Vous êtes sur la page 1sur 9

CAUSALITY: PRODUCTION AND PROPAGATION 155

'When
penetration of the nail into the wood. we say that a horse pulls a cart,
IX we mean that the force exe¡ted by'the horse produces the motion of the carl.
When we say that lightning stats a forest fi¡e we mean that the electrical
CAUSALITY: PRODUCTION dis€harge pfodùces ignition. lvhen we say that a peßori's embarrassment was
due to a thoughtlëss remark we mçan that an inappropriate comment p¡o-
AND PROPAGATIONX duced psychological discomfort. Such examples of causal ploduction occur
frequeritly in everyday contexts,
WESLEY C. SALMON Causal propagation (or transmission) is equally familiar. Experiences
which we had earlier in our lives affect ou¡ cunent b€haviour. By means of
memory, the influence of these past events is transmitted to the present. A
sonic boom makqs us aware of the passage of a jet airplane overhead; a
A standard picture of causality has been around at least since the time of
disturbance in the air is propagated from the upper atmosphere to out location
Hume. The gene¡al idea is that we have two (or more) distinct ev€nts which
ori the g¡ound. Signals transmitted from a broadcasting station are ¡eceived
bear some sort of cause-€ffect relation to one aDother. There has, of course,
by the radio in our home. News or music teaches us because electromagnetic
been considerable controvefsy regarding the natu¡e of both the relation and
waves a¡e p¡opagated from the rarismitter to the receiver. ln 1775 some
the relata, It has sometimes been maintained, for instance, that facts or
Massachusetts farmers 'fi¡ed the shot heard "lound the world"'. As all of
p¡opositions (¡athe¡ than events) are the sorts of entities which can constitute
these examples show, what happens at one place and time can have signifì-
the relata. It has long been disputed whether individual events or only classes
cant influenca upon what happeos al other placas and times. This is possible
ofevents can sustain cause--effect felations. The relation itselfhas sometimes
because causal influence can be propagated through time and space. Al-
been taken to be that of sr¡fficient condition, somerimes necessary condition,
though çausal p¡oduction and caus4l propagation are intimately related to one
or perhaps a combinatio[ of the two,l Some authors have even proposed that
another, we should, I believe, resist any temPtation to try lo reduce one to
ce¡tain sorts of statistical relations constitute causal relations.2
the other.
It is my conviction that this standard view, in all of its well-known vari-
âtions, is profoundly mistaken, and that a radically diffê¡ent notion should
be developed, I shall not attempt to mount arguments against the standard
conceptio¡;3 instead, I shall present a rather different apploach for purposes
. 2. PRocEss Es
of comparison. I hope that the altemative will stand on its own merits.
One of the fundamental ðhanges which I Propose in approaching causality is
to take pfocasses rather than events as basic entities. I shall not attemPt any
rigorous definition of processes; Éthe¡, I shall cite examples and make some
1, Two BAstc coNcEpTs
very informal remarks. The main diffe¡enca between evenß and processes is
that events arc ¡elatively tocalized in space and time, while p¡ocesses have
There are, I believe, two fundamental causal conceprs which need to be
müch greateÎ tempolâl dumlion, and in many cases, much greater spatial
oxplicated, and if that can be achieved, iv€ will be in a position to deal with
extent. In space-time diagrams, events ate ¡€presented by points, while
the problems of causality in genemì. The two basic concepts arc production
processes ¿ue repiesented by lines. A baseball colliding with a window would
and propagdtìon, and both are familiar to common sense, Wheri we say that
count as an evçnt; the baseball, iravelling from the bat to the window,
the blow of a hammer drives a nail, we mean that the impact produces
would constitute a process. The activatiol of a photocell by a pulse of light
. would be an event; the pulse of light, travelling, perhaps from a distant
This ñaterial is bås€d upon work supponed by the National Science Foundatìon ùnde. Cranr
No. SES-7809146. sta¡, would be a p¡ocess. A sneeze is an event. The shadow ofa cloud moving
' Se€ Johí L, Mackic, rrr¿ C¿rñ¿n¡ oî the Uniwße (O\fotd, 1911). across the landscape is a p¡ocess. Although I shall deny that all processes
'? Scê W€sLy C. Ssl¡non,'Prcbåbilistic Causâlity', P¿c¡fr Philosophicat Quofle.t,6l (1980), qualify as causal processes, what I mean by a process is similar to what
50-74.
I Some ¡c Siven in'Pmbtbìlistic C¡usâlity'. Bert¡and Russell cbaracterized as ø causal lize: 'A causal line may always
156 WES LEY C. SALMON CAUSALITY: PRODUCTION AND PROPACATION 157

