to object to is the idea that the m eaning of observation
term s m ight have nothing to do w ith the objective features of the
w orld that typically elicit or cause their sp ontaneous use. A nd he cites
the possibility that at least som e of an observation term 's sem antic
properties m ight be determ ined nonholistically, p erhaps by the causal
connections just alluded to. H e then concludes, "In light of this, I
propose simply not to grant that all the sem antic properties of
sentences/beliefs are determ ined by their theoretical context. A nd
G ranny proposes n ot to grant that to o " (p. 30).
This is not good enou gh, for tw o reason s. W e do not require
F od or's concession that all of the sem antic properties of sentences or
beliefs are determ ined by their theoretical con text. So long as som e of
the sem antic properties of any observation sentence are inevitably
d eterm ined in that fashion, such sentences will still be stuck w ith a
significant burd en of prejudicial theory. To achieve a truly theory-
neutral foundation for know ledge, Fod or needs a class of sentences,
or term s, none of w hose sem antic properties is dep en den t on theory.
Second, there are decisive reasons in support of the claim that at
least som e of any observation term 's sem antic properties m u st be
determ ined by the netw ork of beliefs that em beds it. C onsider the
following argum ent.
If a term 'F' is to be a m eaningful observation term , then its p re
dication in ‘F a’ m u st have som e m aterial consequences: it m u st im ply
som e fu rther sentences, it m u st be incom patible w ith som e others,
and so forth. The sentence 'Fa' will clearly have this p rop erty if it is
asserted in a con text w here general sen ten ces such as '(x)(Fx=>Gx)',
‘(x)((Fx & Hx)=> ~ K x )', and so forth, are already assu m ed . 'Fa' will
then im ply 'Ga', be incom patible w ith '(Ha & K a)', and so forth.
But if 'F' figures in no such background beliefs or assum ptions
w h atsoev er, then 'Fa' will be entirely w ithout consequence or sig
nificance for anything. It will have no bridges to link its assertion or
denial w ith the assertion or denial of any other sentence. It will be a
w heel th at turns nothing, a coin w eightless in every balance, an
assertion em p ty by any m easure. Less figuratively, its assertion will
be com putationally inert. It will be w ithout com putational significance
for the v ery cognitive system that asserts it.
M eaningful observation term s, therefore, will always be em bedded
w ithin som e set of assum ptions. A nd since there is no analytic/
synthetic distinction, those assum ptions will alw ays be speculative
and corrigible. M eaningful observation term s, w e seem bou n d to co n
clude, will alw ays be laden with theory.
It will be pointed out, of course, that, even in the absence of any
b ackground assum ptions, 'Fa' will have a h ost of purely fo rm a l co n
seq u en ces, such as T v ~ P ', 'Q=>Fa', and so on. But the p attern of
these trivial consequences is exactly the sam e for 'Fa' as it is for any
oth er putative observation sentence: 'Ga', 'Ha', etc. These con se
quences thu s cannot serve to bestow any distinct significance on 'Fa'.
It is the m aterial consequences of 'Fa' (i.e., the ones that flow from
substantive, or nonform al, background assum ptions) that do that.
W hich is an oth er w ay of stating the central claim of the netw ork
theory: w h at determ ines the m eaning of any term is the peculiar clus
ter of beliefs in w hich the term figures, and the peculiar p attern of
inferences they m ake possible.
G iven the clear inevitability of an ideological com p on en t in the
m ean ing of any observation term , one m ay begin to w on d er at the
relative contribution of that com ponent as against an oth er possible
com p onent of m eaning, a causal com p onen t. I have criticized causal

accou nts of m eaning elsew here (C hurchland and Churchland 1983;
C hurchland 1986e), so I shall here restrict m yself to som e brief
rem arks and illustrations.
C onsider again the Friends of Caloric discussed earlier. In their
spon tan eou s use of the vocabulary of caloric theory, they are re
sponding to exactly the sam e feature of the objective w orld to w hich
w e respon d w ith the vocabulary of 'h o t', 'co ld ', etc. But our resp ec
tive observation sen ten ces— 'That is hot' versu s 'That h as a high
caloric-fluid p ressu re'— certainly do not m ean the sam e thing. They
are com m itted to radically different ontologies and to system atically
different behaviors. The alien's understan din g of "th erm al p h e
n o m en a" is very different from ours. W h at w e have, then, is tw o
expressions that are typically elicited by the sam e cause, but w hich
differ w idely in their m eaning.
W e can also argue for the com plem entary conclusion. Consider the
m eaning of the "o b serv atio n " vocabulary of the Friends of Caloric, as
u sed by them (1) in this w orld, and (2) in a different possible w orld in
w hich there really is a fluid substance that an sw ers to their collected
beliefs, a fluid w hose changing p ressure causes all the familiar sen sa
tions in them , a w orld, in short, at w hich the theory of caloric-fluid is
true. N ow , in our w orld the spon tan eou s application of their term
'high caloric-fluid pressu re' is typically cau sed by the high m ean
m olecular kinetic energy of the object felt. In the other possible
w orld, its application is typically cau sed by a high caloric-fluid p res
sure. But that expression, along w ith the entire theory that em beds it,
m ean s exactly the sam e thing in that w orld as it does in this. The only
difference is that in the other w orld the Friends of Caloric are m ore
fortunate: in that w orld their beliefs about caloric— the very sam e, or
sem antically identical, beliefs— are all true!
Tw o paragraphs ago, w e had the sam e cause of sp ontaneous ap
plication, b u t different m eanings. In the last p aragrap h , w e h ave the
sam e m eanings, b u t different cau ses of sp on tan eou s application.
M oreover, w hat covaries with m eaning in the first case is the su r
rounding netw ork of beliefs. A nd w hat is cocon stan t with m eaning in
the second case is the surrounding netw ork of beliefs. The m eaning
of an observation term , I therefore conclude, has nothing w hatever to
do w ith the typical cause of its spontaneous application. Its m eaning
is fixed by other factors entirely: specifically, by the surrounding net
w ork of beliefs.
This m axim al claim that ideology com pletely exh au sts m eaning is
n ot n eed ed to resist Fod or's position, but I am strongly inclined to
defend the claim in any case. The argu m en t just given form s part of
the reason. The robust causal connection betw een a high m ean
m olecular kinetic energy and the alien's application of the term 'high
caloric-fluid p ressu re' contributes nothing that I can discern to the
sem antic conten t of the quoted expression. N either does it provide a
reference for that expression (it is quite em p ty of reference), nor does
it guarantee the truth of any of its typical applications (they are all
false). All th at causal connection does is prom pt the aliens spon
taneou sly to deploy the expression at issue. Its m eaning, its reference,
and the truth of its applications (to the extent they have any) arise
from oth er sources entirely.
A n oth er reason for looking skeptically at causal accou n ts of m ean
ing arises from their regular failure to assign sem antic contents that
are consistent w ith the ideology of the speaker. The bronze-age re
port 'Thor is hurling heavenly fire and pounding his h am m er' m ay
be a reliable indicator that (i.e ., have the calibrational content that)
there is a sudden large-scale flux of electrons w ith accom panying
atm ospheric shock w ave. But this latter description w ould radically
m isrepresent the m eaning of the b ronze-age native's rep ort. That
m eaning is fixed by the assem bled beliefs about Thor, fire, and so
forth, em braced by the native. The point here is just the obvious one
that w e can system atically misconceive w hat w e p e r c e iv e , a fact that
causal accou n ts of m eaning are constitutionally inclined to suppress.
A further reason that causal accounts of m eaning are suspect is
that, on such accounts, sem antic con tent becom es com putationally
im potent. O ne's com putational econom y has access to the form al and
structural features of one's observation judgm ents or rep o rts, but it
has no access to their distal causal an tecedents. The com putational or
inferential consequences of an observation judgm ent, therefore, will
be indepen dent of its "c o n te n t," if sem antic con ten t is assigned accord
ing to distal causal antecedents.
A final problem w ith causal accounts is the am biguity of their
assignm ents of m eaning. A n u tterance typically has a unique m ean
ing, at least given context, but an utterance is alw ays the last m em ber
of a long sequence of internal and external causal an teced en ts. W hich
of the m an y causal anteced en ts that m ake up the chain supplies the
"re a l" m eaning of the utterance, and on w hat principle is the selec
tion m ade?
Part of the initial appeal of causal accounts of m eaning derives from
the fact that, w hen w e translate or assign content to the observation
reports of an other speaker, w e typically exp ect his observation re
p orts, as translated by us, to be roughly reliable indicators of the
w orld around him as perceived by us. Thus the im pulse to assign
con ten t on a causal/calibrational basis. But while quite reasonable in
p edestrian circum stances, this translational policy ignores the possi

bility that the speaker has a system atically different, and p erhaps sys
tem atically m istaken, conception of observational reality. Translating
his u tteran ces according to their external causes as reckoned by us will
thus am ou nt to a system atic m isrepresentation of the alien's m ean
ing, a "W h ig tran slation ," as it w ere. W orse yet, this policy ignores or
precludes the possibility that w e too m ight be system atically m istaken
in our observational conceptions. It is therefore am using that a causal
ap p roach to m eaning should be advanced in criticism of the claim that
observation is plastic and theory laden, for the causal theory w as in
itially plausible only to the degree that it ignored the theoretical com
m itm ents implicit in our observation jud gm en ts, and suppressed the
possibility of system atic falsehood in our observational conceptions.
Such factors m u st inevitably return to h au n t causal accounts. A nd
they do. Judgm ental error em erges as the single greatest problem for
causal accounts of m eaning, even in their loci classici: D retske 1981;
F o d o r1 9 8 8 a .
This concludes m y discussion of causal approach es to m eaning. To
the extent, often very limited, that our term s do find real referents or
extensions in the w orld, it is because of the global structure and the
global virtues of the fram ew ork that em beds those term s. Crudely,
there m u st exist an assignm ent of individuals to our singular term s,
and an assignm ent of extensions to our general term s, such that a
significant proportion of the beliefs or assum ptions em bedding those
term s are true on that assignm ent. W hat cou n ts as "sign ificant" will
here go u nexplored, but pragm atic considerations surely dom inate.
This view is consistent, w e should note, w ith a fierce skepticism
about the ultim ate integrity of the notion of reference. N othing
g uarantees that a system of beliefs will alw ays, or ever, determ ine a
unique set of relations betw een its term s and elem ents in the w orld
(see Pu tn am 1975, 1981). If genuine reference m ust be unique, then
very likely it is chronically un d erdeterm ined by the structural p rop
erties of sets of beliefs. Som e will see in this a m otive for trying to add
in a causal com p onen t to the theory of m eaning in order to take up
the slack left by unaided netw orks of belief. For exam ple, the 1985
P u tn am succum bed to this im pulse; the 1981 Putn am did n ot. But I
suggest that the real lesson of the Twin E arth cases, and of the later
argum ents inspired by the Skolem -Low enheim theorem , is quite
different and rather m ore sobering. The real lesson is that the folk-
sem antical notion of "referen ce" is w ithout any real integrity. Refer
ence is uniquely fixed neither by netw orks of belief, nor by causal
relations, nor by anything else, because there is no single and uniform
relation that connects each descriptive term to the w orld in anything

like the fashion that com m on sense supposes.
These last rem arks touch on them es that cannot be p u rsu ed here
(see ch ap ter 13 and Stich 1989). H appily, reference is not at the center
of the debate over theory-ladenness. It is m ean ing-as-u nd erstan d in g
that occupies that position. M y aim in this section has b een to reestab
lish the stron g presum ption that any observation term , to the extent
that it is m eaningful at all, m u st be em bedded in a netw ork of corrigi
ble assum ptions.
2.3 Sensational plasticity versus conceptual plasticity

O ne possible w ay to defend Fodor w ould be to concede the theory-
d ep en den t character of our observational concepts and jud gm en ts,
and try to insist on no m ore than the theory-in d epen d en t ch aracter of
our sen sations. Fod or him self seem s to be sketching a position of this
sort late in his p aper w hen he urges the rigidity of "th e look of
th in gs" versu s the penetrability and plasticity of "h o w things are
judged to b e " (p. 40).
But this defense will not take us any distance at all. For one thing, if
all Fod or w ishes to insist on is uniform ity in the ch aracter of our
sensations through changes in our doxastic com m itm en ts, then his
argu m en t is largely an ignoratio. It fails to address the m ajor epistem o
logical tradition at issue, w hose central them e has alw ays been the
theory-laden ch aracter, not of our sensations, but of our observation
al con cep ts and observational judgm ents.
A n d there is a very good reason for the centrality of th at them e.
Thinkers in the tradition at issue (Popper, Feyerabend, H an son , etc.)
have been primarily concerned with the refutation or corroboration of
theories. But sensations them selves neither confirm nor refute any
theory. Sensations belong to the w ron g logical sp ace: it is only an
observation ju d g m en t, or belief, or report that can be logically consistent
or inconsisten t w ith any theory (Popper 1959). Thus the chronic co n
cern, th rou g h o u t the positivist and postpositivist periods, with the
possibility of a theory-neutral observation vocabulary. W h eth er sen sa
tions them selves m ight be infected or modified by theory w as rarely,
if ever, an issue.
M y o w n 1979 position, to cite one target of F od or's, sim ply assum es
the generally constant character of our sen sory resp o n ses to the
en vironm en t. The plasticity that excited m e there w as confined to
the conceptual fram ew orks w ithin w hich w e m ake our judgm ental
resp o n ses to the passing contents of our sensory m anifold. A ccord
ingly, if rigidity in the character of our sensations is all Fod or is
con cern ed to defend, then I do not u n derstand his objection to,

and dism issal of (pp. 2 8 -2 9 ), the alternative perceptual possibilities
sketched in m y 1979 (p. 30). For that sketch m akes no assum ptions
about the plasticity of our sensations. It is conceptual plasticity that is
there at issue.
To be sure, sensational plasticity w ould constitute an additional
argu m en t for the plasticity of perception. A t least one au th or has
cautiously advan ced a claim of this kind (Kuhn 1962, pp. 1 2 0 -1 2 1 ).
A n d I, for an other, am now willing to defend it vigorously (recall the
exam ples in figure 12.1). So there is a genuine point to attacking it, as
Fod or does. But it is w rong to rep resent or regard this attack, su ccess
ful or otherw ise, as aim ed at the principal argu m en ts in favor of
theory-lad en ness. Those argu m en ts have typically b een based on
oth er grou n ds entirely: on the plasticity of our conceptual respon ses
to sensory activity.
3 Conclusion
I shall spare the reader a reprise of the various conclusions already

reached. Instead, let m e try to evoke a general picture of the situation.
The central issue of this p ap er is n ot an argu m en t about the obscure
etiology of a certain class of beliefs or the arcane sem antics of a certain
class of term s. The real d isagreem ent is about the fundam ental ch ar
acter of the h u m an epistem ic situation and the long-term possibilities
for the evolution of the h um an spirit.
O ur epistem ic situation I assert, is one in w hich even the hum blest
jud gm en t or assertion is alw ays a speculative leap, not just in its
assertion over its denial, b ut also in the background conceptual
fram ew ork in w hich that judgm en t is con stitu ted, in preference to the
infinity of oth er conceptual fram ew orks that one m ight have used
instead. In the case of perceptual jud gm en ts, w hat the senses do is
cause the p erceiver to activate som e specific representation from the
an teced en t system of possible rep resentation s— that is, from the con
ceptual fram ew ork— that has been b rough t to the perceptual situa
tion by the perceiver. A perceptual ju d gm en t, therefore, can be no
better, though it can be w orse, than the broad system of rep resen ta
tion in w hich it is constituted.
This m ean s that perceptual jud gm en ts are evaluable at tw o distin
guishable levels. The first con cern s the propriety of the jud gm en t as
evaluated by the local standards of the fram ew ork that em beds it
(W as the observer in a position to m ake it? W as the observation
carelessly m ad e? Is it inconsistent w ith inform ation already in han d ?).
These corresp on d to w hat C arnap (1956) has called "in ternal ques
tio n s." The second level concerns the adequacy of the em bedding
fram ew ork overall, as a system adequate to rep resent the range of
nom ologically possible configurations that the objective w orld m ight
assum e. These correspond roughly to w hat C arnap has called "e x te r
nal qu estion s" (Do the categories of m y fram ew ork capture the
objective divisions in reality? Do the basic generalizations of m y
fram ew ork exp ress genuine laws of nature?).
External questions are rath er daunting. Local standards of evalua
tion are b oth inapplicable and question b egging, and global standards
are vagu e and elusive. G ood positivist that he w as, C arnap reacted by
denying that such questions are factual in ch aracter, claim ing that the
decision to u se a given conceptual fram ew ork is ultim ately just a p rac
tical question, to be decided on pragm atic grounds. But C arn ap w as
m istaken in seeing a fundam ental difference betw een the tw o kinds
of questions, and b etw een tw o kinds of grou n d s for accep tan ce. So-
called "extern al qu estions" are just large-scale theoretical questions,
to be decided on empirical and system atic grou nd s like anything else.
A nd so-called "p rag m atic" considerations attend epistem ic decisions
at every level of inquiry, even the m ost hum ble and m u n d an e.
In fact, external questions are confronted by hum ans and dealt with
on a daily basis, by scientists inventing and evaluating new fram e
w orks for understan din g this or that dom ain, and also by infants
and children, w ho m ust evolve a conceptual fram ew ork adequate
to cond u ct a life in the Lebensw elt of con cu rren t h um an society. This
m ean s that our conceptual fram ew orks can and regularly do und er
go change, both within the lifetime of an individual and in society
as a w hole, over historical periods. To use a H egelian figure, the
jou rn ey of the h um an spirit is essentially the story of our evolving
conception of the w orld, and of our ow n place w ithin it. O ur eyes are
little different from a baboon's or a ch im p an zee's, but our p erceptual
know ledge is profoundly superior to theirs. O ur m otor system s are
little different from those of any other prim ate, but our practical cap a
bilities and intentional actions encom pass universes quite closed to
them . The m ain difference lies in the dram atically superior con cep
tual fram ew orks w e have evolved epigenetically, and not w ithout
m isad ventu re, over the course of the last 500,00 0 years.
If w e h ave com e this far, m u st the journ ey end h ere? M anifestly
not. The long aw akening is potentially endless. The h u m an spirit will
continue its breathtaking adventure of self-reconstruction, and its
percep tu al and m otor capacities will continue to develop as an inte
gral part of its self-reconstruction. But only if w e try hard to see new
opportunities, and only if w e w ork hard at leaving old fram ew orks
behind.
Chapter 13
Conceptual Progress and Word-World Relations:
In Search of the Essence of Natural Kinds
The problem of natural kinds form s a busy crossroad s w here a n u m

ber of larger problem s m eet: the problem of universals, the problem
of induction and projectibility, the problem of natural laws and de re
m odalities, the problem of m eaning and reference, the problem of
intertheoretic reduction, the question of the aim of science, and the
problem of scientific realism in general. N or do these exh au st the list.
N ot surprisingly then, different w riters confront a different “ problem
of natural k in d s," depending on w hich background issue is for them
the principal issue at stake. The issues of essentialism , m eaning, and
reference, for exam ple, have tended to dom inate recent discussions
of natural kinds (Kripke 1972; Putnam 1975, 1981; M ellor 1977;
C hurchland 1979; Shapere 1982). But evidently these issues are only
p art of the puzzle.
The p resen t paper outlines in fairly broad strokes a view of natural
kinds, an d of our sem antic access to them , that contrasts fairly sh arp
ly w ith the recently popular view associated w ith P u tn am and
Kripke, w herein the term s of natural language can be and generally
are firmly attached to real, natural classes of objects, even w here w e
m ight hold system atically false beliefs about those objects. This view ,
it seem s to m e, has had a largely retrograd e effect on discussions
concerning the relation(s) betw een language and the w orld, and there
is need to reconceive the m atter from a broader and m ore naturalistic
perspective.
I propose to be skeptical here about tw o things in particular. First, I
shall criticize the idea that there is a theory-neutral or intension-
in d ep en d en t relation that connects w ords to unique natural sections
of the w orld. A nd second, I shall argue that on the best available
accou n t of natural kinds, m ost of the kinds w e regard as natural kinds
are not natural kinds at all, but m erely "p ractical" kinds. The set of
This paper first appeared in the Canadian Journal o f Philosophy 15, no. 1 (March
1985).
genuinely natural kinds contains at m ost a tiny elite of v ery basic

physical properties, and m ay indeed be em p ty entirely. The positive
u p sh ot of the discussion is that w e need to reappraise our conception
of the h um an m edium of cognitive rep resentation and its relation to
the w orld, and w e need to reconsider our conception of the m ost
basic aims of rational cognition. I shall try to address these issues as
w e proceed.
1 N atu ral Kinds and Scientific P rogress: The P utnam -K ripke View
W h at w ants rejecting in the Putnam -K ripke story is not just the speci
fic m odels of how the extensions of natural-kind term s are secured
(by a dem onstrative ostension or dubbing cerem on y, plus som e re
cursive device), but also the background vision of the aims of science,
and of the nature of scientific progress, in w hich that claim is m ad e.
Their choice of exam ples is significant: gold, w ater, and heat dom i
nate the discussion. These are all kinds that are well entrench ed in
the com m on vocabulary, and all have enjoyed a relatively sm ooth
reduction to som e category provided by m icrotheory. W ater is H 20
m olecules, gold is A u atom s, and heat is m olecular kinetic energy.
The picture w e get, if w e generalize innocently from these exam
ples, is one of a stable anteced en t taxon om y of reality, a fram ew ork of
co m m on sen se term s each in firm referential contact with an objective
ly real class, w here the advance of science provides us w ith nothing
m ore than a new and m ore penetrating accou n t of w hat unites that
already palpated class. The aim of science em erges as the explication of
natural classes already secured, rath er than as the discovery of novel
classes that finally do display natural integrity, classes that m ay well
fail to correspond , even roughly, to the specious categories w e
antecedently em braced.
This conservative picture tends to suppress the im portant fact that
the an teced en t taxonom y provided by com m on sense is as richly
theoretical, conjectural, and provisional as are the m ore obviously
"th eo retical" taxonom ies that m ay, or m ay not, reduce it. A nd it tends
to suppress the fact that the "m an ifest" taxon om y of com m on sense
has neither a better, nor a firm er, nor even a different handle on the
hidden natural kinds than do the novel theoretical taxonom ies that
aspire to replace it.
Since all know ledge is conjectural and theoretical, any fram ew ork
of com m onsense term s at any stage of our history will have the sam e
tenuous sort of claim to have as extensions the real natural classes.
That claim will have the sam e sort of grounding at any stage, nam ely,
the explanatory and theoretical success of the current taxonom ic
In Search of the Essence of Natural Kinds 283
fram ew ork. A n d each stage m ust anticipate being su p ersed ed , soon

er or later, by a quite different taxonom y of natural kinds, w ith great
er explan atory and predictive su ccess. This m eans that a central aim
of science is to find out w hich of the infinite range of possible natural-
kind taxon om ies is truly the right taxonom y. To m ake the "exp lica
tion " of current categories a fundam ental aim of science is to claim for
those categories an integrity that they m ay not possess.
The fluidity and fragility of com m onsen se taxon om y can be seen
even in the relatively stable exam ples at issue. The "g o ld " of the late
m edieval period included rather m ore than w e co u n ten an ce as gold
today, since "g o ld " w as then conceived prim arily in p h en om en o
logical term s that adm itted sundry alloys and ersatzes of gold into the
class. This w as n ot simply a mistake on the part of our an cestors. As
they con ceived of things, "g o ld " w as som ething that cam e in various
g rades, w hich trailed off sm oothly into the b aser m etals. A n d from the
point of view of alchemical theory, this w as only to be exp ected , since
the hidden principle responsible for the characteristics of high-grade
" g o ld " w as thou gh t to be a spirit that displayed varying d egrees of
m aturity, a spirit that w as w ed in varying degrees of h arm on y w ith
the oth er spirits that ensoul a m etal. C onceived within m edieval com
m o n sen se, the extension of their term w as w ider than ours. C on
ceived w ithin alchem ical theory, it had no extension at all. In neither
case did it have the sam e extension as our term " g o ld ."
The extension of the term "w a te r" has presum ably u n d ergon e a
similar evolution, as v ery primitive peoples cam e to experience phase
transitions, cam e to see the p overty in a notion of " w a te r" that holds
w etn ess to be an essential feature, and cam e to a broader notion of
w ater that included ice and steam as variant form s of the stuff. M ore
recently, the extension of this term has expanded again, to include
m ore than just H 20 , since w e have learned that deuterium will form a
com p oun d , heavy w ater, that is distinct from , but strikingly similar
to, the m ore com m on com pound form ed from hydrogen.
The case of com m on sen se heat and tem p eratu re provides us w ith a
m ore dram atic exam ple. A s I have argued elsew here (1979, pp. 23-24)
the co m m onsen se notions of hot, w arm , and cold are em pirically in
coh eren t, in that they attem pt to im pose a one-dim ensional con
tinuum of properties w here nature supplies three distinct and diver
gent continua— degree of heat energy, am ount of heat en ergy, and rate
o f flo w of heat en ergy— none of w hich corresp on d s adequately to the
com m on sen se conception. O ur com m onsense term s here are n ot just
different in extension from the therm odynam ic term s th at displace
them ; they are entirely em pty of extension, despite their usefulness
in our quotidian affairs, since nothing in nature answ ers to the col
lected law s of "com m on sen se th erm od y n am ics."
I am keenly aw are that this claim is initially im plausible, but I stand
by it. Rather than enter again upon the intricacies of its defense,
how ever, let m e here address w h at makes it an upsetting claim. It
u p sets because it holds up the p rosp ect that entire dom ains of our
co m m on sen se observational vocabulary m ight be utterly w ithout
reference or extension. It holds up the sp ecter of large chunks of
language, even an entire language, unconnected to the w orld by w ay of
the reference of its singular term s and the extension of its general
term s. This is not only shocking to contem plate; it leaves it a m ystery
how any language could function in such a disconnected predicam ent.
Shocking or no, let us contem plate it. (W e will address the m ystery
of how a language can function in that predicam ent in a m o m en t.)
W ithout im pugning anything that currently passes for com m on
sense, w e can easily appreciate the possibility of such m assive ref
erential disconnection if w e imagine a radically false scientific theory
taught and used as an observation fram ew ork by ordinary people. To
take an easy exam ple, suppose w e w ere to use the fram ew ork of calor
ic flu id to conceptualize and talk about therm al ph en om en a. (This
possibility is explored at length in m y 1979, p p . 1 6 -2 5 .) M ore specifi
cally, suppose that w e w ere trained to exp ress our spontaneous
observation judgm en ts and rep orts in the vocabulary of caloric theory,
e .g ., "T h is porridge has a high-caloric fluid pressu re, M o m m y ."
Since you are antecedently convinced that caloric fluid does not exist
and that the vocabulary of caloric theory is em pty of reference and
extension, you will agree that such a linguistic regim e w ould m eet the
conditions at issue: a vocabulary functioning in an "ob servation al"
m od e, but w holly w ithout reference or extension.
There is no question that the vocabulary of caloric theory could
function sm oothly in the role described. (In fact, it w ould function
even m ore sm oothly than our cu rrent co m m on sen se conceptions.)
A nd the reason it could do so is that, while it w ould be disconnected
from the w orld referentially, it w ould rem ain richly connected to the
w orld causally, by w ay of our conceptual and linguistic respon ses to
sensory input, and by w ay of our m otor respon ses to our deliber-
ational output. The prim ary connection betw een any language and
the w orld is thus a causal one, and successful functioning requires
only that the language constitute a useful subsystem of our sen sor
im otor p ath w ays. If a language also enjoys referential connections
w ith the w orld, that is a fortunate and a secondary achievem ent, one
that relatively few linguistic fram ew orks can boast. M assive referen
tial disconnection, therefore, is a real possibility, even in a sm oothly
functional vocabulary, and this rem ains true w hatever w e decide

about the specific case of com m onsense tem perature.
So it is n ot unthinkable that our current com m onsense conceptions
of hot and cold are em pty of extension. H ow ever, let us suppose that
m y earlier judgm en t against our com m onsense therm al con cep ts is
too h arsh ; let us suppose that som e principle of charity bids us count
therm od ynam ic tem p eratu re (degree of h eat energy) as the real ref
eren t of our earlier u sage of 'hot' and 'co ld '. The irony is that therm o
dyn am ic tem p eratu re has turned out to enjoy no uniform essence in
any case. Tem perature is m ean m olecular KE, in a gas. But in a clas
sical solid, tem p eratu re is m ean m axim um m olecular KE. In a plasm a
it is a com plex mix of differently em bodied energies (ions, electrons,
p h oton s), depending on just how high the tem perature is. A nd in a
v acu um it is a specific w avelength distribution am ong the electro
m agn etic w aves coursing through that vacu u m . (These different in
stances of "tem p e ra tu re " form a coherent family b ecause diverse
instantiations can nevertheless be in m utual energetic equilibrium.)
A familiar property has fragm ented again, and discovers new and
confusing instances. Even a vacuum can have a tem perature! O ur
taxonom ies form , dissolve, and reform , even as w e w atch.
Reflecting on conceptual changes of the m agnitude w e have been
discussing m ay produce the occasional spell of vertigo, b ut the p re
ceding exam ples illustrate that pow erful conceptions can function
and flourish despite m assive referential disconnection, that com p et
ing theories need n ot be talking about "th e sam e th in g s," and that
rational choices betw een theories can be m ade, and rational progress
can be achieved, even though there is no sam eness or continuity
of reference, either synchronically across com peting theories or dia-
chronically across successive theories.
The com peting idea that our com m on sen se term s do enjoy stable
and objective natural classes as their extensions, by w ay of a Putnam -
ean indexical/recursive form ula, despite confusion or system atic
falsehood in our beliefs about them , is perhap s an appealing idea,
since it seem s to m ake reference to real classes an easy m atter in a
difficult w orld. But it m akes it far too easy for it to bear m u ch scrutiny.
To see this, consider a seven teen th -cen tu ry incarnation of Putnam ,
intoning the following form ula over a small fire:
"P h log isto n is w hatever bears the 'sam e-spirit' relation to the
phlogiston right here" (speaker points to the vaporous outp ou r
ings above the fire).
W h eth er this secures anything at all in the w ay of a natural extension
will be a function of the theoretical integrity of the notions of "p h lo
gisto n " and "is the sam e spirit a s ." A s it h ap p ens, neither notion has
any genuine integrity, and so nothing at all is secured. The sam e is
true of oth er exam ples:
"C aloric fluid is w hatever bears the 'sam e substance' relation to

the caloric I feel right here" (speaker holds up a w arm coffee cup).
"H eav en ly crystal is w hatever b ears the 'sam e-su b stan ce' rela
tion to the crystal up there" (speaker points to the crystal sphere
that divides the superlunary from the sublunary realm s).
"P a rty drink is w hatever bears the 'sam e-liquid' relation to the
party drink in m y h an d " (speaker holds up a glass of g ran dm a's
random ly concocted pink p arty punch).
P u tn am 's form ula, evidently, provides no g uarantee that its intona

tion secures a natural class for an extension, nor that it secures
anything at all for an extension. N or should this be surprising. As
W ittgenstein taught us, conceptually unaided pointing singles out
nothing. But once w e include a specific conceptual surround, the
secured extension is partly a function of our intensions, our general
assum p tions, our theories. A similar point holds for the recursive
clause of P u tn am 's form ula. Expressions such as 'sam e substance'
and 'sam e spirit' are obviously laden w ith th eory, and the sam eness
relation there invoked will vary from one period of scientific history
to an other. P u tn am 's form ula, therefore, does not constitute a non-
intensional or theory-neutral m eans of securing a natural class as
an extension for a given term , nor need it su cceed in securing any
natural class at all.
W e can and m ust agree that Putnam sen ten ces occasionally play
som e role in delineating the extension of som e term s. The sentences
that collectively m ake up any term 's intension are usually general
sen ten ces, but singular sentences occasionally acquire a com parable
im p ortan ce (see C hurchland 1979, pp. 7 2 -7 3 ). For exam ple, the
sen ten ce, 'The stuff in our lakes, rivers, and oceans is w ater' is a
sem antically im p ortant sentence. That sentence thus plays som e role
in determ ining the extension of our term 'w a te r'. It provides a (cor
rigible) sufficient condition for being w ater. But the sem antic role of
such singular indexical sentences is in no w ay p riv ileg ed : though they
m ay play a role in fixing the extension of a given term , they are just
further elem ents in the intension of that term . (Indeed, the exam ple
just cited has lost som e of its sem antic im portance as a result of the
expansion of chem ical theory during the last three cen tu ries.)
W h eth er a term has a nonem pty extension rem ains conditional on
w hether the collected assum ptions that constitute its intension are, at
least for the m ost part, true. A n d exactly w hich extension gets secured
is a function of just w hat those collected assum ptions a re .
Since a term 's intension (the set of sem antically im p ortan t sen
tences in w hich that term figures) m ay occasionally include indexical
or dem onstrative sentences, it m ust be conceded that exactly w hat
extension gets secured will be determ ined in part by w h ere, and in
w hose brain, that intension is asked to function as p art of an ongoing
cognitive econ om y. Extension is determ ined by intension plus con
text. If this is a concession to P utnam , then I think w e m u st m ake it.
H ow ever, none of this implies that our natural-kind term s enjoy a
firm and stable referential connection to a natural class of objects in
d ep endently of the intensions that govern the use of those term s.
N one of this guarantees that our term s have a n on em p ty extension.
N one of this implies that the extensions of our term s do n ot differ as
they ap p ear in com peting theories. A nd none of this implies that the
extensions of our term s do n ot change as w e change w hich theories
w e em brace.
A ccordingly, w e m u st learn to live, after all, w ith a roughly Fregean
conception of the relation betw een intensions and extensions, even in
the case of natural kinds. There is no royal road to securing natural
classes as extensions for our term s. O ur science m ust achieve refer
ence the old-fashioned w ay: it m ust earn it, by slow plodding
throu gh a variety of very different conceptions of the w orld, in hopes
of fin d in g a theory that successfully carves nature at h er system atic
joints, a theory w hich displays that success in its unexcelled explana
tory and predictive pow ers. O ur referential access to natural kinds is
by w ay of the corrigible system s of sentences w e accept, and of w hat
an sw ers to them in the w orld. This does m ake our access to natural
kinds fluid, uncertain, and problem atic. But that is as it should be.
The extensions of our term s are stably fixed neither by analytic truths,
as in the o rth od ox em piricist tradition, nor by indexical/recursive
pointings, as in the Putnam -Kripke alternative. They are n ot stably
fixed by anything, since they are not stably fixed at all.
This concludes m y discussion of our access to natural kinds. Let m e
now ad d ress the question of the essence of natural kinds. W h at are
natural kinds? W hat distinguishes them from less substantial kinds of
kinds?
2 N atu ral K inds as Law -B ound K inds: Som e Virtues, Consequences, and
D ifficulties
C om m onsen se and P u tn am 's and Kripke's discussion exam ples