be regarded as the peÌsistence of som€thing-a person, a table, a Photon, or change the spot at some point; for example, wc çan place å red filter ât the
what not. Throughout a given causal line, there may be constancy ofquality, wall wíth the result that the spot of light becomes red at that point. But if we
constancy of structure. or a gradual change of eilher, bul not sudden changes make such a modification in the travelling spot, it will not be transmitted
of any considerable magnitude'.¿ Among the physicauy iinportant processes beyond the point of inte¡aclion. As soon as the light spot moves beyond the
are wav€s and material objects which pe¡sist through time. As I shall use point at which tho red filter was placad, it will become white again. The mark
terns, eygn a material object at rest will qualify as a process. . . can be made, but it will not be t¡ansmitted. We have a 'p¡ocass' which, in
We need to make a distißction between what I shall call cøusal processes the absence of any intervention, consists of a white spot moving regularly
aÍd pseudo-processes: . . . causal processes ar€ those which are capable of along the wall of the b.uilding, If we intervene at somc point, the 'p¡ocess'
transmitting signals; pseudo-processes aÍe incapable of doing so. will be mod¡{ied øt th.at point, but it will continue on beyond that point
Consider a simple example. Suppose that we have a very large circular just as if no intervention had occured. We can, of course, r-nakg the spot
building-a so¡t of super-Astrodome, if you will-with a spotlight mounted red at other places if we .{ish. We can install a red lens in the spotlight,
at its centre. When the light is tumed on in the othenvise darkened buildìng, but that does not constitute a ,oc¿l intervention at an isolated point in the
it casts a spot of light upon the wau. If we tum the light on for a brief proceös itself. We can put ¡ed filters at many places along the wall, but that
mornent, and then off again, a light puls€ llavels from the light ro the wall. would involve zany intewentions rather than a single one. We could get
This pulse of light, travelling ùom the spotlight to the 'rall, is a paradigm of someone to ¡un aroùnd the wall holding a red filter in ftont of the 6pot
what we mean by a causal process, Suppose, further, that the spotlight is continuously, but that would not constitute an interventioî at a s¡ngle poinr
mounted on a mechanism which makes it rotate. If the light is tumed on and in the 'process', . . .
set into rotatio¡r, the $pot of light which it casts upon the wall will move A given process, whether it be causal o¡ pseudo, has a certain degree of
ûound the outer wall in a highly regular fashion. This 'process'-the moving unifomity-we may say, somewhat loosely, that i¡ exhibits a certain struc-
spot of light-seems to fulfrl the conditior¡s Russell used to charactedze ture. The difference between a causal process and a pseudo-process, I am
causal lines, but it is not a causal process. It is a paradigm of what we mean suggesting, is that the causal p¡ocess transmits its own structr¡¡e, while the
by a pseudo-process. pseudo-process does flot, The distinction betweel proc¿sses which do and
The basic method for distinguishing causal processes from pseudo- those ìvhich do not ftansmit thèir own structures is revealed by the mark
processes is the criterion of ma¡k t¡a¡smission. A causal ptoccss is capable criterion. If a process-a causal prccess-is transmitting its own structure,
of transmitting a mark; a pseudo-procass is not. Consider, ñrst, a pulse of tben it will be capable of transmitting modifications in that structure. Radio
light which travels from the spotlight to the wall. lf we place a piece of broadcasting present,s a clear example. The transmitting station sends a ca.Ì-
red glass in its path at any point b€tween ahe spotlight and the wall, the rier wave which has a c€dain structure---characterized by amplitude and
light pulse, which was white, becomes and remairis ted until it reaches frequency, among othe¡ things-and modifications of this wave, in the fofm
the wall. A single intervention at oÍe point in the process t¡srisfoms it
of modulations of amplitude (AM) or frequency (FM), arc imposed for the
in a way which persists from that point on. If we had not intervened, the purpose of broadcasting. P¡ocesses which transmit their own struçtue are
light pulse would have remâined white du ng its ettire joumey f¡om capable of tr¿nômitting marks, sigfials, info.mation, energy, and causal in-
the spotlight to the wall, If we do iDtervene locally at a single place we
fluence. Such p¡ocesses are the means by which causal influence is propa.
can produce a change which is t¡ansmitted from the point of intervention
gated in our ì{orld. Causal influenc€s, trdnsmiited by radio, may set your foot
onwa¡d. We shall say, thercfore, that the light pulse constilutes a causal
to tapping, or induc€ someone to purchase a different brand of soap, or point
process, whether il is modified ot not, since in either case it is capable of
a television camera aboard a spacæcraft toward the rings of Satum. A cáusal
transmitting a mark, Clearly, light pulses can serve as signals and can ttans-
influencc t¡ansmitted by a flying aÍow c¿¡n pierce an apple on rhe head of
mit messaBes.
Witliam Tell's son. A causal influence tra¡smitted by sound waves can mak€
Now, let us considet lhe spot of light which moves arou¡d the wall as the
youI dog come runfling. A causal influence transmitted by ink marks on a
spotlight rotates. There âre a number of ways in which w€ can intervene to
piece of paper can gladden one's day or break someone's heart. Pseudo-
ad processes can do no such things.
' HuM¡ Knob'l¿dqe: Iß Scop¿ Li,nits (New Yotk, 1948),459-
I58 WES LEY C. SALMON C.AUSALITY: PRODUCTION ÀND PROPÀGATION 159