su ggest a conception of natural kinds as any kind w ith a hidden
m icroessence that unites all of its instances. But that conception is

parochial and ill considered, because it invites one to think of natural
kinds as successfully revealed only or prim arily in sm ooth interthe-
oretic reductions, in w hich com m on sen se or an tecedent categories
survive by reason of finding close analogues in the m icro realm .
A better view conceives natural kinds as just kinds com prehended
by genuine laws of nature. That is, the natural kinds are the law -bound
kinds. This view sees natural kinds as revealed prim arily w hen a new
taxon om y displaces an older schem e entirely. (Such displacem ent can
occasionally include a reduction of the old by the n ew , but it need
n ot.) The identification of natural kinds rem ains a basic aim of sci
ence, on this view , but that is because identifying the laws of nature
is a basic aim of science, and because the taxon om y of kinds and
their em bedding law s m ust em erge from inquiry together.
This conception of natural kinds is appealing in its simplicity, but it
does p resen t serious puzzles of its ow n. M ost of these concern the
notion of a natural law and the m atter of their hierarchical relations. It
is all very well to say that a natural law is a spatiotem porally u n re
stricted, counterfactual-supporting, explanation-sustaining universal
conditional. But w h en it com es to cases, confusion gathers swiftly.
W e can all agree that the sentence
(1) A ny m ass free of net forces m oves, if at all, in a straight line

w ith con stant velocity
is a law of n ature, and hence that m ass, fo rce, and velocity are natural
kinds or properties. But now consider the sentence
(2) A ny chair free of net forces m oves, if at all, in a straight line

w ith constant velocity.
Is this a law of nature? Is chair thus a natural kind? If it is, then a lot of
things are going to count as laws, and a lot of kinds are going to be
natural: toothbrushes, spaceships, toasters, etc. Surely chair is not a
natural kind. But how do w e deny natural-law status to (2)? Sentence
(2) follows from (1) and
(3) A n y chair is a m ass used for sitting on.
W e can all agree that (3), though true, is n ot a natural law, and
p erh ap s here lies the key to denying natural-law status to (2).
Suppose w e identify som e natural laws as basic law s, on grounds of
their relative inexplicability, and then stipulate that nothing else
cou nts as a genuine natural law unless it is a purely logical con se
quence of some one or more-basic laws. This will rule out (2) as a natural
law, since it is neither basic, nor a consequence of purely natural laws
alone: the nonnatural "law ” (3) is also required for its deduction. This
solution has independent plausibility. To explain a specific chair's
rectilinear m otion through space, one does not point to the fact that it
is a chair, and then cite (2). R ather, one points out th at it has mass,
and then one appeals to (1). It seem s to be (1) that carries the explana
tory clout, n ot (2).
This accom m odation w ith the problem has considerable appeal,
but if w e buy it, w e m u st be prepared to live w ith its consequences.
Specifically, natural kinds turn out to constitute a very small and
exclusive com pany, m uch smaller than one m ight have thought
initially. Let m e ap proach this claim by w ay of som e exam ples.
In learning elem entary m echanics, one learns to deduce lawlike ex
pressions for the rotational inertia of various kinds of bodies. Starting
from the relatively basic law that the rotational inertia / of a point
particle of m ass m at a distance r from the axis of rotation is m r 2, one
can d ed u ce general expressions for the rotational inertia of bodies
w ith a m ore com plex configuration. If M is the m ass of the body, and
R and L the relevant radius and length, w e have, variously,
F or any hoop, I = M R 2
F o r any solid cylinder, I = M R 2/2
F o r any thin rod, I = M L 2/ 12
For any solid sphere, I = 2M R 2/5
For any dumbbell, I = 2 M R 2/5 + M L 2!A
F or any hollow spherical shell w ith a thin rod stuck through it
tran sversely, I = 2 M R 2/3 + ML2/12
and so forth (see figure 13.1).

Evidently, there are an indefinite variety of shap es, and for any
specified shape a putative law can be dedu ced , using only the
"d efin ition " of the relevant shape or configuration of point m asses,
the basic inertial law cited, and the sum m ative/integrative laws of
m athem atics. If the universally quantified sentences listed above are
genuine natural law s, then hoop, dum bbell, hollow spherical shell with
thin rod stuck through it transversely, and an unlim ited num ber of other
inelegant kinds are all going to count as natural kinds. Fortunately,
these generalizations do not follow from basic laws alone: they all
require a definition or stipulation of the form
(4) A spherical shell is a set of point m asses uniform ly dis

tributed an equal distance R from a center point,
(5) A dumbbell is a set of point m asses co n cen trated in two
solid spheres of radii R a fixed distance L apart,
Figure 13.1
Rotating bodies and axes of rotation
etc. A nd it is plausible to insist that no such stipulation constitutes a

natural law. A ccording to the view being defended, then, the above
generalizations are n ot genuine natural law s, and so the kinds they
com p reh en d need not be natural kinds.
W ell and good. Even so, generalizations such as those listed, and
the kinds they com preh en d, are not to be sniffed at. They constitute
the bulk of our general and practical know ledge, and they can be
extrem ely useful. L et us therefore call such universal statem en ts, as
follow from the conjunction of a genuine law w ith som e stipulative
definitions, a practical law. A nd let us call the kinds that figure in them
practical kinds. These will be kinds constituted or recognized by us
because their instances are very com m on in the environm ent, or be
cause they have som e practical relevance for u s. In the sam e spirit, w e
m ight even speak of practical explanations as those w hich are sustained
by practical laws and w hich concern the behavior of practical kinds
of objects. These will not be genuine explanations, but, w here they
apply, a genuine explanation will be hidden in the w ings.
So far our intuitions have not been unduly strained. But they will
be seriously strained w hen it is appreciated that m ost of w hat are
traditionally counted as natural kinds are m erely practical kinds. Ti
gers, elm s, apples, roses, p erhaps even w ater and gold— all turn out
not to be natural kinds, as characterized above. The reason is that all
of them turn out to be of a piece w ith h oop s, rod s, sp heres, and
dum bbells. All of them are m ore or less arbitrary configurations of
г 1
sm aller or m ore basic bits. N one of the generalizations in w hich they

figure is a genuine law of nature. N one of them , fairly clearly, is a
basic law . M oreover, none of them , I em phasize, can even be deduced
from basic law s save in conjunction w ith a nonlawlike stipulation sp e
cifying h ow instances of that kind (tigers, elm s, etc.) are constituted
from m ore basic elem ents of reality.
Take the natural kind tiger, for instance. A nd let us sup p ose that
the following is as close as w e can get to a genuine natural law about
tigers.
(6) A n y live, adult tiger burns up at least 1000 Kcal p er day.
This "p ractical" law is n ot beyond explanation. From a specification

of the general physical and biochem ical constitution of a tiger, one
can d ed u ce from basic therm odynam ical laws that a system of that
constitution will use en ergy at such a rate. But the specification of the
general physical and biochem ical constitution of a tiger is on all fours
with the specification of the distribution of point m asses in a spherical
shell w ith a thin rod stuck through it transversely. N either specifica
tion is a natural law; both specify only one configuration out of an
infinite num ber of equally possible configurations of m ore basic ele
m ents, each of w hich configurations w ould have its o w n peculiar
physical properties, equally a consequence of basic law as applied to
that arbitrary case.
O ne d oes not naturally think of biological species as being as arbit
rary as I am here insisting, since our w orld p resents us w ith only a
fixed subset of the infinite num ber of possible species. But that subset
is an acciden t of evolutionary history. W e could have had a com plete
ly different set of species, any one of countless other sets, just as
w e can have countless bodies differently configured so as to have
different rotational inertias.
If the kinds tiger, elm , and apple turn out to be m erely practical
kinds, so also, it seem s, do w ater and even gold. A n instance of either
substance is only one am ong endlessly m any possible configurations
of m ore basic elem ents, and its physical properties are determ ined by
the law s governing those m ore basic elem ents. From a specification
such as
(7) A n y H 20 m olecule consists of tw o h yd rogen atom s valence

bonded to one oxygen atom ,
in conjunction with electrodynam ical law s, one can p erh ap s deduce,
(8) A n y volum e of H 20 m olecules is tran sp aren t to light.
But (7) seem s to be of a piece w ith (4) and (5), and thus (8) em erges as
a m erely practical law about a m erely practical kind: H 20 m olecule. The

sam e sorts of considerations, repeated at the subatom ic level, reveal
Au atom as a m erely practical kind also. A nd so for all chem ical com
p oun d s and all of the chem ical elem ents.
In sum , the only genuine natural kinds ap p ear to be those com p re
h en d ed by absolutely the m ost basic law s of our science. On the view
here outlined, m ass, length, duration, charge, color, energy, m om entum ,
and so forth all turn up safely as natural kinds or properties. But
precious little else does. This austere result m ight m ove us to try to
liberalize the conditions on natural kinds, b ut any attem pt to do so
threaten s to let back in such kinds as hoop, chair, dum bbell, and sphere
w ith a stick through it.
A briefly plausible suggestion here is im portantly m istaken. W hat
discrim inates the chem ical and elem ental kinds from these arbitrary
inertial configurations, w e m ight suppose, is that under natural co n
ditions the subatom ic particles spontaneously form them selves into
the elem ents, and under natural conditions those elem ents sp on
taneously form them selves into the chem ical com p oun d s. They can
thu s cou n t as natural kinds after all.
D espite an initial appeal, this does not draw the desired distinction.
W h eth er and w hich elem ents get form ed is highly sensitive to the
details of the environm ent: tem p eratu re, p ressu re, relative abun
dances of different particles, and so forth. There is nothing inevitable
about the kinds that dom inate our environm ent. W ere the w hole uni
verse un d er the sam e gravitational squeeze that grips the m atter of a
neu tron star, none of the familiar chem ical elem ents w ould exist. O ur
w orld w ould be a sym phony of purely nuclear chem istry rath er than
of electron chem istry. Equally im portant, u n d er appropriate environ
m ental circum stances, m atter w ill form itself spontaneously into
hoop s (Saturn's rings), thin rods (sticks), spheres (planets), or even
dumbbells (close binary stars). In general, if you specify a shape, I can
con coct an environm ent adequate for its sp on tan eou s form ation.
Either all of these are natural kinds then , or none of them are. I tenta
tively conclude that none of them are. Save only the elect few from
the m ost basic physics, all kinds are m erely "p ra ctica l" kinds.
For m y ow n part, I could rest content w ith this view . I m ight even
end m y discussion here. But tw o residual w orries preclude peace of
m ind. The first concerns the basis and the integrity of the distinction
betw een laws and nonlaw s. A nd the second con cern s the existence of
basic or ultim ate natural laws.
The first and lesser w orry stem s from the fact that natural laws
differ tangibly from other true, unrestricted universal conditionals
only in their functional and relational properties: they support
explanations, counterfactuals, and m odal statem en ts; th ey are con

firmed by their instances; they are, or turn out to be, an integrated
p art of a system of other law s. O n these differences w e can agree, but
there seem s to be nothing in the w orld that w e can point to as the
distinguishing feature of lawful regularities.
O ne ap p roach to this problem locates the distinguishing feature of
laws in their relations to things other than the real w orld — in their
relations to each m em ber of the set of nom ically possible w orlds, for
exam ple. Law s are distinguished by being true in all of th em , while
accidental universals are true only in som e.
If this ap proach is to avoid putting the cart before the h orse, then
w e m u st be realists w ith respect to the "s e t of nom ically possible
w o rld s." This appeals to m e hardly at all. Solving the problem of the
sem antics of lawlike generalizations seem s insufficient m otivation for
expanding our ontology by a factor of at least aleph-1 additional
w orlds. Better to take the view that this w orld, the real o n e, itself
contains objective necessities and possibilities, w hich are reflected in
the cou rse of actual events, but are logically und erd eterm in ed by
those events. This need n ot be a hopelessly m ysterious view . It m ay
be that som e objective features of the real w orld find expression only
in certain global features of our system of beliefs, in the coagulation of
laws into coheren t system s of laws. But I will not insist on such an
account. Indeed, I rem ain deeply puzzled about the distinction b e
tw een nom ic and nonnom ic universals, and about the m odalities in
general. N ot puzzled enough to give up the notion of natural kinds,
h ow ever.
The secon d w orry is m ore serious. The account of natural kinds
defended above uses the notion of "b asic" law s of n atu re. It m akes
the tacit assu m ption that there exists som e final, uniquely true theory
w hose law s exp ress the basic regularities in the universe and w hose
predicates denote its m ost basic kinds. It w as n ot su pposed that w e
will ev er possess such a utopian theory; only that our currently best
theories give us our current b est shot at reality's basic law s and basic
kinds. A n d that basic laws and kinds are there to be aim ed at.
This assu m p tion is highly problem atic. I do not know h ow to de
fend it, n or, I suggest, do you. Reflection will reveal its possible falsi
ty, as follows. First, consider the possibility that, for any level of order
discovered in the universe, there alw ays exists a deeper taxon om y of
kinds and a deep er level of order in term s of w hich the lawful order at
the an teced en t level can be explained. It is, as far as I can see, a w hol
ly em pirical question w heth er or not the universe is like this, like an
"exp lan ato ry on ion " w ith an infinite num ber of concentric explana
tory skins. If it is like this, then then are no basic or ultim ate law s to
w hich all successful investigators m ust inevitably be led, and accord

ing to the account of natural kinds defended above, there are no n atu
ral kinds at all. All kinds are m erely practical kinds.
A ccording to m y accou n t then, it is an em pirical question w heth er
there are any natural kinds at all. It depends on w heth er the universe
is or is not an infinitely layered explanatory onion. This too I can live
w ith. But the w orries do n ot end here. Even if the w orld is not an
explanatory onion, there is still no guarantee that there exists a
unique and final theory (that is, a set of sentences) flawlessly ad e
quate to its com plete description. O n the con trary, it m ay be that the
cognitive m edium of h um an natural language suffers certain fun
dam ental structural limitations in its capacity for representing the in
tricacies of the universe. A nd it m ay be that a wholly new m edium of
rep resentation will be required if the h u m an race is ever to m ake
cognitive p rogress p ast a certain level of com paratively paltry
apprehension.
This seem s very likely to m e, since I think it w ould be a m iracle if
evolution had already fitted us out w ith cognitive equipm ent that is
structurally adequate to representing the deep est m ysteries and sub
tlest intricacies of the cosm os. (O n this point I differ substantially
w ith David Lew is, w ho is inclined tow ard the contrary position that,
give or take a few residual w rinkles, the Final Theory will be in our
h an d s w ithin a cen tu ry. N ot only do I not share his optim ism here, I
think that the idiosyncratic linguistic stru ctu res w e call theories, and
even the idiosyncratic neural system s w e call brains, will prove to
have fundam ental shortcom ings qua m edia for m odels of reality.) I
think it is far m ore likely that w e shall have to contrive m ajor physical
ch an ges in, and enhan cem en ts to, our cu rrent cognitive equipm ent, if
w e are to transcend its inherent limitations. This m ight leave us with
a m edium or system of representation in w hich nothing answ ers to the
notion of a universal generalization or the notion of a predicate. W hat
science w ould then be pursuing, I can not say. But I assum e it w ould
no longer be pursuing natural laws and natural kinds. W hich leads
m e to suspect the ultim ate integrity of these notions, right here and
now .
A pair of exam ples m ay help to un d erscore the point just m ad e.
D espite w h at one m ight think, the particular cluster of cognitive ta
lents found in us n eed not characterize any or all intelligent species.
W e tend think of the ability to m anipulate linguistic structures, and
perh ap s m athem atical structures, as being alm ost definitive of high-
grade cognition. But even w ithin our ow n species there are h u m an s,
of otherw ise exem plary cognitive talents, w h o are wholly unable to
en gage in such activities. Victims of global aphasia are utterly unable to
use or to com prehend language, and this is not a m otor or a sensory

deficit, but a specifically cognitive deficit. A n d yet they live am ong us
w ithout m u ch assistance, and display an intelligence that is undim i
nished in m ost oth er respects. Victims of acalculia show a parallel
global deficit w ith regard to the capacity to m anipulate num bers,
even to count in the single-digit range or to m ake the m ost elem entary
num erical judgm ents of less than or greater than. A nd yet such p eo
ple get by w ith surprising and cunning com pensations: one congeni
tal acalculia victim becam e a millionaire in the real estate business,
using an acute eye for prom ising p roperty and a set of m utually iso
lated accou ntan ts to do his figures and banking for him . All of w hich
illustrates that it is possible for a being to be highly intelligent and yet
h ave no capacity at all for m anipulating language or m ath em atics.
This su ggests in turn that there m ay be im portant and fundam ental
cognitive talents for w hich w e are genetically unequipped and of
w hich w e are totally u n aw are, talents w e m u st acquire if w e hope to
un d erstan d m ore deeply the universe around us.
To su m m arize, our access to natural kinds is fluid, conceptually
m ed iated , and uncertain. The familiar m ultitude of putative natural
kinds em braced by com m on sense, and by the m any derivative sci
ences, are at best m erely practical kinds. G enuine natural kinds form
a very sm all, aristocratic elite am ong kinds in general, being found
only in the m ost basic laws of an all-em bracing physics. A n d if there
are no su ch laws, or if the hum an cognitive m edium should turn out
to be a representational cripple, then p erh aps there are n o natural
kinds at all.
Chapter 14
Moral Facts and Moral Knowledge
1 The E pistem ology and O ntology o f M orals
M oral know ledge has long suffered from w hat seem s an unflattering
contrast w ith scientific or other genuinely "factu al" form s of know l
edge. It is not hard to appreciate the appearan ce. O ne has no obvious
sense organ for m oral facts, as one does for so m any of the facts dis
played in the m aterial w orld, and so there is an im m ediate epistem o
logical problem about m oral facts. H ow does one ap p reh en d them ?
C on n ected to this epistem ological problem is an ontological problem .
For em pirical statem ents, one typically finds an objective configura
tion of objects or properties to w hich the statem ent, if tru e, corres
ponds. F or statem ents of m oral truth, such as 'O ne ough t to keep
on e's p rom ises', one seem s not to find com parable objective con
figurations, lying in obliging corresp on dence. A nd even if one rejects
the corresp on d en ce conception of truth implicit in this objection, the
w idely accep ted principle that "o u g h t" cannot be derived from " is "
w ould seem to leave the truth of m oral statem ents grou n d ed in
som ething other than the w ay the m aterial universe h ap p ens to be
configured.
This ap p earan ce has tended to provoke one of tw o possible reac
tions. O n the one h and, w e have a long history of noncognitivist and
oth er deflationary or overtly skeptical accou n ts of m oral "k n ow l
e d g e ." A n d on the other, w e have a long history of inflationary
accou n ts th at try to locate the ground of m oral truth in abstract gen er
al principles that are som ehow certified by reason as opp osed to
em pirical fact. This rough division does not capture every ap p roach :
utilitarianism , perh ap s, is both nondeflationary and nonrationalist.
But the division captures a good deal. M y ow n inclination is to resist
the ap p earan ce that tends to produce these tw o reactions, an d thus to
avoid the m otivation for both pathologies. M oral truths, I shall argue,
are rou gh ly as robust and objective as oth er instances of tru th , but
this objectivity is not secured by their being grounded in pu re reason
or in som e other nonem pirical support. It is secured in som ething
very like the w ay in w hich the objectivity of scientific facts is secured.
W h at m otivates this suggestion is the novel accou n t of know ledge

and conceptual developm ent em erging from neural-netw ork m odels
of cognitive function (see chapters 5 and 9 to 11, especially ch apters 9
and 10). These m odels provide a portrait of know ledge in w hich m ost
of the old epistem ological concepts and con trasts, the ones used to
characterize m oral know ledge in particular as "n o n fa ctu a l," have
sim ply disappeared. The gram m atical and sem antic differences that
app ear to distinguish m oral from factual propositions no longer loom
large, because propositions no longer appear as the prim ary m eans of
know ledge representation in any case. On these neurocom putational
m odels, know ledge acquisition is primarily a p rocess of learning h o w :
how to recognize a w ide variety of com plex situations and h ow to
respond to them appropriately. The quality of on e's know ledge is
m easured not by any uniform corresp on den ce betw een internal sen
tences and external facts, but by the quality of on e's continuing
perform ance. From this perspective, m oral know ledge does not
autom atically suffer by contrast w ith oth er form s of know ledge.
To the con trary, praxis now appears prim ary. It is theoria that stands
in n eed of explanation.
M y aim in this brief chapter is to sketch a view of m oral, social, and
political know ledge that locates it correctly in the epistem ological and
ontological schem e of things. It is an essay in m etaeth ics, but the view
proposed will allow us to m ake som e observations about the ch arac
ter of the principal substantive m oral theories and about the nature of
the conflicts betw een them . Certainly it has consequences concerning
the nature and the possibility of m oral progress. Let m e begin m o d
estly, h ow ever, by sketching a view of the developm ent of m oral
know ledge in children, a view m otivated by the neural-netw ork m o d
els at issue. My aim at this stage is only to su ggest that (a) a child's
acquisition of an elem entary m oral consciousness is not prim arily a
m atter of his internalizing a set of discursive principles, and (b) such
acquisition is a genuine case of learning som ething about the objec
tive w orld. I shall aspire to bolder claims tow ard the end of this essay.
2 M oral Prototypes and M oral D evelopm ent
A s is evident from the w orkings of m ultilayered feed-forw ard n et

w orks, the discrim ination and recognition of certain features of the
perceptual environm ent m ay require the p resen ce of one or m ore
layers of "h id d e n " units, layers that intervene betw een the initial
layer of transducing input units and the outp u t layer w here the d e
sired discrim ination is finally coded. W ithout one or m ore such in
tervening layers, a netw ork's discrim inative capacities will be limited
Moral Facts and Moral Knowledge 299
to only the sim plest of environm ental features (see ch apter 9, section
5). This m ean s that, for m an y properties in the w orld, it is the entire
n etw ork that m anages to m ake the relevant discrim inations: the job is
n ot done by the units of the input layer alone.
Let u s turn now to biological brains. It has becom e evident that
very few , if any, of the properties w e com m only regard as observa
tional are distinguished by the peripheral cells alone. M ost of them
involve the activity of several subsequent layers in the processing
h ierarchy. A nd this is certainly true of relatively subtle properties.
C onsider you r observing that the sky is threatening, that a b an an a is
ideally ripe, that the car's engine is still cold, that M ary is em bar
rassed , that the lamb chops on the grill are ready, that the class is
bored, th at an infant is overtired, and so forth. These are the sorts of
im m ediate and autom atic discrim inations that one learns to m ake,
and on w hich on e's practical life depends. To be sure, th ey are am-
pliative discrim inations relative to the often m eager peripheral sti
m ulation that triggers them , and they are highly corrigible for that
very reason . But they are not the result of applying abstract general
principles, nor the result of draw ing covert discursive inferences, at
least in a w ell-trained individual. They rep resent the norm al and
alm ost instantan eous operation of a m assively parallel netw ork that
has b een trained over time to be sensitive to a specific range of en
vironm ental features.
The discrim ination of social and m oral features is surely an instance
of the sam e process, and it is m ade possible by training of a similar
kind. Children learn to recognize certain prototypical kinds of social
situations, and they learn to produce or avoid the behaviors proto-
typically required or prohibited in each. Young children learn to recog
nize a distribution of scarce resources such as cookies or candies as a
fa ir or u nfair distribution. They learn to voice com plaint in the latter
case, an d to w ithhold com plaint in the form er. They learn to recog
nize that a found object m ay be som eone's property, and that access is
limited as a result. They learn to discrim inate unprovoked cruelty, and
to dem an d or expect punishm ent for the tran sgressor and com fort for
the victim . They learn to recognize a breach o f prom ise, and to how l in
p rotest. They learn to recognize these and a hundred oth er p roto
typical sodal/m oral situations, and the w ays in which the embedding
society generally reacts to those situations and expects them to react.
H ow the learning child subsequently reacts to a novel social situa
tion will be a function of w hich of her m an y prototyp es th at situation
activates, and this will be a m atter of the relative similarity of the new
situation to the various p rototyp es on w hich she w as trained. This
m eans that situations will occasionally be am biguous. O ne and the
sam e situation can activate distinct p rototyp es in distinct observers.

W hat seem s a case of unprovoked cruelty to one child can seem a case
of just retribution to another. M oral argum ent then consists in trying
to red u ce the exaggerated salience of certain features of the situation,
and to enhan ce the salience of certain oth ers, in order to change
w hich p rototyp e gets activated. The stored prototypes them selves
regularly undergo change, of course, as experience brings ever new
exam ples and the child's social/m oral consciousness continues to
develop.
W h at the child is learning in this process is the structure o f social
space and how best to navigate one's w ay through it. W h at the child is
learning is practical w isdom : the wise adm inistration of her practical
affairs in a com plex social environm ent. This is as genuine a case of
learning about objective reality as one finds an yw here. It is also of
fundam ental im portance for the character and quality of any indi
vidual's life, and not everyone succeeds equally in m astering the rel
ev an t intricacies. O n these points I exp ect little disagreem ent. W hat
is problem atic is w heth er this p rocess am ou nts to the learning of
genuine M oral Truth, or to m ere socialization. W e can hardly collapse
the distinction, lest w e m ake m oral criticism of diverse forms of social
organization impossible. W e w ant to defend this possibility, since, as
N ietzsche w ould be the first to point out, the socialization described
above can occasionally am ou nt to a cow ardly acquiescence in an arbit
rary and stultifying form of life. C an w e specify un d er w hat circum
stances it will am ount to som ething m ore than this?
3 Praxis, Theoria, and Progress
W e m ay begin to an sw er this question by noting that an exactly paral

lel problem arises with regard to the learning of Scientific Truth. In
school and university w e are taught to recognize a panoply of co m
plex prototypical situations— falling bodies, forces at equilibrium,
oxidation, nuclear fission, the green hou se effect, bacterial infection,
e tc .— and w e are taught to anticipate the prototypical elem ents and
effects of each. This is unquestionably a process of learning. O n this
there will be little disagreem ent. But it is just as clearly a p rocess of
socialization, a p rocess of adopting the conceptual m achinery of an
an teced en t society, thereafter to function sm oothly w ithin it. W h at is
problem atic is w heth er this process am ou nts to the learning of
genuine Scientific Truth, or to m ere socialization. W e dare not
collapse the distinction, lest w e m ake scientific criticism of diverse
theories im possible. W e w ish to defend that possibility, since, as his
tory record s, such socialization can am ou nt to a m yopic acquiescence
in a foolish and im potent w orld view . C an w e specify u n d er w hat

circu m stan ces it will am ou nt to som ething m ore than this?
W e m u st be careful not to set our hurdle too high here. A decade of
scrutiny has found the traditional form s of scientific realism to be very
wnrealistic in their optim istic expectations of universal con vergen ce
on a unique set of Final Truths (Laudan 1981). The lessons of scientific
history, the collapse of naive argum ents, and the lessons of cognitive
neurobiology (chapter 9, section 7) all invite a different accounting of
the p ro sp ects of the scientific enterprise. A certain am ou nt of con cep
tual radiation, rather than con vergen ce, seem s the m ore likely and the
m ore healthy long-term fate of science. For the enterprise m ay have
no unique final goal, just as biological evolution has no unique final
goal. M oreover, the evaluation of our cognitive resou rces and com
m itm ents at any given time will have to look beyond such parochial
virtues as Tarskian truth, since that is a feature unique to the p aro
chial elem ents of h u m an language, w hich is a peripheral m edium of
rep resentation even for hum an cognition. It is unlikely to be of any
fundam ental significance in evaluating cognitive activity in creatures
generally.
The radical character of these recent view s aside, there rem ains
every reason to think that the norm al learning p rocess, as instanced
both in individuals and in the collective enterprise of institutional sci
ence, involves a reliable and dram atic increase in the am ou nt and the
quality of the inform ation w e have about the w orld. H ow to rechar
acterize our cognitive virtues is still an unsettled question, but the
neurocom putational fram ew ork discussed in the earlier chapters
show s us how to begin, and it leaves the existence of high-grade
learning, intricate w orld m odeling, and fertile cognitive futures in no
serious doubt.
W h en such pow erful learning netw orks as hum ans are confronted
w ith the problem of how best to perceive the social w orld, and how
best to con du ct on e's affairs w ithin it, w e have equally good reason to
expect that the learning p rocess will show an integrity com parable
to that show n on other learning tasks, and will p rodu ce cognitive
achievem ents as robust as those p rodu ced anyw here else. This
expectation will be especially apt if, as in the case of "scien tific"
know ledge, the learning process is collective and the results are
transm itted from generation to generation. In that case w e have a
continuing society und er constant pressure to refine its categories of
social and m oral perception, and to m odify its typical respon ses and
exp ectation s. Successful societies do this on a system atic basis. A
body of legislation accum ulates, w ith successive additions, deletions,
and m odifications. A body of case law accum ulates, and the techni
que of finding and citing relevant preced en ts (which are, of course,

prototypes) becom es a central feature of adjudicating legal disputes.
Ju st w hat are the m em bers of the society learning? They are learn
ing how best to organize and adm inister their collective and indi
vidual affairs. W h at factors provoke change and im provem ent in their
typical categories of m oral perception and their typical form s of b e
havioral response? That is, w hat factors drive m oral learning? They
are m any and various, but in general they arise from the continuing
social experience of conducting a life u n d er the existing m oral
fram ew ork. That is, m oral learning is driven by social experience,
often a long and painful social experience, just as theoretical science is
driven by experim ent. M oral know ledge thu s has just as genuine a
claim to objectivity as any other kind of em pirical know ledge. W h at
are the principles by w hich rational people adjust their m oral con cep
tions in the face of unw elcom e social experience? They are likely to be
exactly the sam e "p rin cip les" that drive conceptual readjustm ent in
science or anyw here else, and they are likely to be revealed as w e
com e to un derstand how empirical brains actually do learn (see
chap ter 11).
F rom w hat little w e already know about learning in neural net
w orks, w e can already see that "su p erem p irical" virtues such as
simplicity and conceptual unity play a role that is com parable in im
p ortan ce to adequately com prehending the data of experience. W e
m ay, therefore, exp ect the sam e to be true in the dom ain of m oral
know ledge, and this expectation is born out by the character of the
m ost prom inent substantive m oral theories. Ju st as scientific theories
do, substantive m oral theories attem pt to find or im pose a unity on
the scattered concepts and convictions that m ake up o n e's m oral con
sciousness. They attem p t to accou n t for the full range of our m oral
understan din g in term s of a small num ber of allegedly m ore fun
d am ental con cep ts, either m oral or nonm oral.
This im pulse to unity w as discussed briefly in ch apter 10, in the
con text of intertheoretic reduction and conceptual unification. There
the p rocess w as characterized as the search for a su perordinate p ro
totyp e, of w hich the target prototyp es can be usefully seen as various
instances. That sam e character is evident in existing m oral theories.
O ne of the m ost primitive attem pts at providing an explanatory unity
to o n e's m oral understanding portrays all m oral im peratives as the
com m and s of a stern (but loving) supernatural father. The stern fa th er
p rototype is one that alm ost everyone p ossesses. It has certain salient
features that m ake it appealing to m any millions of people. A n d it
provides a m oral content they can u n d erstan d. If w e are all His chil
dren , for exam ple, then every m an is your b roth er and every w om an
you r sister. This activates the "sibling" p rototyp e m ost of u s possess,

w hich d em and s better treatm ent for siblings than one accords to n on
family m em b ers, and so forth. In these w ays, and oth ers, one is in
vited to reconceive on e's society as being an instance of a family, w ith
all of the perceptions and im pulses that com e with th at dom estic
prototyp e.
Less primitive attem pts at explanatory understanding appeal to
different prototyp es. A m ore arresting account bids one reconceive
on e's relation to society as being an instance of the p rototyp e, party to
a contract, w ith all that that suggests. A different but closely related
accou n t bids one conceive all of one's m oral decisions und er the
p ro to ty p e, m axim izing private benefit under collective constraints. A third
attem p ts to unify all of one's m oral p recep ts by seeing th em as in
stances of universalizable rules. A fourth claims unity for them as
m axim izin g gen eral utility. A nd so on. This im pulse to unity is en
tirely healthy, in our m oral understanding as well as in o u r scientific
un d erstan din g. The virtue of unified theories here, as elsew here,
is that th ey help to d eepen our understanding and help us to deal
better w ith novel social situations. It rem ains an open question, on
w hich I shall not here tend er com m ent, as to w hich of the superordi
nate p rototyp es just m entioned effects the deepest and m ost lasting
unification of our understanding.
My point by now is clear, or as clear as I can m ake it in short com
pass, and I shall bring exploration of these parallels to a close. The
ap p roach of this ch apter does not tell us anything substantive about
w hich m oral theory is correct. But it does indicate that m oral know l
edge is as genuine as know ledge elsew here, and that m oral progress
is possible. There is no reason w hy our m oral consciousness and
m oral u n d erstan din g should not continue to im prove and deepen
indefinitely, just as our nonsocial perception and our theoretical
science m ay do so. For in fact w e do have an organ for understan din g
and recognizing m oral facts. It is called the brain.
References
Allman, J. M ., et al. 1981. "Visual Topography and Function." In C. N. Woolsey, ed.,