It is evident, I think, that the propagation or transmission of causal in- P(,4rC)>P(Arð) * (¡)
fluence from one place and time to another frrust play a fundamental role in
the causaf structure of úe world. As I shall argue below, causal processes P\B:C)> P@E) ø)
constitute precisely the causal connections which Hume sought, but was For reasons which will be made cle¡T below, we shall stipulate that none of
unable to find.5 the probabilities occuning in these ¡elations is equal to zero or orie. Although
it is not immediately' obvious, conditions (l)-(4) entail
P(A.B')> P(A\xP(Bl' tsl
3. coNJ u NcrIvE FoRKs
ihese relations apply quitc straightforwardly in concrete situations Given
In order to approach the second basic causal concept, ptoductíon, it v¡ill be two effects.4 and 8, which occur together more frequently than they would
flecessa¡y to consider the nature of causal forks. There are thÌee types with if they were sratistically independent ofone another, there is some prior evenl
which we must deal-namely, conjunctive, interactive, and perfect forks. All C which is a cause of ,4 and is also a cause of¡, and which explains the lack
three types are concemed with situations in which a common cause gives rise of independence between Á and 8. In the case of plagiarism, the cause C is
to two o¡ more effects which are someho\Ì, correlated with one another. The the présenc€ of the term paper in the file to which both students had access.
point of departure for this disçussion is Reichenbach's principle oÍ the com. In the c¿se of simultar¡eous illness, the cause C is the common meal which
mon cause, and his statistical çharacterization of the conjuctive fork as a included the poisonous mushrooms. . . .
device !o elaborale thât fundamental causal principle.o To say of two events X and f that they occùned independently of one
The principle of the common cause stares, roughly, that when improbable another means that thcy occur together with a probability equal to lhe product
coincidences recur too frequently to attribute them to chance, they can be of the probabilities of the¡r separate occurrences; i.e.,
explained by reference ao a common causal antecedent. Consider some fam- P (x.Y, P (x) x'P
= (Y) (6)
iliar examples, If two students in a class tum in ide[tical te¡m papers, and if Thus, in the examples \¡/e have considered, as relation (5) states, the two
we can rule out the possibility tha[ eirher copied dircctly from the orher, then
effects A and B are not independent. However, given the occu¡rence of the
we search for a common caùse-for example, â paper in a f¡atemity file from common cause C,.4 and B do occur independently, âs the relationship among
which borh of them copied independently of each othe¡. If two friends, who the conditional probabilities in equation (l) shows. Thus, in the case of
have spent a pleasant day in the country togethe¡, both suffer acute gastro- illness, the fact that the p¡obability of both individuals being ill ar the $ame
intestinal distress in the evenjng, we may find rhat thei¡ illûesses can be time is greate¡ than the product of the probabilities of their individual ill-
traced to poisonous mush¡ooms they collected and consumed. Many such
nesses is explained by the common meal, In this example, we are assuming
examples have been mentioned in the literature, and othels come ¡eadily to that the fact that one pelson is afflicted does not have any direct c¡usal
mind. , . .
influence upon the illness of the other. Moleove¡. I9t us assume fo¡ the sak€
In an attempt to characterize the sttucturc of such examples of common of simplicity úat, in this situation, there are no othq potenlial common
causes, Reichenbach inrroduced the notion of a conjunctive fork, defined in
causes-of severe gastro-intestinal illness.e Then, in the absence of tlle com-
mon cause C-tha! is, when C obtains-, and I are also independent of one
terms of the following four conditions:'
P(A.Btc) = P P(BJC)
(AtC)x (t) another, as the lelationship among the conditional probabilities in equatioir
(2) states. Relations (3) and (4) simply assert that C is a positive cause of A
P(A.B\C) = P(AtC)x P@ta) Q) and 8, since the p¡obability of each is greater in the presence of C than iri
5 the absence of C,
ln Wesley C. SåtBon, 'An "At-At" Thcory of Caùsål Íîlhte¡.e', Philôsophy ol Scíenc., 44
(191'll, 215-24,1h^ve Ãîempted to provide å deråiled a¡alysis of rhe norio¡ of rr¿nsñissior or There is another useful way to look ate4uations (l) and (2). ReÆslling that'
propaeation of causal influenø by cåùsal proc€sses, ånd â justificârion lor the claim thar they
legirimâtely qualiry connecrions. accordiDg to the multiplication theorem,
' "s.åusrl
Hans R€ichenbåch, 1r¿ D ircction o¡ îìne (Berkêl.y ånd Los Angelês, I 956), s€rr. I 9.
I lbid,, sæt, 19, The variableç á, a, C which âpp€¡r in lhe probâbÍny cxpBsior¡s aÌe þkcn by ¡ Reichenbach, ib¡d, 160-1,
Reichenbach ro denoteclas$s, ard rhe probabiliries th€hslvd åre undeßrood as sråtisrical frcquen. "lfotherpotentialcomnoncaus€scxistwccanfomapr¡titionCLCr.Ci.ândthecor.espond-
ing lclâtions eill obtai¡.
160 WESLEY C. SA LMON CAUSALITY: PRODUCTION AND PROPAGATION 16l
P(4.8{Ú) = P(ÂlC) x P(B|Á.C) (7t It m¿y initially seem counterintuitive to say that the results on the two dice
we see that, provided P(,4|C) * 0, equation (l) entails a¡e statistically independent if the electlomagnet is off, and they a¡e statisti-

P(BIC) = P(BIA.C). (8)