Cortical Sensory Organization, vol. 2. Clifton, N .J.: Humana Press.
Arbib, M ., and Amari, S. 1985. "Sensorimotor Transformations in the Brain." Journal o f
Theoretical Biology 112:123-155.
Ballard, D. H. 1986. "Cortical Connections and Parallel Processing: Structure and Func
tion." Behavior and Brain Sciences 9, no. 1: 67-90.
Barto, A. G. 1985. "Learning by Statistical Cooperation of Self-Interested Neuron-like
Computing Elem ents." Human Neurobiology 4:229-256.
Bartoshuk, L. M. 1978. "Gustatory System ." in R. B. Masterton, ed., Handbook o f
Behavioral Neurobiology, vol. 1, Sensory Integration, pp. 503-567. New York: Plenum
Press.
Bear, M. F., Cooper, L. N ., and Ebner, F. F. 1987. "A Physiological Basis for a Theory of
Synapse M odification." Science 237, no. 4810.
Bechtel, W ., and Abrahamsen, A. Forthcoming. Connectionism and the Mind: An Intro
duction to Parallel Distributed Processing. Oxford: Basil Blackwell.
Block, N. 1978. "Troubles with Functionalism." In C. W. Savage, ed., Perception and
Cognition: Issues in the Foundations o f Psychology, Minnesota Studies in the Philoso
phy of Science, vol. 9, pp. 261-325.
Block, N ., and Fodor, J. 1972. "W hat Mental States Are N ot." Philosophical Review
81:159-181.
Bloor, D. 1976. Knowledge and Social Imagery. Boston: Routledge & Kegan Paul.
Brandt, R., and Kim, J. 1967. "The Logic of the Identity Theory." journal o f Philosophy
64, no. 17: 515-537.
Brody, B. A. 1972. "Towards an Aristotelian Theory of Scientific Explanation." Philoso
phy o f Science 39:20-31.
Bromberger, S. 1966. "W hy-Questions." In R. Colodny, ed., Mind and Cosmos, pp.
86-111. Pittsburgh: Pittsburgh University Press.
Bruner, J. 1973. "O n Perceptual Readiness." In J. Anglin, ed., Beyond the Information
Given, pp. 7-42. New York: W. W. Norton & Co.
Campbell, K. 1983. "Abstract Particulars and the Philosophy of M ind." Australasian
Journal o f Philosophy 61, no. 2 :129-141.
Carnap, R. 1956. "Empiricism, Semantics, and O ntology." In R. Carnap, Meaning and
Necessity, 2d edition, pp. 205-221. Chicago: University of Chicago Press.
Churchland, P. M. 1970. "The Logical Character of Action Explanations." Philosophical
Review 79, no. 2: 214-236.
Churchland, P. M. 1975a. "Two Grades of Evidential Bias." Philosophy o f Science 42,
no. 3: 250-259.
Churchland, P. M. 1975b. "Karl Popper's Philosophy of Science." Canadian Journal o f
Philosophy 5, no. 1: 145-156.
306 References
Churchland, P. M. 1979. Scientific Realism and the Plasticity o f Mind. Cambridge: Cam
bridge University Press.
Churchland, P. M. 1980. "Plasticity: Conceptual and Neuronal," Behavior and Brain
Sciences 3, no. 1:133-134.
Churchland, P. M. 1981a. "Eliminative Materialism and the Propositional Attitudes."
Journal o f Philosophy 78, no. 2: 67-90.
Churchland, P. M. 1981b. "Functionalism, Qualia, and Intentionality." Philosophical
Topics 12, no. 1: 121-145.
Churchland, P. M. 1982. "Is Thinker a Natural Kind?" Dialogue 21, no. 2: 223-238.
Churchland, P. M. 1984. Matter and Consciousness. Cambridge: MIT Press.
Churchland, P. M. 1985a. "The Ontological Status of Observables: In Praise of the
Superempirical Virtues." In P. M. Churchland, and C. A. Hooker, eds., Images o f
Science, pp. 35-47. Chicago: University of Chicago Press.
Churchland, P. M. 1985b. "Reduction, Qualia, and the Direct Introspection of Brain
States." Journal o f Philosophy 82, no. 1: 8-28.
Churchland, P. M. 1986a. "Som e Reductive Strategies in Cognitive Neurobiology."
Mind 95, no. 379: 279-309.
Churchland, P. M. 1986b. "O n the Continuity of Science and Philosophy." Mind and
Language 1, no. 1: 5-14.
Churchland, P. M. 1986c. "Firewalking and Physics." The Skeptical Inquirer 10, no. 3:
284-285.
Churchland, P. M. 1986d. "Cognitive Neurobiology: A Computational Hypothesis for
Laminar Cortex." Biology and Philosophy 1, no. 1: 25-51.
Churchland, P. M. 1986e. "Sem antic Content: In Defense of a Network Approach." The
Behavioral and Brain Sciences 9, no. 1: 139-140.
Churchland, P. M. 1988a. "Folk Psychology and the Explanation of Human Behavior."
Proceedings o f the Aristotelian Society, suppl. 62:209-221.
Churchland, P. M. 1988b. "Perceptual Plasticity and Theoretical Neutrality: A Reply to
Jerry Fodor." Philosophy o f Science 55, no. 2: 167-187.
Churchland, P. M. 1989a. "O n the Nature of Theories: A Neurocomputational Perspec
tive." In C. W. Savage, ed., The Nature o f Theories, Minnesota Studies in the Phi
losophy of Science, vol. 14. Minneapolis: University of Minnesota Press.
Churchland, P. M. 1989b. "Simplicity: The View from the Neuronal Level." In N.
Rescher, ed., Aesthetic Values in Science. Forthcoming.
Churchland, P. M ., and Churchland, P. S. 1983. "Content: Semantic and Information-
Theoretic." The Behavioral and Brain Sciences 6, no. 1: 67-68.
Churchland, P. M ., and Hooker, C. A., eds. 1985. Images o f Science: Essays on Realism
and Empiricism. Chicago: University of Chicago Press.
Churchland, P. S. 1978. "Fodor on Language Learning." Syntheses 38, no. 1: 149-159.
Churchland, P. S. 1980a. "A Perspective on Mind-Brain Research." Journal o f Philosophy
77, no. 4: 185-207.
Churchland, P. S. 1980b. "Language, Thought, and Information Processing." No Us
14:147-169.
Churchland, P. S. 1980c. "Neuroscience and Psychology: Should the Labor Be Di
vided?" Behavioral and Brain Sciences 3, no. 1: 133.
Churchland, P. S. 1981. "Is Determinism Self-Refuting?" Mind 90:99-101.
Churchland, P. S. 1986. Neurophilosophy: Toward a Unified Understanding o f the Mind-
Brain. Cambridge: MIT Press.
Cynader, M., and Berman, N. 1972. "Receptive Field Organization of Monkey Superior
. Colliculus." Journal o f Neurophysiology 35:187-201.
Davidson, D. 1973. "O n the Very Idea of a Conceptual Schem e." Proceedings and
Addresses o f the American Philosophical Association 47.
References 307
Dennett, D. C. 1971. "Intentional System s." Journal o f Philosophy 68, no. 4: 87-106.
Reprinted in D. C. Dennett, Brainstorms, pp. 3 -2 2 (Montgomery, Vt.: Bradford
Books, 1978).
Dennett, D. C. 1975. "Brain Writing and Mind Reading." In K. Gunderson, ed.,
Language, Mind, and Knowledge, Minnesota Studies in the Philosophy o f Science, vol. 7,
pp. 403-415. Minneapolis: University of Minnesota Press.
Dennett, D. C. 1981. "Three Kinds of Intentional Psychology.” In R. Healey, ed.,
Reduction, Time, and Reality, pp. 37-61. Cambridge: Cambridge University Press.
Reprinted in D. C. Dennett 1987.
Dennett, D. C. 1987, The Intentional Stance. Cambridge: MIT Press.
Desmond, N ., and Levy, W. 1983. "Synaptic Correlates of Associative Potentiation/
Depression: An Ultrastructural Study in the Hippocampus." Brain Research 265:21-
30.
Desn.^nd, N ., and Levy, W. 1986. "Changes in the Numerical Density of Synaptic
Contacts with Long-Term Potentiation in the Hippocampal Dentate Gyrus." Jour
nal o f Comparative Neurology 253:466-475.
Dretske, F. 1981. Knowledge and the Flow o f Information. Cambridge: MIT Press.
Dreyfus, H. 1979. What Computers Can't Do. New York: Harper & Row.
Dreyfus, H ., and Dreyfus, S. 1986. Mind over Machine. New York: Macmillan.
Fales, E. 1982. "Natural Kinds and Freaks of N ature." Philosophy o f Science 49, no. 1:
67-90.
Feyerabend, P. K. 1963a. "Materialism and the Mind-Body Problem ." Review o f
Metaphysics 17:49-66. Reprinted in C. V. Borst, ed., The Mind/Brain Identity Theory,
pp. 142-156. Toronto: Macmillan, 1970.
Feyerabend, P. K. 1963b. "H ow to Be a Good Empiricist." In B. Baumrin, ed., Philoso
phy o f Science, vol. 2, pp. 3-19. New York: Interscience Publications. Reprinted in
H. Morick, ed., Challenges to Empiricism (Belmont, California: Wadsworth, 1972),
pp. 164-193.
Feyerabend, P. K. 1965. "Reply to Criticism: Comments on Smart, Sellars, and Put
n am ." In M. Wartofsky, ed., Boston Studies in the Philosophy o f Science. New York:
Reidel. Reprinted in P. K. Feyerabend, Realism, Rationalism and Scientific Method,
Philosophical Papers, vol. 1 (Cambridge: Cambridge University Press, 1981), pp.
104-131.
Feyerabend, P. K. 1969. "Science without Experience." Journal o f Philosophy 66, no. 22:
791-794.
Feyerabend, P. K. 1970. "Against Method: Outline of an Anarchist Theory of Knowl
edge." In Radner and Winokur, eds., Analyses o f Theories and Methods o f Physics and
Psychology. Minnesota Studies in the Philosophy o f Science, vol. 4, pp. 17-130. Min
neapolis: University of Minnesota Press.
Feyerabend, P. K. 1980. "Consolations for the Specialist." In I. Lakatos, and A. Mus-
grave, eds., Criticism and the Growth o f Knowledge. Cambridge: Cambridge Universi
ty Press: 197-230.
Fodor, J. A. 1968. Psychological Explanation. New York: Random House.
Fodor, J. A. 1975. The Language o f Thought. New York: Crowell.
Fodor, J. A. 1984. "Observation Reconsidered." Philosophy o f Science 51, no. 1: 23-43.
Fodor, J. A.1988a. Psychosemantics: The Problem o f Meaning in the Philosophy o f Mind.
Cambridge: MIT Press.
Fodor, J. A. 1988b. "A Reply to Churchland's 'Perceptual Plasticity and Theoretical
N eutrality.'" Philosophy o f Science 55, no. 2: 188-194.
Fodor, J. A. and Pylyshyn, Z. 1981. "How Direct Is Visual Perception? An Examination
of Gibson's 'Ecological A pproach.'" Cognition 9, no. 2: 139-196.
308 References
Friedman, M. 1974. "Explanation and Scientific Understanding." Journal o f Philosophy

71, no. 1: 5-18.
Gazzaniga, M. S., and LeDoux, J. E. 1975. The Integrated Mind. New York: Plenum
Press.
Giere, R. 1988. Explaining Science: A Cognitive Approach. Chicago: University of Chicago
Press.
Glymour, C. 1988. "Artificial Intelligence Is Philosophy." In J. Fetzer, ed., Aspects o f
Artificial Intelligence. Dordrecht, Holland: D. Reidel.
Glymour, C. 1985. "Explanation and Realism." In P. M. Churchland, and C. A. Hook
er, eds., Images o f Science, pp. 99-117. Chicago: University of Chicago Press.
Goldberg, M ., and Robinson, D. L. 1978. "Visual System: Superior Colliculus." in R.
Masterson, ed., Handbook o f Behavioral Neurobiology, vol. 1, pp. 119-164. New York:
Plenum Press.
Goldman, A. Forthcoming. "Interpretation Psychologized." Mind and Language.
Gonshor, A., and Jones, G. M. 1976. "Extreme Vestibulo-Ocular Adaptation Induced
by Prolonged Optical Reversal of V ision." Journal o f Physiology 256:381-414.
Gordon, B. 1973. "Receptive Fields in Deep Layers of Cat Superior Colliculus." Journal
o f Neurophysiology 36:157-178.
Gordon, R. 1986. "Folk Psychology as Simulation." Mind and Language 1, no. 2: 158-
171.
Gorman, R. P., and Sejnowski, T. J. 1988a. "Learned Classification of Sonar Targets
Using a Massively-Parallel N etwork." IEEE Transactions: Acoustics, Speech, and Sig
nal Processing.
Gorman, R. P., and Sejnowski, T. J. 1988b. "Analysis of Hidden Units in a Layered
Network Trained to Classify Sonar Targets." Neural Networks 1:75-89.
Gregory, R. 1966. Eye and Brain. London: McGraw Hill.
Gregory, R. 1970a. The Intelligent Eye. New York: McGraw Hill.
Gregory, R. 1970b. "The Grammar of Vision." Listener 83, no. 2133: 242-246. Reprinted
in Gregory 1974.
Gregory, R. 1974. Concepts and Mechanisms o f Perception. New York: Charles Scribners
and Sons.
Grover, D ., Camp, J., and Belnap, N. 1975. "A Prosentential Theory of Truth." Philo
sophical Studies 27, no. 2: 73-125.
Haldane, J. 1988. "Understanding Folk." Proceedings o f the Aristotelian Society, suppl.
62:222-246.
Hanson, N. R. 1961. Patterns o f Discovery. Cambridge: Cambridge University Press.
Harman, G. 1965. "Inference to the Best Explanation." Philosophical Review 74:88-95.
Harman, G. 1973. Thought. Princeton: Princeton University Press.
Hartshorne, C., and Weiss, P. 1935. Collected Papers o f Charles Sanders Peirce. Cam
bridge: Harvard University Press.
Hatfield, G. 1989. "Computation, Representation, and Content in Noncognitive
Theories of Perception." In S. Silvers, ed., Representation, pp. 255-288. Dordrecht:
Kluwer.
Hebb, D. O. 1949. The Organization o f Behavior. New York: Wiley.
Hempel, C. 1965. "Studies in the Logic of Confirmation." In C. Hempel, Aspects o f
Scientific Explanation, pp. 3-52. New York: Free Press.
Hesse, M. 1966. Models and Analogies in Science. Notre Dame: Notre Dame University
Press.
Hinton, G. E. Forthcoming. "Connectionist Learning Procedures." Artificial Intelligence.
Hinton, G. E., and Sejnowski, T. J. 1986. "Learning and Relearning in Boltzmann
M achines." in D. E. Rumelhart, J. McClelland, and the PDP Research Group, Para
llel Distributed Processing: Explorations in the Microstructure o f Cognition. Cambridge:
MIT Press.
References 309
Hooker, C. A. 1975. "The Philosophical Ramifications of the Information Processing

Approach to the Mind-Brain." Philosophy and Phenomenological Research 36:1-15.
Hooker, C. A. 1981. "Towards a General Theory of Reduction." Dialogue 20:38-59,
201-236, 496-529.
Hooker, C. A. 1985. "Surface Dazzle, Ghostly Depths: An Exposition and Critical Eval
uation of van Fraassen's Vindication of Empiricism against Realism ." In P. M.
Churchland, and C. A. Hooker, eds., Images o f Science, pp. 153-196. Chicago: Uni
versity of Chicago Press.
Hooker, C. A. 1987. A Realistic Theory o f Science. Albany: State University of New York
Press.
Hopfield, J. J., and Tank, D. 1985. "'N eural' Computation of Decisions in Optimization
Problem s." Biological Cybernetics 52:141-152.
Hornik, K ., Stinchcombe, M ., and White, H. 1989. "Multi-Layer Feedforward Net
works Are Universal Approximators." Neuronetworks 2, forthcoming.
Hubei, D. H., and Livingstone, M. S. 1987. "Segregation of Form, Color, and Stereop-
sis in Primate Area 18." Journal o f Neuroscience 7:3378-3415.
Hubei, D. H ., and Wiesel, T. N. 1962. "Receptive Fields, Binocular Interactions, and
Functional Architecture in the Cat's Visual Cortex." Journal o f Physiology 160:106-
154.
Huerta, M. F., and Harting, J. K. 1984. "Connectional Organization of the Superior
Colliculus." Trends in Neuroscience 7, no. 8: 286-289.
Jackson, F. 1982. "Epiphenomenal Q ualia." Philosophical Quarterly 32, no. 127:127-136.
Jackson, F. 1986. "W hat Mary Didn't Know." Journal o f Philosophy 83, no. 5: 291-295.
Kanaseki, T., and Sprague, J. M. 1974. "Anatomical Organization of Pretectal Nuclei
and Tectal Laminae in the C at." Journal o f Comparative Neurology 158:319-337.
Kelso, S. R., Ganong, A. H ., and Brown, T. H. 1986. "Hebbian Synapses in Hippocam
p u s." Proceedings o f the National Academy o f Sciences 83:5326-5330.
Kitcher, P. 1981. "Explanatory Unification." Philosophy o f Science 48, no. 4: 507-531.
Kitcher, P. 1989. "Explanatory Unification and the Causal Structure of the W orld." In
P. Kitcher, ed., Scientific Explanation, Minnesota Studies in the Philosophy of Scien
ce, vol. 13.
Konishi, M. 1986. "Centrally Synthesized Maps of Sensory Space." Trends in Neuro
science 9, no. 4: 163-168.
Kottenhoff, H. 1957. "Situational and Personal Influences on Space Perception with
Experimental Spectacles." Acta Psychologica 13:79-97.
Kripke, Saul. 1972. "Naming and Necessity.” In D. Davidson and G. Harman, eds.,
Semantics o f Natural Language, pp. 253-355. Dordrecht, Holland: D. Reidel.
Kuhn, T. S. 1962. The Structure o f Scientific Revolutions. Chicago: University of Chicago
Press.
Lakatos, I. 1970. "Falsification and the Methodology of Scientific Research Pro
gram m es." In I. Lakatos and A. Musgrave, eds., Criticism and the Growth of
Knowledge. Cambridge: Cambridge University Press.
Lakoff, G. 1987. Women, Fire, and Dangerous Things. Chicago: University of Chicago
Press.
Land, E. 1977. "The Retinex Theory of Color V ision." Scientific American, December,
108-128.
Laudan, L. 1981. "A Confutation of Convergent Realism ." Philosophy o f Science 48, no.
1: 19-49.
Lehky, S., and Sejnowski, T. J. 1988a. "Computing Shape from Shading with a Neural
Network M odel." In E. Schwartz, ed., Computational Neuroscience. Cambridge: MIT
Press.
310 References
Lehky, S ., and Sejnowski, T. J. 1988b. "Network Model of Shape-from-Shading: Neu

ral Function Arises from Both Receptive and Projective Fields." Nature 333 (June
2): 452-454.
Leikind, B. J., and McCarthy, W. J. 1985. "A n Investigation of Firewalking." The Skep
tical Inquirer 10, no. 1: 23-35.
Lewis, D. 1983. "Postscript to 'Mad Pain and Martian P a in .'" Philosophical Papers, vol.
1. New York: Oxford.
Linsker, R. 1986. "From Basic Network Principles to Neural Architecture: Emergence of
Orientation Colum ns." Proceedings o f the National Academy o f Sciences 83:8779-8783.
Linsker, R. 1988. "Self-Organization in a Perceptual N etw ork." Computer, March.
Livingstone, R. B. 1978. Sensory Processing, Perception, and Behavior. New York: Raven
Press.
Livingstone, M. S ., and Hubei, D. H. 1987. "Psychophysical Evidence for Separate
Channels for the Perception of Form, Color, Movement, and D epth." Journal o f
Neuroscience 7:3416-3468.
Llinas, R. 1975. "The Cortex of the Cerebellum ." Scientific American, January, 56-71.
Llinas, R. 1986. "'M indness' as a Functional State of the Brain." In C. Blakemore and S.
Greenfield, eds., Mind and Matter. Oxford: Blackwell.
Lycan, W. G. 1985. "Tacit Belief." In R. J. Bogdan, ed., Belief. Oxford: Oxford Universi
ty Press.
Margolis, J. 1978. Persons and Minds. Dordrecht: Reidel.
Mays, L. E., and Sparks, D. L. 1980. "Saccades Are Spatially, Not Retinocentrically,
Coded." Science 208:1163-1165.
Mcllwain, J. T. 1975. "Visual Receptive Fields and Their Images in the Superior Collicu
lus of the C at." Journal o f Neurophysiology 38:219-230.
Mcllwain, J. T. 1984. Abstracts: Society fo r Neuroscience 10, part 1: 268.
McKloskey, M. 1983. "Intuitive Physics." Scientific American, April, 122-130.
Mellor, D. H. 1977. "Natural Kinds." British Journal for the Philosophy o f Science 28:229-
312.
Meredith, M. A ., and Stein, В. E. 1985. "D escending Efferents from the Superior Col
liculus Relay Integrated Multisensory Inform ation." Science 227, no. 4687: 657-659.
Merzenich, M ., and Kaas, J. 1980. "Principles of Organization of Sensory-Perceptual
Systems in M amm als." Progress in Psychobiology and Physiological Psychology 9:1-42.
Minsky, M. 1981. "A Framework for Representing Knowledge." In J. Haugeland, ed.,
Mind Design. Cambridge: MIT Press.
Minsky, M ., and Papert, S. 1969. Perceptrons. Cambridge: MIT Press.
Mooney, R. D ., et al. 1984. "Dendrites of D eep Layer, Somatosensory Superior Collicu-
lar Neurons Extend Into the Superficial Layer." Abstracts: Society fo r Neuroscience
10, part 1: 158.
Munevar, G. 1981. Radical Knowledge. Indianapolis: Hackett.
Musgrave, A. 1985. "Realism versus Constructive Empiricism." In P. M. Churchland
and C. A. Hooker, eds., Images o f Science, pp. 197-221. Chicago: University of
Chicago Press.
Nagel, E. 1961. The Structure o f Science. N ew York: Harcourt, Brace & World.
Nagel, T. 1974. "W hat Is It like to Be a B at?" Philosophical Review 83, no. 4: 435-450.
Nemirow, L. 1980. Review of Thomas N agel, Mortal Questions. Philosophical Review 89,
no. 3: 473-477.
Pellionisz, A. 1984. "Tensorial Aspects o f the Multi-Dimensional Approach to the
Vestibulo-Oculomotor Reflex." In A. Berthoz and E. Melvill-Jones, eds., Reviews in
Oculomotor Research. Elsevier.
Pellionisz, A ., and Llinas, R. 1979. "Brain Modelling by Tensor Network Theory and
Computer Simulation: The Cerebellum, Distributed Processor for Predictive Coor
dination." Neuroscience 4:323-348.
References 311
Pellionisz, A., and Llinas, R. 1982. "Space-Time Representation in the Brain: The Cere
bellum as a Predictive Space-Time Metric Tensor." Neuroscience 7, no. 12: 2949-
2970.
Pellionisz, A., and Llinas, R. 1985. "Tensor Network Theory of the Metaorganization of
Functional Geometries in the Central Nervous System ." Neuroscience 16, no. 2:
245-274.
Pfaff, D. W ., ed. 1985. Taste, Olfaction, and the Central Nervous System. New York: Rock
efeller University Press.
Pickering, A. 1981. "The Hunting of the Quark." Isis 72:216-236.
Pickering, A. 1984. Constructing Quarks: A Sociological History o f Particle Physics. Chicago:
University of Chicago Press.
Popper, K. 1959. The Logic o f Scientific Discovery. New York: Harper & Row.
Popper, K. 1972. Objective Knowledge. New York: Oxford University Press.
Popper, K ., and Eccles, J. 1978. The Self and Its Brain. New York: Springer Verlag.
Putnam, H. 1964. "Robots: Machines or Artificially Created Life?" Journal o f Philosophy
61, no. 11: 668-691.
Putnam, H. 1970. "Is Semantics Possible?" In H. Kiefer, and M. Munitz, eds., Lan
guages, Belief, and Metaphysics. Albany: State University of New York Press.
Reprinted in H. Putnam, Mind, Language and Reality, pp. 139-152 (Cambridge:
Cambridge University Press).
Putnam, H. 1971. "The Nature of Mental States." In D. Rosenthal, ed., Materialism and
the Mind-Body Problem. New Jersey: Prentice-Hall. Reprinted in H. Putnam, Mind,
Language, and Reality, pp. 429-440 (Cambridge: Cambridge University Press).
Putnam, H. 1975. "The Meaning of 'M eaning.'" In K. Gunderson, ed., Language, Mind
and Knowledge, Minnesota Studies in the Philosophy of Science, vol. 7. Reprinted in
H. Putnam, Mind, Language and Reality, pp. 215-271 (Cambridge: Cambridge Uni
versity Press).
Putman, H. 1981. Reason, Truth, and History. Cambridge: Cambridge University Press.
Pylyshyn, Z. 1980. "Computation and Cognition: Issues in the Foundation of Cogni
tive Science." Behavioral and Brain Sciences 3, no. 1: 111-134.
Quine, W. V. O. 1960. Word and Object. Cambridge: MIT Press.
Quine, W. V. O. 1969. "Natural Kinds." In W. V. O. Quine, Ontological Relativity and
Other Essays. New York: Columbia University Press.
Randi, J. 1982. Flim-Flam! Psychics, ESP, Unicorns, and Other Delusions. Buffalo, N .Y.:
Prometheus Books.
Risset, J. C., and Wessel, D. L. 1982. "Exporation of Timbre by Analysis and Syn
thesis." In D. Deutsch, ed., The Psychology o f Music. New York: Academic Press.
Robinson, D. A. 1972. "Eye Movements Evoked by Collicular Stimulation in the Alert
M onkey." Vision Research 12:1795-1808.
Robinson, H. 1982. Matter and Sense. New York: Cambridge University Press.
Rock, I. 1983. The Logic o f Perception. Cambridge: MIT Press.
Rogers, L. J., and Miles, F. A. 1972. "Centrifugal Control of Avian Retina: Effects of
Lesions of the Isthmo-Optic Nucleus on Visual Behaviour." Brain Research 48:147-
156.
Rorty, R. 1965. "Mind-Body Identity, Privacy, and Categories." Review o f Metaphysics
19:24-54.
Rosch, E. 1981. "Prototype Classification and Logical Classification: The Two Sys
tem s." In E. Scholnick, ed., New Trends in Cognitive Representation: Challenges to
Piaget's Theory. New Jersey: Lawrence Erlbaum.
Rosenberg, C. R., and Sejnowski, T. J. 1987. "Parallel Networks That Learn to Pro
nounce English Text." Complex Systems 1:145-168.
Rosenblatt, F. 1959. Principles o f Neurodynamics. New York: Spartan Books.
312 References
Rumelhart, D. E., Hinton, G. E., and Williams, R. J. 1986a. "Learning Representations

by Back-Propagating Errors." Nature 323:533-536.
Rumelhart, D. E., Hinton, G. E., and Williams, R. J. 1986b. "Learning Internal Repre
sentations by Error Propagation." In D. E. Rumelhart, J. L. McClelland, and the
PDP Research Group, Parallel Distributed Processing: Explorations in the Microstruc
ture o f Cognition, vol. 1. Cambridge: MIT Press.
Sacks, J. G ., and Lindenberg, R. 1969. "Efferent Nerve Fibers in the Anterior Visual
Pathways in Bilateral Congenital Cyctic Eyeballs." American Journal o f Opthalmology
68:691-695.
Salmon, W. 1966. The Foundations o f Scientific Inference. Pittsburgh: University of
Pittsburgh Press.
Salmon, W. 1970. "Statistical Explanation." In R. Colodny, ed., The Nature and Function
o f Scientific Theories, pp. 173-231. Pittsburgh: Pittsburgh University Press.
Salmon, W. 1971. Statistical Explanation and Statistical Relevance. Pittsburgh: Pittsburgh
University Press.
Salmon, W. 1978. "W hy Ask, 'W hy'?" Proceedings and Addresses o f the American Philo
sophical Association 51:683-705.
Salmon, W. 1984. Scientific Explanation and the Causal Structure o f the World. Princeton:
Princeton University Press.
Sayre, K. M. 1986. "Intentionality and Information Processing: An Alternative Model
for Cognitive Science." Behavioral and Brain Sciences 9, no. 1: 121-138.
Schank, R., and Abelson, R. 1977. Scripts, Plans, Goals, and Understanding. New Jersey:
John Wiley and Sons.
Scheffler, I. 1963. The Anatomy o f Inquiry. New York: Knopf.
Schiller, P. 1984. "The Superior Colliculus and Visual Function." In I. Darian-Smith,
ed., Handbook o f Physiology, vol. 3, pp. 457-504.
Schiller, P., and Sandell, J. H. 1983. "Interactions between Visually and Electrically
Elicited Saccades before and after Superior Colliculus and Frontal Eye Field Abla
tions in the Rhesus M onkey." Experimental Brain Research 49:381-392.
Schiller, P., and Stryker, M. 1972. "Single-Unit Recording and Stimulation in Superior
Colliculus of the Alert Rhesus M onkey." Journal o f Neurophysiology 35:915-924.
Searle, J. 1980. "M inds, Brains, and Program s." Behavioral and Brain Sciences 3, no. 3:
417-457.
Sejnowski, T. J., Kienker, P. K., and Hinton, G. E. 1986. "Learning Symmetry Groups
with Hidden Units: Beyond the Perceptron." Physica D 22D:260-275.
Sellars, W. 1956. "Empiricism and the Philosophy of M ind." In H. Feigl and M.
Scriven, eds., The Foundations o f Science and the Concepts o f Psychology and
Psychoanalysis, Minnesota Studies in the Philosophy of Science, vol. 1. Minneapo
lis: University of Minnesota Press. Reprinted in W. Sellars, Science, Perception and
Reality, pp. 127-196 (London: Routledge and Kegan Paul, 1963).
Shapere, Dudley. 1982. "Reason, Reference, and the Quest for Knowledge." Philosophy
o f Science 49, no. 1: 1-23.
Shoemaker, S. 1975. "Functionalism and Q ualia." Philosophical Studies 27:291-315.
Smith, D. V. et al. 1983. "Coding of Taste Stimuli by Hamster Brain Stem N eurons."
Journal o f Neurophysiology 7:541-558.
Stanton, P. K., and Sejnowski, T. J. In preparation. "Associative Long-Term Depres
sion in the Hippocampus: Evidence for Anti-Hebbian Synaptic Plasticity."
Stein, В. E. 1984. "Development of the Superior Colliculus." In W. M. Cowan, ed.,
Annual Review o f Neuroscience 7:95-126.
. Stich, S. P. 1983. From Folk Psychology to Cognitive Science: The Case against Belief. Cam
bridge: MIT Press.
Stich, S. P. 1989. The Fragmentation o f Reason. Cambridge: MIT Press.
References 313
Strawson, P. F. 1958. "P ersons." In H. Feigl, M. Scriven, and G. Maxwell, eds., Con
cepts, Theories, and the Mind-Body Problem, Minnesota Studies in the Philosophy of
Science, vol. 2, pp. 330-353. Minneapolis: University of Minnesota Press.
Sturgeon, N. 1985. "Moral Explanation." In D. Copp, and D. Zimmerman, eds., Moral
ity, Reason, and Truth. Southern Journal o f Philosophy, suppl. to vol. 24.
Suppe, F. 1974. The Structure o f Scientific Theories. Chicago: University of Illinois Press.
Sur, M ., Garraghty, P., and Roe, A. 1988. "Experimentally Induced Visual Projections
into Auditory Thalamus and Cortex." Science 242, no. 4884: 1437-1441.
Taylor, C. 1970. "Mind-Body Identity: A Side Issue?" In C. V. Borst, ed., The Mind/
Brain Identity Theory, pp. 231-241. Toronto: Macmillan.
Taylor, C. 1987. "Overcoming Epistemology." In K. Baynes, J. Bohman, T. McCarthy,
eds., After Philosophy: End or Transformation? pp. 464-488. Cambridge: MIT Press.
Teller, P. 1974. "O n W hy-Questions." Nods 8:371-380.
Thompson, R. F. 1986. "The Neurobiology of Learning and M emory." Science 233:941-
947.
Van Essen, D. C., and Maunsell, J. 1983. "Hierarchical Organization and Functional
Streams in the Visual Cortex." Trends in Neuroscience 6:370-375.
Van Fraassen, B. 1977. "The Pragmatics of Explanation." American Philosophical Quarter
ly 14:143-150.
Van Fraassen, B. 1980. The Scientific Image. Oxford: Oxford University Press.
Van Fraassen, B. 1981. Critical Notice of Paul Churchland, Scientific Realism and the
Plasticity o f Mind. Canadian Journal o f Philosophy 11, no. 3: 555-567.
Wilkes, K. 1981. "Functionalism, Psychology, and the Philosophy of M ind." Philo
sophical Topics 12, no. 1.
Wilkes, K. 1984. "Pragmatics in Science and Theory in Common Sense." Inquiry 27, no.
4: 339-361.
Wilson, M. 1980. "The Observational Uniqueness of Some Theories." Journal o f Philoso
phy 77, no. 4: 208-232.
Wilson, M. 1985. "W hat Can Theory Tell Us about Observation?" In P. M. Churchland
and C. A. Hooker, eds., Images o f Science, pp. 222-242. Chicago: University of
Chicago Press.
Wolter, J. R. 1965. "The Centrifugal Nerves in Human Optic Tract, Chiasm, Optic
Nerve, and Retina." Transactions o f the American Opthalmological Society 63:678-707.
Wolter, J. R., and Lund, O. E. 1968. "Reaction of Centrifugal Nerves in the Human
Retina." American Journal o f Opthalmology 66:221-232.
Zeki, S. 1980. "The Representation of Colours in the Cerebral Cortex." Nature 284:412-
418.
Zeki, S. 1983. "Colour Coding in the Cerebral Cortex: The Reaction of Cells in Monkey
Visual Cortex to Wavelengths and Colours." Neuroscience 9, no. 4: 741-765.
Zipser, D ., and Elman, J. D. 1987. "Learning the Hidden Structure of Speech." Journal
o f the Acoustical Society o f America 83, no. 4: 1615-1626.
Index
Abduction, 219 Belnap, N ., 5