cally independent if it is on, büt that ov€rall they are not independent.
Neve¡theless, they are, indeed, non-independent, and this non-independence
In Reichenbach's terminology, this says tbat C screens off A from 8. A arises ftom a clustering of sixes which is due simply to the fact that in a
similar argument shows that C screens off B from A. To screen off meqnà to subset of the class of all tosses ihe probability of 6 is enhanced for each die.
make statistically irrelevant. Thus, accord¡ng to equation (l), the common The dependency a¡ises, not because of any physical interaction between the
cause C makes each of the two effects A and È statistically in€levant to dice, but because of special background conditions which obtain on certain
one another. By applying the same argument to equation (2), we can eas¡ly of the tosses, The same consideration applies to the ea¡lier, less contrived,
6ee that it entails that the absence of the common cause also screens off A cases, When the two stu-dents each copy from a paper in a fraternity file, there
from 8. is no di¡ect physical interaction between the process by which one of the
To make quite clear the nature of the conjunctive fork, I should like to use papers is produced and that by which rhe othcr is produced-ín fact, if either
an example deliberately contrived to exhibir the relationships irivolved. Sup- student had been aware that the other was using that source, the unhappy
pose we have a pair of dice which are rolled together. If the fiÌst die comes coincidenc! might have been avoided, Likewise, as explicitly mentioned in
to rest with side 6 on top, that is a¡ event of rhe type ,Á; if the second die the mushroom poisoning example, the illness of one friend had no effect
comes to rest with side 6 upp€rmost, that is an event of type B. these dice upon thc illness of rhe othe!. The coincidence resulted from the fact that a
arc like standard dice except for the fact that each on€ has a tiny magnet common set of background conditions obtained, namely¡ a commo¡r food
embedded in it. In addition, the table on which they are throvn has a power- supply from which both ate. . . .
ful electÌomagnet embedded in its surface, This magnet can be tumed on or Reichenbach claimed-conectly, I believe-that conjuricrive forks possess
off with a concealed swítch. ff the dice are rolled when the electromagnet is an important asymmetry, Just as we can have two effects which arise out of
on, it is considered an inshnôe of the common cause C; if the magnet is off a given common cause, so also mày we ñnd a common effect resulting from
when the dice are tossed, the event is designated as ô. Iæt us further assume two distinct causes, For example, by getting results on lwo dice which add
that, when the electromagnet is tumed off, these dice behave exactly as up to seven, ofie may win a prize. Reichenbach distinguished three situatíonsi
standard dice. The probability of getting 6 with eirhe¡ die ís l/6, and the
p¡obabitity of getting double 6 is 1/36.10 If rhe electromagnet is tumed on, E
let us assume, the chance of gctting 6 with either die is l/2, and the prob-
,^r
ability of double 6 is I /4. It is eåsily seen that conditions ( 1)-(4) are fulfilted.
Let us make a further stipulatiori, .,vhich will simplify the arithmetic, but
Áq Pa Ad\¡A
which has no other bearing upon the essential features of the example-
namely, that half of the tosses of this pair of dice are made with the elecrro-
\"/
c c
magn€t tumed on, and half are made with it turned off, \ e might imagine
some sort of random device which conbols the switch, and which re¿lizes (i) (ii) (¡iÐ
this equi-probability condition. We can ¡eadily see that the overall prob-
Êbility of6 on each die, regardless of whether the elechomagnet is on or off, Fra. E
is l/3. In addition, the overall probability of double 6 is the arithmetical
average of l/4 and 1/36, which equals 5/36. If rhe occl¡¡rence of 6 on one (i) a common cause C giving rise to two separate effects,,A and 8, without
die were independent of 6 occuFing on the other, the overall probability of any common effect arising frcm A and B çonjointly; (ii) two events A and B
double 6 would be ll3 x l13 = I 19 + 5136. Thus, the example satisfies rela- which, in the absence ofa common cause C, jointly produce a common effect
tion (5), as of couIse it must, in addirion to rclations (t)-(4). E; and (iii) a combination of (i) and (ii) in which the events Á and haveI
both a common cause C arld a common effeÆt E. Hç chatacterized situations
ro We are
assuming lh¿t tbe n¿Anê! in onê die dols not afTccr rhê blhaviour of rhc olh6.lic. (i) and (ii) as 'open forks', while (iii) is closed on both ends. Reichenbach's
WES LEY C,,SALMON CAUSALITY: PRODUCTION AND PROPAGATION 163

asymmetry thesis was that situations of type (ii) never represenf conjunctive the mechanisms by which causal influence is transmitted from the cause to
forks; conjunctive fo¡ks which are opeû are always open to the future and each of the effects. These causal conneçtions are an essential part of the
never to the past, Since the statistical relations which are found in cofljunctive causal fork, and without them, ¡lle event C at the vertex of a conjunctive fork
forks are said to explain otherwise impfobable coincidences, it follows that cannot qualify as a common cause.
such coincidences are explained only in terms of common causes, never
common effects, I believe that an even stronger claim is war¡anted-though
I shall not try to argue it here-namely, that conjunctive forks, whether open 4. INTER AcrI vE FoRKs
or closed by a fourth event, alwayE point in the same ternporal direction.
Reichenbach allowed that in situations of type (iii), the two evenrs A and B Therc is another, basically different, type of common cause which violates
along with their common effect E could form a conjunctive fork. Here, of the statistical conditions used to define the co[junctive fork. Consider a
course, there must also be a common cause C, ând it is C rather than E which simple exampl€, Two pool balls, the cue ball and the 8-ball, lie upon a potil
explains the coincidental occurrence of ,4 and 8. I doubt that, even in these table. A relarive novice attempús a shot which is i¡tended to put the 8-ball
circumstances, A, B, and E can form a conjunctive fork.¡l into one of the far come¡ pockets, but given the posilions of the balls, if the
It would be a mistake to suppose that the staristical rèlations given in 8-ball falls irito one comer pocket, the cue ball is almost ce ain to go into
conditions (l)-(4) afe sufficient to cha¡acteÍize common causes in their role the other far corrier pocket, ¡esulting in a 'scratch'. Let A stand for the 8-ball
as explanations of co¡¡elated effects, as an example, due to Ellis Cr¿snow, dropping inlo the one corner pocket, let B stand for the cue ball droPPíng
clearly demonstrates.l2 Conside¡ a man who usu;lly añives at his office at into the other comer pocket, and let C sta[d for the collision between the cue
about 9.00 a.m., makes a cup of coffee, and set¿les down !o ¡ead the morning ball and the 8-ball which occurs \ryhen the player executes the shot. We may
paper. On some occasions, however, he a.rrives promptly at 8.00 a.m., and ¡easonåbly assume that fhe probability of the 8-ball going into the pocket ís
ori tbesg very same momings his secretary has arrived somewhat earlie¡ and 1/2 if the player tries the shot, and that the probability of the cue ball going
prepared a f¡esh pot of coffee. Moreover, ori just these momings, he is met i¡to the pocket is also about 1/2, It is immediately cvident that,4, B, and C
at his office by one of his associatès who normally works at a differerit do not constitute a conjunctive foÍk, fo¡ C does ¡lot screen,A and B from one
location. Now, jf we consider the fact that the coffee is al¡eady made when another. Given that the shot is attempted the probability that the cue ball wíll
he arrives (,4) and the fact that his associate shows up on that moming (8) fall into the pocket (approx. ll2) is 4ot equal to the probability that the cue
as the coincidenôe to be explained, th€n it might be noted that on such batt will go into the pocket given that the õhot has be€n attempted and that
momings he always catch€s the 7.00 a,m. bus (C), while on other morniògs the 8-ball has dropped into the other f comer pocket (¿pprox. 1).
he usually takes the 8,00 a.m. bus (õ). In this example, it is plausible enough In discussing the conjunctive fork, I took some pains to point out that forks
to suppose that 4,8, and C form a conjunctive fork satisfying (lF(4), bur of that sort occur in situatiofls in which separate and d istinct processes, which
obviously C cannot be considered a cause either of A or of B. The actual do not directly interact, adse out of special background conditions. In the
common cause is an entirely different €vent C, namely, a telephone appoint- example of the pool balls, however, there is a di¡ect inreraction-a colli-
ment made the day befote by his secreta¡y. C'is, in fact, the common cause sion-between the two causal processes which consist in portions of the
of ,4, B, and C. histories of the two balls. For this ¡eason, I have suggested that forks which
In o¡de¡ to distinguish the cases in which the event C in a conjunctive fork a¡e exemplified by such cases b€ called ial eractive forks.tt S]dr¡ce the common
constitutes a bonâ fide common cause from those in which it dogs not, let us causg C does riot statistically screen the two effects A and B from one another,
add the condition that there must be a su¡table causal process connecting C interactive fo¡ks violate condition (l) in the definition of conjunctive forks
with A and anothel connecting C with 8. These causal processes constitute The best way to look at interactive fo¡ks, I believe, is in te[ms of spatio-
tçmpolal intersections of processes. In some cases, two procçsses may inler-
¡r The rdd€r
is urgcd to c¡mpare t¡e illumiflating Â@ount ofc¿usal âsymmelry offercd by Pâlrl sect withoùt prcducing any lasting modifiçalion in either. This \\,i11 happen,
W. Hu'nphEys in 'Probãbilisric Crusâlily aÍd Muhiple Causario¡)', in PcÉr D- Asquith ard Romld
N. Ci€rc (cds.), PSÁ /980 (&sr knsins, Mich., l98l),2T37.
': I håd pr€viously Áitriblr.d this cmneor¡s view to R€ichenbach. bur Paul Humphrcys kindly r¡ Se¿ ìVcsley C. Saltnon, 'why Ask '1r'hy?"?-An lnquiry Conc€ming scic¡tific ErPlenâtion
poi¡r.d our rh¡! my ârtliburio ras iDco.r€cr, Proceedi¡ss ¿'d Á¿dreises o¡tne Aneìca^ Phitosoph¡cdt Asso.¡dtion,5116( ú8 1978),68'-?05 "
164 WESLEY C. SALMON CAUSÀLITY: PRODUCTION AND PR OPA G ATION 165