A braham sen, A ., 216 Berm an, N ., 94
Abstracta, 126 Bernoulli, D ., 242
Acalculia, 295 Block, N „ 2 9 -3 9
Action explanation, 2 ,1 1 4 , 115, 121 Bloor, D ., 248
A ctivation vector, 7 3 ,1 2 7 ,1 6 2 ,1 6 3 , Boltzm ann, L ., 242
1 6 4 ,1 6 8 ,1 7 4 , 1 7 8 ,1 8 1 ,1 8 9 , 206, Boltzm ann learning, 2 4 4 -2 4 5 , 248
232 Boyd, R ., 141
Artificial intelligence (AI), xiii, 14, Brandt, R ., 59
112, 1 2 2 ,1 2 5 , 127, 130, 1 35,154, Brody, B. A ., 224
156, 171 ,1 7 2 Brom berger, S ., 224
A lchem y, 9 ,1 3 - 1 5 ,1 8 Brow n, T. H ., 247
A llm an, J. M ., 96 Bruner, J ., 256
Am ari, S ., 101
A m pere, A .-M ., 242 Caloric, 48, 125, 141, 2 6 9 -2 7 5 , 284,
Am pliative inference, 145, 211. See 286
also Inference to the best explana Cam p, J., 5
tion Cam pbell, K ., 62
A nalogy, 3 ,1 0 6 ,1 1 9 ,1 2 0 ,1 2 6 , 219, Carnap, R „ 146, 278, 279
220, 226 C erebellum , 39, 68, 82, 98, 1 00,101,
Analytic/synthetic distinction, 273 1 0 7 ,1 8 2 -1 8 5
A nom alies, 8, 191-193, 241 Cerebral hem ispheres, 20, 21, 79, 80,
A phasia, 294 82, 94, 98
Arbib, M ., 101 Chinese nation, 34, 36, 37
A ristarchos, 249 Chinese room , 39 ff
A ttractors, 206 Chom sky, N ., 19, 109
Autom ated science, 250 Chords, musical, 53, 55, 65, 6 6 ,1 0 6 ,
A utonom y, 196 268
Axon. See Neuron Churchland, P. M ., 1, 2, 25, 28, 31,
45, 46, 49, 54, 55, 57, 7 5 ,1 0 3 , 111,
Back propagation, 172, 1 7 3 ,1 7 4 ,1 8 5 - 115, 124, 130, 131, 146, 154, 157,
187, 2 4 3 -2 4 6 , 248. See also 1 5 8 ,1 5 9 ,1 9 0 ,1 9 7 , 222, 256, 272,
Learning 274, 281, 286
Ballard, D. H „ 98 Churchland, P. S ., 7, 22, 23, 45, 46,
Barto, A . G ., 186 57, 109, 274
Bartoshuk, L. M ., 105 Clim bing fiber, 185
Bear, M . F ., 186 Cognitive im penetrability, 255 ff
Bechtel, W „ 216 Cognitive space, 232, 250, 253
Behaviorism , 3, 207 C oherence, 6, 27, 1 0 9 ,1 3 9 , 1 45-147
316 Index
Color, 24, 52, 6 5 -7 0 , 73, 102-105, Einstein, A ., 107, 242

123, 257, 260, 261, 292 Electrom agnetic (EM) waves, 30, 51,
C om putation, 77, 78, 9 2 ,1 0 0 , 101, 52, 53, 56, 57, 238, 285
127, 154, 156, 158, 1 72,179, 184, Elim inative materialism , 1, 6, 9, 11,
185, 204, 205, 209 15, 16, 21, 45, 125-127
Conceptual change, 7, 155, 178,195, Elm an, J., 167, 246
2 3 1 -2 3 4 , 236, 239, 241, 248 Em ergent properties, 5 0 -5 5 , 64
Conceptual configurations, 250 Empirical adequacy, 139, 142,143,
Conceptual fram ew ork, 1 -4 , 25, 28, 1 4 5 ,1 4 6 , 147, 149, 227, 228
5 0 -5 5 , 60, 64, 66, 7 5 ,1 0 6 ,1 2 0 ,1 2 1 , Em piricism , 139, 149, 150, 227, 228
147, 2 3 1 -234, 237, 253, 255, 268, Equivocation, 6 2 -6 3 , 6 6 -6 9 , 7 1 -7 2
270, 271, 277-2 7 9 Error, 14, 42, 60, 124, 165-177, 180,
Conceptual redeploym ent, 232, 2 3 6 - 1 8 5 ,1 8 6 ,1 8 9 -1 9 4 , 201, 202, 222,
242 2 4 3 -2 4 6 , 248, 253
Conceptual space, 231, 237, 239, 249, Euclid, 215, 216
253 Evolution, 16, 19, 26, 8 2 ,1 0 9 , 147,
Connectionism , 129, 130. See also 1 7 7 ,1 7 8 , 241, 278, 283, 294, 301
Neural netw orks Expert system s, 251 ff
C ontent Explanation
calibrational, 43, 275 in folk psychology, 2 -6 ,1 1 1 - 1 2 7 ,
translational, 43, 44 21 7 -2 1 8
C onvergent realism , 1 4 0,194, 2 9 3 - deductive-nom ological (D-N) model
295 of, 121, 199, 200, 215, 217, 223, 224,
Coordinate transform ation, 8 6 -9 5 , 227, 229
92, 97, 9 8 ,1 0 0 ,1 0 1 statistical-relevance (SR) model of,
C orrespondence rules, 4 7 -4 9 , 51 2 2 4 -2 2 5
Cortex, 55, 62, 64, 68, 69, 73, 7 8 -8 2 , prototype-activation (PA) model of,
9 1 -9 8 ,1 4 7 ,1 7 0 ,1 7 1 ,1 8 3 , 266, 267 2 0 9 -2 2 3
Crick, F ., xvii evaluation of, 22 0 -2 2 3
Cynader, M ., 94 pragm atics of, 22 6 -2 2 8
and perception, 19 8 -1 9 9 , 22 8 -2 2 9
D alton, J., 12 functional, 214
D avidson, D ., xii, 249 causal, 2 1 3 -2 1 4
D eductive-nom ological (D-N) ex moral, 2 1 6 -2 1 7
planation. See Explanation reductive, 21 4 -2 1 6
D elta rule, 1 7 2 ,1 7 3 ,1 7 4 ,1 8 4 ,1 8 6 , Explanatory understanding, 119—
1 8 7 ,1 9 2 ,1 9 3 , 202 1 2 1 ,1 2 5 , 1 97-200, 21 0 -2 1 4 , 218,
D endrite, 185 2 2 3 -2 2 9 , 303
D ennett, D. C „ xii, 1 0 ,1 0 9 ,1 2 5 -1 2 7 Extension, 49, 283, 284, 285, 286, 287
D escartes, R ., 238, 241
D escending pathw ays, 185, 266 Facial recognition, 107
D esm ond, N ., 237 Feigl, H ., xi
D retske, F., 276 Feyerabend, P. K ., xi, 1, 55, 76, 111,
D reyfus, H ., 251 147, 154, 190, 277
D reyfus, S ., 251 First-person account of mind, 74 ff
D ualism , mind/body, 26, 63, 71, 72 Fodor, J. A ., xii, 7, 1 0 ,1 1 , 2 9 ,1 5 8 ,
D yskinesia, 90 190, 229, 2 5 5 -2 5 7 , 270, 2 71-274,
27 7 -2 7 8
Eccles, J., 10 Folk psychology (FP), 2 - 6 ,1 7 , 22, 28,
Ectoplasm , 71 -7 2 35, 44, 45, 54, 55, 7 5 ,1 1 1 -1 1 8 ,1 2 1 ,
Index 317
122, 125-127, 135, 1 5 0 ,1 5 1 , 217, Identities, cross-theoretic, 4 7 -5 0 , 6 5 -

218, 225, 240 66
Foundations of empirical know l Identity theory, 59
edge, 6, 188, 255, 259, 268, 273 Illusory contours, 260-261
Fram e problem , 1 9 9 -2 0 0 ,1 5 5 -1 5 6 Incom m ensurability, 9, 43, 258, 268
Fresnel, A . J., 242 Indexicals, 285, 286, 287
Friedm an, M ., 227 Inference to the best explanation,
Functional persistence, 90, 95, 101, 2 1 1 ,2 1 8
178-179, 205 Instrum entalism , 126, 145, 150, 151,
Functionalism , 9 -4 5 158
Functors, predicate-form ing, 115 Intension, 42, 43, 60, 62, 281, 286, 287
Intensional context, 59
G anong, A. H ., 247 Intentionality, 2, 3, 4, 23, 28, 34, 3 9 -
Gazzaniga, M. S ., 20 44
G ib s o n ,]., 228 -2 2 9 Interneuron, 94, 184
G iere, R ., 158 Introspection, 1, 2, 3, 2 3 -3 4 , 47, 5 4 -
Glym our, C ., 1 42,172 61, 6 4 -6 7 , 75, 7 6 ,1 0 6
G oldberg, M „ 94, 97 Inverted spectrum . See Q ualia, gerry
Goldm an, A ., 119 m andered
G onshor, A ., 263
G ordon, B ., 94 Jackson, F „ 47, 55, 6 1 -7 5 ,1 0 2
G ordon, R ., 119 Jones, G. M ., 263
Gorm an, R. P ., 167 Jordan, L ., xvii, 108
Gradient descent, 166 ,1 9 2 , 201 Joule, J. P ., 242
Gregory, R „ 1 9 ,1 9 8 , 228, 256, 261
Grover, D ., 5 Kaas, J., 96
Kanaseki, T ., 93
Haldane, J„ 118 Kelso, S. R „ 247
H anson, N. R ., 256, 277 Kelvin, Lord, 242
H arm an, G ., 218 Kepler, J., 48, 241
Harting, J. K ., 94 Kim, J., 59
H artshorne, C ., 228 Kitcher, P ., xvii, 193, 227, 229
Hatfield, G „ 229 K onishi, M ., 96
H eat, 13, 213, 242, 282, 283, 285. See Kottenhoff, H ., 262
also Therm odynam ics Kripke, S ., 281, 282, 287
H ebb, D. O ., 186, 187, 246, 247, 248 Kuhn, T. S ., 1 2 2 ,1 5 6 ,1 5 8 ,1 5 9 ,1 9 1 ,
H egel, G „ 279 19 2 ,1 9 3 , 219, 229, 241, 256, 264,
Hem pel, C ., 154,199 278
H esse, M ., 219
H idden units, 1 6 2 ,1 6 3 ,1 6 8 ,1 7 0 ,1 7 2 , Lakatos, I., 154
1 7 9 -1 8 1 ,1 9 3 , 2 0 1 -2 0 8 , 216, 222, Lakoff, G „ 122
225, 233, 235, 236, 244, 247, 251, Lam inar organization, 78, 93
252 Land, E ., 6 5 ,1 0 3 ,1 0 4
H inton, G . E ., xiv, 17 2 ,1 8 6 , 243 Language, 1 0 ,1 6 ,1 7 ,1 8 ,1 9 , 20, 21,
H ooker, C. A ., xii, 49, 1 42,158, 159 26, 42, 46, 53, 54, 64, 69, 7 0 ,1 0 9 ,
H opfield, J. J., 186 1 1 3 ,1 3 2 ,1 5 4 ,1 5 5 ,1 9 5 , 227, 249,
Hornik, K „ 172 251, 268, 281, 284, 294, 295, 301
Hubei, D .H ., 69, 171,174 Lateral geniculate nucleus (LGN), 55,
Huerta, M . F ., 94 69, 70, 73, 82, 266, 267
H um e's problem , 143-145 Laudan, L „ 140, 157, 301
H uygens, C ., 237, 240 Lavoisier, A ., 12
318 Index
Law s, 2, 5, 1 5 ,1 8 , 58, 62, 63, 71, 72, M ooney, R. D ., 94

115 ,1 4 4 , 199, 241, 28 7 -2 9 2 , 301 M ossy fibers, 1 01,182
Learning M uller-Lyer illusion, 257, 261, 262
by back propagation, 172, 173, 174, M unevar, G ., 194
18 5 -187, 2 4 3 -2 4 6 , 248 M usgrave, A ., 142
Boltzm ann learning, 24 4 -2 4 5 , 248
H ebbian learning, 24 6 -2 4 8 Nagel, E., 47, 223
Bayesian learning, 154-155 . N agel, T ., 47, 5 5 -6 6 ,1 0 2
LeD oux, J. E ., 20 N atural kinds, 25, 26, 39, 28 1 -2 8 3 ,
Lehky, S ., 170 2 8 7 -2 9 5
L eibniz's law, 58, 62, 63 Naturalization of epistem ology, 188
Levy, W ., 237 Nem irow , L ., 62, 69
Lew is, D ., 62, 69, 294 NETtalk, 1 6 9 ,1 7 0 ,1 7 4 ,1 7 5 ,1 7 6 , 205,
Lindenberg, R ., 266 207
Linear versus nonlinear networks, Neural networks
1 0 1 ,1 2 7 ,1 6 1 ,1 7 2 , 237 feedforw ard, 1 2 3 ,1 2 4 ,1 6 2 ,1 6 5 , 181,
Linsker, R ., 247 1 8 2 ,1 8 7 ,1 9 4 ,1 9 7 , 200, 201, 221,
Livingstone, M. S ., 69 251, 298, 302. See also Matrix,
Livingstone, R. B ., 266 neural
Llinas, R ., xiii, xvii, 79, 98, 100, 101, recurrent, 208, 236
1 0 7 ,1 8 3 ,1 8 5 N euron, 123, 1 3 1 ,1 5 9 ,1 6 0 ,1 7 0 ,1 7 4 ,
Lund, O . E „ 266 184, 200, 209, 210, 220, 232
Lycan, W . G ., 109 N ew -look theories of perception,
2 2 8 -2 2 9 ,2 5 6
M argolis, J., 10 N ew ton, I., 238, 240, 241, 264
M aterialism , 1, 6, 8 -1 0 , 15, 47, 56, 61, N orm ative notions, 1 0 ,1 5 ,1 6 ,1 7 ,1 8 ,
63, 66, 67, 7 0 -7 4 111, 1 1 3 ,1 5 0 ,1 5 1 ,1 5 8 ,1 9 6 ,1 9 8 ,
Matrix, neural, 35, 78, 92, 9 8 -1 0 2 , 223, 297 ff
1 2 7 ,1 2 9 ,1 8 2 ,1 8 3 , 264, 268 N um erical attitudes, 4 ,1 1 5
M aunsell, J., 69
M axwell, J. C ., 238, 241 O bservables, 31, 52, 53, 5 6 ,1 1 7 ,1 1 8 ,
M ays, L. E ., 95 1 3 9 -1 4 8 ,1 9 0 , 269, 270
M cllw ain, J. T ., 95 O bservation. See Perception
M cKloskey, M ., 116 O ntology, 1, 5, 6 ,1 8 ,1 9 , 48, 50, 55,
M eaning, 3, 21, 22, 3 9 -4 3 , 255, 2 7 1 - 7 2 ,1 1 6 -1 1 8 ,1 2 5 ,1 2 6 ,1 3 9 -1 4 8 ,
277, 281. See also Sem antics 151, 255, 256, 293, 297
M edial geniculate nucleus (M GN), O stensive learning, 205
267 O ther m inds, problem of, 3, 120
M ellor, D. H „ 281
M ental states Pain, 2 3 -3 5 , 44, 45, 59, 113,
elim ination of, 1, 6, 9, 11, 1 5 ,1 6 , 21, Papert, S ., xiii, 171
4 5 ,1 2 5 -1 2 7 Paradigm, 48, 5 1 ,1 2 2 , 154, 158, 191,
reduction of, 4 7 -5 5 , 7 4 -7 6 ,1 0 2 -1 0 8 1 9 2 ,1 9 3 , 240, 264
type identity of, 2 3 -2 9 Parallel distributed processing
M eredith, M. A ., 97 (PDP), 3 9 ,1 9 7 ,1 9 8 ,1 9 9 , 200, 2 0 4 -
M erzenich, M ., 96 207, 2 1 6 -2 1 8 , 220, 225, 228, 229,
M iles, F. A ., 266 252
Mine/rock netw ork, 1 6 5 ,1 6 7 ,1 6 9 , Parallel fibers, 1 07,185
203 Partitions. See V ector space
M ines, 1 2 3 ,1 6 4 ,1 6 5 , 202 Peirce, C. S ., 219, 228
M insky, M ., xiii, 171
Index 319
Pellionisz, A ., xiii, xvii, 79, 9 8 ,1 0 0 , practical, 214