for example, \¡/hen both processes are pseudo-processes. If the paths of two P(A.Btc)> P(AtC) x P(Blc) (e)
airplanes, flying in different directions at different altiiudes on a clea¡ day,
Moreover, given a causal interaction of the foregoing son, I shall say that the
cross one another, the shadows on the ground may coincide momentarily,
change in each process is produced by the interaction with lhe other p¡ocess.
But as soon as the shadows have passed the intersection, both move on as if
no such int€¡sectíon had ever occuned. In the case of the two pool balls, I have now characte¡ized, at least partially, the two fundamental causal
concepts mentioned at the outset. Causal processes are the means by which
however, the intersection of their paths results in a change in the motion of
each which would not have occuÍed if they had not collided. Energy and causal influence is propagated, and changes in processes arc produced by
momentum are transfgrred from one to the other; their respective states of causal interactions. We are now in a position to see the close ¡elationship
motion are altered. Such modifications occur, I shall maintain, only when between these basic notions. The distinction between causal plocesses aod
two çausal processes intercect. If either or both of the intersecting processes pseudo-processes was formulated in terms of the criterion of mark trans-
are pseudo-processes, ¡ro such mutual modifiçation occurs. Howeve¡, it is missiori. A mark is a modification in a process, and if that modiflcadon
entirely possible for two causal processes to intersect without any subsequent persists, the mark is transmitted. Modifications in processes occur when they
modification in either. Ba¡ring the extremely improbable occuÍence of a intersect with other processes; if the modifications persist beyond th€ point
particle-particle type collision between two photons, light lays normally pass of intersection, then the intercection constitutes a causal interaction and thc
right through one another without any lasdng effect upon either one of them. interac!íon has produced maÍks which ate transmitted For example, a pulse
The fact that two inte¡secting processes a¡e both causal is a necessary but not of white light is a process, and a piece of red glass is another process lf these
sufficier¡t condition of the producrion of lasting changes in rhem. two processes intersect-i,e, if the light pulse goes through the red glass-
When two caùsal pfocesses intg¡sect and suffer lasting modifications after then the light pulse becomes and remains red, *hile the fílter undetgoes an
tho intersection, there is some co¡relation between the changes which occur increase in energy as a result of absorbing some of the light which impinges
in them. In many cases-and perhaps all-energy and/or momentum transfer upon it. Although the newly acquired energy may soon be dissipated into the
occurs. and the coFelations between the modificat¡ons are direct conse4uen- $urrounding environment, the glass retains some of the addod energy for
ces of the respective consewation laws.la This is nicely iltustrated by the some time beyond the actual moment of ioteraction.
Compton scattering of an energetic photon from an electron which can be we may, therefore, turn the presentation around in the following way. We
considered, for practical puryoses, initially at resr. The differenc€ in energy live in a world whiçh is futl of p¡oc€sses (causal or pseudo), and these
between the íncoming photon l,v and the scattered photon r¡v'is equal to the processes unde¡go frequent intersections with one another' Some of these
ki$etic energy of the recoiling electron. Similarly, the momentum change intersgctions constitute causal interaclions; otheß do not lf an intersection
in the photon is exactly compensated by the momentum change in the occurs which does not qualify as an interaction, we can draw no conclusion
elect¡on.l5 as to whether the processes involved are causal or pseudo. If two processes
When two processes inte$ect, and they unde¡go correlated modificatioís intersect in a manner which does qualify as a causal intefaction, then we may
\ryhich p€rsist after the intersection, I shall say that the intersection constitules conclude that both proc€sses are causal, for each has been marked (i e.
a causal interdctíon This is the basic idea behind what I want to take as a modified) in ihe intersection with the other, and each procass t¡ansmits the
fundamental causal concept. Let C stand for the event consisling of the mark beyond the point of intersection. Thus, each p¡ocess shows itself cap-
intgrsection of two processes, [.et,4 stand for a modification in one and I able of fansmitting marks, since each one has transmitted a mark generated
fo¡ a modifica¡ion in the other. Then, in many cases, we find a relation in the intersection, Indeed, the ope¡ation of marking a process is accom-
analogous to equation (l) in the definition of the coojunat¡ve fork, except plished by means of a causal interaction \tith another process. Although we
that the eqùalíty is replaced by an inequality: may often take an active rolg in producing a mark in o¡der lo ascenait
whether a process is caùsal (or for some other purpose), it should be obvious
¡¡ For å v¿luable discussion of rhe role of ènergy and momerrum r.¿nsfer ¡n cîusâti¡y, s€e David that buman agency plays no essential part in the charactelization of causal
Fai¡,'Causarior ¡nd rhe Flow of EnerSy , E¡l¿nrrr6, 14 (1979), 219-50.
'' As €rplâ¡ned in Srlmoî, Why Àsk 'Why?"?', thc example of Compron scá erint has rhe processes or causal interactions. ì#e have every reason to bqlieve that the
advrntage of b€ìn8 iñedu.ibly ltatistic6l, orìd thùs, nor Âralysablè, cv.n in pr¡nciplê. ås â pc¡fær world abounded in causal processes and causal interactions lorlg before therc
foÎk (discussed b€low)-
were ¿rny huma¡ agents to perfom expe¡iments.
WBSLEY C. S A LMON cAUSALfTY: PRoDUCTIoN AND PRoPAGATION
Our ordinary causal lângu¿ge is infused with temporal asymmetry, but we
5. RELATIoNS BETWEEN coNJUNcrIvE AND INTERAC. vE FoRKS should be careful in applying it to å¿r¡c causal concepts lf, for example, we
say that two processes are modified as a result of their inte¡action, the wo¡ds
Suppose that we haye a shooting-gallery with a number oftargets. Tbe famous suggest that we have already determined which are the states ofthe processes
sha¡pshoote¡, Annie Oakley, comes to this gallery, but it preserits no challenge prior to the interaction, and which are the subsequent states. To åvoid beg-
to her, for she can invariably hit the bull's-eye of any target ar which she aims, ging temporal questions, we should say that hryo processes intersect, and each
So, to make the situation interesting, a hardened st€el knife-edge is installed of the processes has different characae¡istics on the two sides of the intersec-
in such a position that a direct hil o¡ the knife-edge will seve¡ rhe bullet in a tion. We do not try to say which part of the ptocess comes earlieÍ ând which
way which makes one fragment hir the bull's-eye of rarger A while the other lator. The same is t¡ue when tfle speak of marking. To erase a mark is the
fragment hits the bull's-eye of target B, If we let ,4 stand for a fragmenr exact temponl ¡everse of imposing a mark; to sPeâk of imposing or erasing
striking the bull's-eye of target A, B fo¡ a fragment striking the bull's-eye of is to prcsuppose a temporal dirqction. In many cases, of course, we know on
target B, and C for the severing of the bullet by the knife€dge, we have an other grounds that certâin kinds of interactíons are ineversible Light filters
interactive fork quite anâlogous to the example of the pool balls. Indeed, we absorb some f¡equencies, so that they transform white light into red Filters
may use the same probability values, setting P(AIC)= P(BIC)= l/2, while do ûot fumish missing fteqùencies to tum red light into white. But unril we
P(AlC.B)= P(BlC.A)- 1. Statistical screening-off obviously fails. have gone into the details of the physics of ir¡eversible processes, it is best
¡;y'e might, however, consider another event C*, To make the situation to think of causal interactions in temporally symmetric te¡rns, aÍd to take the
concrete, imagine that ìve have installed between the knife-edge and the causal connections which are furnished by causal p¡ocesses as symm€t¡íc
targets a steel plate with two hol€s in it. If the shot at the knife-edge is good,
connections. Causal processes and causal intemctions do not fumish t€mpo-
then the two fragments of the bullet will go through the two holes, and each fulfil that function.
¡al a,symmetry; conjunctive fo¡ks
f¡agment will strike its respective bull's-eye with p¡obability virtually equal
to L Let C* be the event ofthe two f¡agments go¡ng through thei¡ respective
holes. Then, we may say,,4, B, and C* will folm a conjìrncrive fork, That 6, PERFECT FoR Ks
happens because C* refe¡s to a situation which is subsequent to the physical
interaction between the pads of the bullet. By the time we get to C*, úe
In dealing with conjunctíve and interactive forks, it is advisabl€ to resttict
bullet has bee¡ cut into two separate pieces, and each is going its way
our a[ention to the cases in which P(AlC) and P(8lC) do not assume either
independently of the other, Even if we should decide ¿o vaporize one of the
of the extreme values zero or one. The main reason is that the relation
fragments with a powerful lase¡, that would have no effect upon the prob- Jir