1 0 1 ,1 0 7 ,1 8 3 ,1 8 5 superordinate, 21 4 -2 1 6
Perception social-interaction, 2 1 6 -2 1 7
and abduction, 219, 228-229 motivational, 21 7 -2 1 8
by netw orks, 1 0 2 -1 0 7 ,1 3 2 -1 3 4 , Purkinje cell, 9 9 ,1 0 0 , 107, 185
1 6 3 -1 7 1 ,1 7 4 -1 8 1 ,1 9 0 -1 9 2 , 2 0 2 - Putnam , H „ 11, 2 6 ,1 2 2 ,1 5 7 , 276,
212, 228-2 2 9 281, 282, 285, 286, 287
etiology of, 257 ff Pylyshyn, Z ., 229, 263
encapsulation of, 257-271
plasticity of, 5 2 -5 7 , 6 4 -6 6 , 7 4 -7 6 , Q ualia, 1, 44, 47, 51, 5 5 -6 7 , 7 7 ,1 0 2 -
1 3 1 -1 3 5 ,1 9 0 -1 9 1 , 231-2 5 0 , 255 ff 1 0 7 ,1 4 8
G ibsonian accounts of, 2 28-229 and functionalism , 2 3 -2 9
new -look accounts of, 22 8 -2 2 9 reduction of, 2 9 -3 1 , 5 6 -5 7 , 6 4 -6 6 ,
theory-ladenness of, 3, 5 2 -5 5 , 6 4 - 1 02-108
66, 7 4 -7 6 ,1 8 8 -1 8 9 ,1 2 0 -1 2 1 , MO- reconception of, 5 2 -5 7 , 7 4 - 7 6 ,1 0 2 -
141, 1 4 6 -1 4 7 ,1 8 8 -1 9 1 ,1 9 8 -1 9 9 , 108
2 2 8 -2 2 9 , 255 ff absent, 2 9 -3 1 , 3 4 -3 9
of caloric fluid, 269-271, 27 4 -2 7 5 , gerrym andered, 2 4 -2 9
2 84-2 8 5 Q uantification, 4, 5, 72
of frequencies and w avelengths, Q uine, W . V ., xii, 234
5 6 -5 7 , 6 4 -6 6 , 268-269
o f m olecular kinetic energy, 60 Rationality, 1 ,1 0 ,1 6 ,1 5 3 ,1 5 5 ,1 9 6 ,
o f the night sky, 5 3 -5 4 , 7 5 -7 6 197
with inverting lenses, 262-265 Reading. See NETtalk
in surgically altered cortex, 267 Reduction, intertheoretic, 4 7 -5 5 ,
Perceptron, 171,172 21 4 -2 1 6
Pfaff, D. W „ 105 of folk psychology, 1, 9 ,1 2 5 -1 2 7 ,
Phenom enological properties, 56, 129-135
283. See also Qualia of qualia, 2 9 -3 1 , 5 6 -5 7 , 6 4 -6 6 , 1 0 2 -
P hlogiston, 14, 125,141, 285 108
Pickering, A ., 248 of propositional attitudes, 1 ,1 0 9 ,
Plasticity, 6 6 ,1 2 9 -1 3 4 ,1 4 4 , 255, 257, 15 3 -1 5 4 , 195
258, 263, 264, 267, 268, 277, 278 Reference, 275, 276, 277, 281, 284,
Popper, K „ 10, 21, 277 285, 287
Practical kinds, 281, 290-2 9 5 Relativity, special theory o f (STR),
Progress, scientific, 1 4 0 -1 4 1 ,1 4 9 - 118, 242
1 5 1 ,1 5 7 , 192-194, 219, 2 31-253, Representation
2 7 8 -2 7 9 and intentionality, 42, 43
Propositional attitudes, 1, 3, 4, 7 ,1 0 , of color qualia, 62, 63, 66, 68, 69, 70,
1 1 ,1 6 -1 8 , 20, 22, 3 1 -3 4 , 4 0 ,1 1 2 - 71, 73
1 1 5 ,1 2 1 ,1 2 6 ,1 2 7 ,1 5 4 ,1 5 5 , 158, in the cortex and cerebellum , 77, 78,
188, 223 79, 88, 9 2 ,1 0 1 ,1 0 2 ,1 0 3 , 106,107,
Prototype activation. See Explana 1 0 8 ,1 0 9
tion, proptotype-activation model by prototypes, 1 2 2 ,1 2 3 ,1 2 7
of of culturally im posed features, 130,
Prototype vector, 169, 203, 206, 2 0 8 - 133, 29 8 -3 0 3
212, 2 1 8 -2 2 1 , 223, 225, 238, 240, propositional vs. vectorial, 154-196,
241 20 6 -2 2 3
generation of, 168-169 evolution in, 2 31-283
property-cluster, 212-213 Retina, 37, 38, 65, 70, 73, 81, 92, 94,
etiological, 213-214 95, 9 6 ,1 0 5 ,1 4 8 , 247, 265, 266
320 Index
Retinex theory. See Land, E. Special theory of relativity (STR),

R etinotopic map, 93 118, 242
Risset, J. C ., 106 Sprague, J. M ., 93
R obinson, D. A ., 94, 95 Stanton, P. K ., 247
R obinson, D. L ., 94, 97 Stars, 53, 75, 249, 292
R obinson, H „ 47, 55, 64, 66, 102 State space, 78, 79, 83, 85, 88, 9 1 -
Rock, I., 198,22 8 , 256 108, 209, 221
Rogers, L. J., 266 State-space position, 85, 90, 92, 100,
Rorty, R ., 1, 55, 111 101, 103, 1 0 4 ,1 0 7
Rosch, E ., 122 State-space sandw ich, 9 0 -9 4 , 9 7 -9 8 ,
Rosenberg, C. R ., 124, 169 100-102
Rosenblatt, F ., xiii, xiv, 171, 172 Stein, В. E ., 97
Rum elhart, D. E ., xiv, xvii, 124, 168, Stich, S. P ., xvii, 45, 1 51,157, 158,
172, 173,186, 197, 202 194, 277
Straw son, P ., 120-121
Sacks, J. G ., 266 Stryker, M ., 94, 95
Salm on, W ., 141, 154, 223, 224, 225 Sturgeon, N ., 217
Sandell, J. H ., 95 Superior colliculus, 82, 92, 93, 94, 95,
Schank, R ., 122 96
Scheffler, I . , 154 Suppe, F ., 157
Schiller, P., 93, 94, 95 Sur, M „ 267
Scientific realism , 139, 149, 151, 281, Synapse, 70, 9 9 ,1 0 0 , 1 27,162, 173,
301 1 7 9 ,1 8 3 , 1 8 5 ,1 8 6 , 201, 204, 205,
Searle, J„ 34, 3 9 -4 4 210, 234, 244, 246, 247
Sejnow ski, T. J., xvii, 124, 1 67,169, Synaptic connection, 70, 73, 94, 9 9 -
1 7 0 ,1 7 2 ,1 7 5 ,1 8 0 ,1 8 6 , 204, 205, 101, 123, 131, 150, 160, 161, 162,
247 1 8 3 ,1 8 4 ,1 8 5 ,1 8 9 , 200, 201, 202,
Sellars, W ., 25, 55, 111,141 210, 232, 236, 250, 253
Sem antics, 2, 3, 4 1 ,1 2 2 ,1 5 1 ,1 9 5 , Synaptic w eight, 70, 1 66,177, 1 8 6 -
256, 271, 278, 293. See also M eaning 190, 195, 201, 202, 204, 210, 232,
Sensations. See Qualia 233, 236, 237, 241, 244, 247, 250,
Sensorim otor coordination, 77, 78, 267
79, 82, 86, 88, 89, 90, 9 1 ,1 0 8
Shape from shading, 205 Tank, D ., 186
Shapere, D ., 281 Taste, 1 05-106
Shoem aker, S ., 30 Taylor, C ., 130
Skolem -Ldw enheim theorem , 276 Teller, P ., 223
Similarity, 26, 27, 102-104, 1 0 7 ,1 2 2 - Tem perature, 4 ,1 5 , 31, 43, 47, 49, 51,
126, 158, 175,180, 1 91-195, 204, 52, 55, 59, 60, 75, 259, 283, 285, 292
205, 216, 220, 221, 222, 225, 234, Theories
247, 299 syntactic view of, 153-157
Sim plicity, 62, 8 2 ,1 3 9 ,1 4 1 ,1 4 2 , 1 4 5 - sem antic view of, 157-158
1 4 7 ,1 5 5 ,1 6 0 ,1 7 9 ,1 8 1 ,1 8 8 ,1 9 3 , K uhnian view of, 1 5 8 ,1 9 1 -1 9 3
195, 288, 302 as w eight-space points, 177-178,
Skepticism , 6 ,1 3 9 , 140, 141, 1 42,143, 1 8 8 -1 8 9 , 232, 267
144, 148,149, 276 as vector-space partitions, 178, 2 3 2 -
Sm art, J. J. C ., xi, 141 234, 240-241
Sm ell, 1 0 3 ,1 0 6 ,1 1 3 ,1 1 8 ,1 2 3 Theory-ladenness. See Perception
Sm ith, D. V ., 105 Therm odynam ics, 17, 46, 48, 111,
Sophocles, 8 116, 242, 270, 284. See also Heat
Sparks, D. L ., 95 Third-person account of mind, 7 4 -7 6
Index 321
Thom pson, R. F., 185

Topographic m ap, 79, 81, 82, 89, 91,
92, 93, 94, 95, 96
Training. See Learning
Transform ation of coordinates, 8 6 -
95
Translation, 42, 43, 276
Truth, 1 3 9 -1 5 1 ,1 5 3 ,1 5 7 , 158, 194,
220, 228, 249, 297, 300, 301
Turing m achine, 3 4 -3 8
Tw in Earth, 276
U nderdeterm ination
of truth, 142,143
of reference, 276
Unity, 7 7 ,1 4 6 ,1 9 7 , 210, 22 3 -2 3 0 ,
241, 242, 302, 303
Utilitarianism , 297
V an Essen, D. C ., 69
V an Fraassen, В. C ., 1 39-151, 157,
179, 226, 227, 228
V ector coding, 39, 9 2 ,1 0 6 , 209. See
also State-space position
Vector space, 76, 169, 1 7 4 -1 7 8 ,1 8 0 ,
181, 1 8 9 ,1 9 1 ,1 9 3 -1 9 5 , 2 03-207,
216, 220, 233, 247
partitions on, 168-169, 174-178,
2 0 3 -2 0 4 , 232 -2 3 4
Vital spirit, 14, 22
W arrant of an explanation, 221, 223

W eisel, T. N ., 171,174
W eiss, P ., 228
W essel, D. L ., 106
W ilkes, K „ 116
W illiam s, R. J., xiv, 172
W ilson, M ., 142
W ittgenstein, L ., 205, 286
W olter, J . R ., 266
Zeki, S ., 69, 103

Zipser, D ., xvii, 167, 246

Paul M Churchland A Neurocomputational Perspective The Nature of Mind and The Structure of Science 1989

Transféré par

Informations du document

Titre original

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

Paul M Churchland A Neurocomputational Perspective The Nature of Mind and The Structure of Science 1989

Transféré par

Droits d'auteur :

Formats disponibles

A

The MIT Press

Library of Congress Cataloguing-in-Publication Data

Churchland, Paul M ., 1942-

The single m ost im portant developm ent in the philosophy of m ind

cause the ap paren t authority of introspection w as by this m eans called

T hough published independently over a period of seven years, the

"S o m e Reductive Strategies in C ognitive N eurobiology" w as w ritten

flow vigorously in the opposite direction. Very shortly it will no

La Jolla, California, 1988

The Nature of Mind

T h is p a p e r first ap p eared in th e Journ al o f P hilosophy 78 (1981), n o . 2.

ferred direction in w hich all things fall, that w eight is an intrinsic

1 W hy Folk Psychology Is a Theory

Seeing our com m onsense conceptual fram ew ork for m ental p h e­

w ith anything that m ere physical p hen om en a m ight conceivably dis­

(1) (x )(f)(m )[((x has a m ass of m) & (x suffers a net force o f /) )

C onsider now the large variety of propositional attitudes: '. . .b e ­

(2) (x)(p)[(x fears that p) (x desires that ~ p )]

(If w e stay w ithin an objectual interpretation of the quantifiers,

different m aterial substrates. It is therefore irreducible to the princi­

2 W hy F olk Psychology M ight (Really) Be False

Since folk psychology is an em pirical theory, it is at least an abstract

can n ot explain or fails even to address, one discovers that there is a

already provide searching descriptions and explanations of hum an

3 A rgum ents against E lim ination

Thus the basic rationale of eliminative m aterialism : FP is a theory,

tinct threads running through contem porary functionalism . The first

w ould preclude our being able to describe the (abstract) organization

4 The C onservative N ature o f Functionalism

A valuable perspective on functionalism can be gained from the fol­

(6) (x)(P)(V )(/x)[((x has a pressure P)

(7) (x)(p)[((x desires w ith all his heart that p)

M oreover, and as w ith norm ative convictions generally, fresh insight

5 Beyond F olk P sychology

“ solids" involved and the exhaustive accounting of all dynam ically

the u se of FP altogether, even in the m arketplace. The dem ise of FP's

m icrow aves radiated from an aerial in the forehead, and to perform

The question-begging nature of this argu m en t does not, I assum e,

Functionalism — con strued broadly as the thesis that the essence of

1 F ou r P roblem s concerning Q ualia

'Q ualia' is a p hilosopher's term of art denoting those intrinsic or

This paper was coauthored w ith Patricia S. Churchland. It first appeared in

A . The problem o f in verted/gerrym an dered qualia

N ow there is no point in trying to deny the possibilities just out­

sic nature com m on to all cases of, e .g ., pain, as it occurs in hu m an s,

voluntary w ithdraw al of som e sort. O ur reaction to all of them is im­

stitute the best-founded and m ost authoritative criterion available for

B. The problem o f absent qualia

it fails to be a sensation-of-w arm th on purely functional grounds.

their ow n the intricacies that w ere eventually uncovered by other

C. The problem o f distinguishing states w ith qualia fro m states w ithout

Y es it can. The picture to be avoided here depicts sensations as

D. The differentiation problem

2 The P roblem o f N onstandard Realizations

A. Q ualia in the C hinese nation

Pluto's orbit packed solidly w ith cheek-to-cheek C hinese (roughly

m ore plausible hom e is the vector coding and parallel distributed

B. Intentionality in the Chinese room

the rest of the propositional attitudes. O n the functionalist's view , the

b ecau se the formal symbol m anipulations by them selves d on 't

cases of w hat I will call intrinsic intentionality, w hich are cases of

tions of intentionality, w hich are w ays that people have of speak­

Seeing our com m onsense conceptual fram ew ork for m ental p h e

w ith anything that m ere physical p hen om en a m ight conceivably dis

C onsider now the large variety of propositional attitudes: '. . .b e

different m aterial substrates. It is therefore irreducible to the princi

A valuable perspective on functionalism can be gained from the fol

N ow there is no point in trying to deny the possibilities just out

voluntary w ithdraw al of som e sort. O ur reaction to all of them is im

tions of intentionality, w hich are w ays that people have of speak

D espite the defenses offered above, w e do wish to direct certain criti

across species; it enjoys direct connections w ith evolutionary etholo

W e m ay begin by rem arking that the classical accou n t of in terth eore

F or Thom as N agel, it is the phenom enological features of our experi

it is impossible to exclude the phenom enological features of ex

Im agine a brilliant neuroscientist nam ed M ary w ho has lived her en

tion of a golf sw ing (perhaps in his language cortex, or in the neigh