ability of the other fragment finding its ta¡get, This example makes quire P(A.Btc)= P(A\C)X P(Btc) = 1 (10)
,,i¡
'ìlr
vivid, J believe, the disûnctioD between the interactive fork, which charac- may represent a limiting case of €ither a conjunctive or an interactive fo¡k,
oven though (10) ís a special case of equation (l) and it violates relation (9).
terizes dircct physical inteEctions, and the conjuncrive fo¡k, which charac-
terizes indeperdent processes afising under special background conditions. Consider the Annie Oakley example once more. Suppose that she retums I
the¡e is a funher impofiant point of contrast between conjì.rnctive and to the special shooting-gallery time after time Given that Practice måkes
interaçtive fo¡ks, Conjunctive forks possess a kind of tempotal asymmetry perfect (at teast in her case), she improves her skill until she can invariably :i
which was described above. Interactive fo¡ks do not exhibit the same son of hit the knife-edge in the manner which results in the two fragments fifiding !l
temporal asymmetry. This is easily seeo by considering a simple collision their ¡espective bull's-eyes. Up until the moment shat she has perf€cted her ¡
between two billiard balls. A collision of rhis type can occur in reverse; if a technique, the resultq of her trials exemplified interactive forks. It wauld be 4
t;l
collision C precedes states ofmotion Á and I in the two balls, then a collision absurd to claim that, when she achieves perfecùion, the splitfing of the bullet
C can occur i¡ which states of motion just like /4 and 8, except that the no longer constitutes a causal intemction, but must now be rega¡ded as a l)
direction of motion ¡s reversed, precede the collision. Caùsal interaclions and conjunctive fo¡k. The essence of the interactive fork is to achieve a high
causal processes do nol, in and of themselves, provide a basis for temporal coFelation betrveen two ¡esults; if the correlatíon is perfect, we can ask fio 'x
asymmetry. more. It is, one might say, an arithmetical accident that whçn perfection
168 WES LEY C. SALMON CAUSÄLITYT PRODUCTIoN AND PRoPACATION 169

occurs, equarion (l) is fulfilled while the ineqùality (9) must be violated. If causal process) from the gun to Ìhe victim, where the bullet then interacts
probability values wefe no¡malized to some value othe¡ than l, that result with the body of the victim,
would not obtain. It therefore sgems best to treat this special case as a third The foregoing characterization of causal processes and various kinds of
type of lotk-the perfect fork. causal forks provides, I believe, a basis for understanding th¡ee fundamental
Conjunctive forks also yield pedect forks in the limit, Consider the aspects of causality:
example of illness due to consumption of poisonous mushrooms. If we l. Causal processes are the means by which structure and order are pro-
assume-what is by no means always the case-that anyone who consumes pagated gr tra'Jlsmilted f¡om oqe space-time region of the univeße to other
a significant amount of the mush¡oom in question is certain to become times and places,
violently ill, then we have anothef instance of a perfect fork. Even wher these 2. Causal ìnteractíors, as explicated in tefms of inte¡active forks, con-
limiting values obtain, howev€r, there is still no direct interaction between stitute the means by which modìÍicol¡ons ¡n structure (which are propagated
the processes leading ¡espectively to the two cases of severc gastro-intestinal
by causal processes) arc produced.
dist¡ess.
3. Conjunctive com¿?on cøuses-as chatacteÌized in terms of conjunctive
The main point to be mad€ conceming pe¡fect forks is that, when the
probabilities take on the limiting values, ìt is impossible to tell from the
forks-play a vital role i\ the prcduction of structure and o¡der, In the
conjunctive fork, it will be recalled, two o¡ more processes, which are physi
statistical relationships alone whether the fork should be considered ¡ntelac- cally independent ofone aûother and which do not interact directly with each
tive or conjunctive. The faca that relations (l)-(4), which a¡e used in the other, arise out of some special set of background conditions. The fact that
characterization of conjunctive forks, are satisfied does not constitute a suf'
such sp€cial background conditions exist is the source ofa coÍelation among
ficient basis for making a judgement about the t€mporal orientation of the the various effects which would be utterly improbâble in the absence of the
fork. Only if we can establish, on separate grounds, that the perfecr fo¡k is a common causal backgrcund.
limiting case of a conjunctive.(rather than an interactive) fork, can we con-
clude that the event at the ve¡tex is a common cause rathet than a common There is a striking diffe¡ence bdtween conjunctive common causes on the
effect, Pefect forks need to be distinguished from the otheÌ two types mainly one hand and causal processes and interactions on the other, Causal processes
to gùard against this possibl€ source of confusion. and causal interactions seem to be govemed by basic laws of natu¡e in ways
which do not apply to conjunctive forks. Consider two paradigms of cau$al
processes, namely, art elecl¡omagnetic wave propagating through a vacuum
7. THE cAUSAL srRUcruRE oF THE woRLD atrd a material particle moying without a¡y net extemal forces acting upon
it. Bar¡ing any causal interactions in both cases, the electromagnetic wave is
I¡ everyday life, when ve talk about cause--effect relatiôhs, we think typi- govemed by Maxwell's equations a¡ld the material particle is govemed by
cally (though not necessarily invariably) of situations in which one event Newton's fi¡st law of motion (or ils counterpart in relativity theory), Causal
(which we call the cause) is linked to another event (which we call the effect) interactions are typified by va¡ious sorts of collisions. The couelations be-
by means of a causal process, Each of the two events which stands ¡n rhis tween the changes which occur in the processes involvcd ¿rc govemed-in
relation is an interaction between two (or more) intersecting p¡ocesses we most, ifnot all, cases-by fundâmental physical conservation laws. Although
say, for cxample, that the window was broken by boys playing baseball ln I am not prepared to argue the case in detail, it seems plausible to suppose

this situaúon, there is a collision of a bat wi(h a ball (an interactive fork), the that sll Íundamental phls¡cøl interact¡ow can be regarded as exemplifica-
t"'tl
motion of the ball through space (a causal process), and a collision of the tions of the interactive fork.
ball with the window (ao interactive fork). For another example, we say that Conjunctive common causes are not nearly as closely tied to the laws of
tuming a switch makes the light go on. In this case, an interaction belween nature. It should hardly require mention that, to the extent that conjunctive "1

a switching mechanism and an electrical ci¡cuìt leads to a process consisting forks involve causal p¡ocesses and causal inteÉctions, the laws of nature
of a motion ofelectric cha¡ges ¡n some wires, which in tum leads to emission apply as sketched in the preceding paragraph. However, ìn contrast to causal I
of tight fiom a filament. Homicide by shooting provides s¡ill another processes and causal interactions, conjunctive forks depend crucially upon ll
tj
example. An interaction b€tween a gun and a ca¡tridge p¡opels a bullet (a de Íacto background conditions. Recall some of the exampleg mentiofied
,t)
170 Vr'ESLEY C. S ALMON CAUSALITY: PRODUCTION AND PROPACATION l'11

above. In the plagiarism example, it is a nonlawful fact that two members Throughout this discussion of causality, I have laid paÌticular stress upon
of the same class happen to have access to the same hle of term papers. In the role of causal p¡ocesses, and I have even suggested the abandonment of
the mushroom poisoning example, iÎ is a non-lawful fact that the two partici- the so-called 'event ontology'. It might be asked wheth$ it would not be
pants sup together out of a common pot. In the twin quas4 €xample, it is possible to carry through the same analysis, within the framework of an event
a de facto ao.nditioî tha[ the quasar and the elliptic galaxy are situated in ontology, by considering processes as continuous series of events. I see no
such a way that light coming to us from two different directions arises reason for supposing that this programme could not be car¡ied through, but
from a source which radiates quite uniformly from extended portions of irs I would be incliried to ask why we should bothe¡ to do so. One impoÍant
surface. , . . source ofdifficulty for Hume, if I understand him, is that he tried to account
for causal çonnections between notl-contiguous events by interpolating inter-
yening events, This approach seemed only to mise precisely the same ques-
8. coNcLUDlNc REM,q RK s tions about causal connections between events, for one had to ask how the
causal influence is transmitted from one intervening event to another along
The¡e has been con$iderable controversy since Hume's time regarding the the chain. The difficulty is circumvenred, I believe, if we look to p(ocesses
question of whether causes must precede their çffects, or whethe¡ causes and to provide the causal connections.ìó Focusing upon processes mther than
effects might be simultaneous with each other. It seems to me that the events has, in my opinion, enormous heuristic (if flot systematic) value. As
foregoing discussìon provides a reasonable resolutìon of this conÍoversy. lf John Venn said in 1866, 'Substitute for the time bonoured "chain of causa-
we are talkiog about the typical cause-€ffect situation, which I characterized tion", so often inÍoduced into discussions upon this subject, the phrase À
above in tsrms of a causal process joining two distinct intefactions, then we "rope of causation", and see what a very different aspect the question will
are dealing with cases in which the cause must p¡ecede the effect, fo¡ causal wea¡.'l?
propagation over a finite lime interval is an essential feature of cases of this
type. If, however, we are dealing simply with a causal interaction-an inter- 'ó see salmon, 'Àn "AGAI" Theory'. " The Losic oîCh¿"ce lLondot, 1866), 320.
section of two or more plocesses which produces lasting changes in each of
them-then we have simultaneity, since each process inte$ects the other at
the same time. Thus, it is the intersection of the white lighl pulse with the
red filter whìch produces the red light, and the light becomes red at the very
time of its passage through the filter. Basically, propagation involves lapse
of time, çhile int€raction exhibits the relation of simulta¡eity.
Another traditional dispute has centred upori the question of whether stâte-
ments about causal relations pertain to individual events, or whether they ïi
hold properly only with ¡espect to classes of events, Again, I believe, the
foregoing account furnishes a straightforward answer. I havs argued thal :tl:
causal processes, in many instances, constitute the causal connections be-
tween cause and effecl. A causal process is an individual entity, and such
,j
entitigs ¡ransmit causal influence. An i¡dividual Process can sustâin a causal
connection between an individual cause and an individual effect. Statement$
about such relations ne€d not be construed as disguised generalizations. At
the same time, it should be noted, we have used statislical ¡glations to
charâcter¡ze conjunctive and interactive fo¡ks. Thus, strictly sPeaking, when
!.
we invoke something like the principle of the common cause, we are impli- )
¡
citly making assertions which involve statistica.l generalizations Causal re-
lations, it seems to me, have both partícular ld general aspects.
t,rj

i,'

fi

Vous aimerez peut-être aussi