Vous êtes sur la page 1sur 4

Anita Kulkarni

November 15, 2017

HPS/Pl 120 Paper 1

In his 1955 book Fact, Fiction, and Forecast, Nelson Goodman poses the “new riddle of
induction”, which is concerned with how to precisely distinguish lawlike generalizations from accidental
generalizations. A lawlike generalization can be confirmed with a sample of evidence, whereas an
accidental generalization cannot be. In other words, given that a statement holds true for a single instance
or piece of evidence, if one knows that it then must generalize to all such instances, then the
generalization is lawlike. If one knows that the statement may not necessarily hold for all such instances,
then a generalization of the statement can only be true by accident. The new riddle of induction thus is
concerned with how one can be sure whether a generalization will be true for all cases given knowledge
of only a limited set of cases. An example Goodman gives to clarifyillustrate this distinction is the
following pair of statements:
1. The fact that one piece of copper conducts electricity confirms the hypothesis that all pieces of
copper conduct electricity. The hypothesis is lawlike since a single piece of copper can confirm it.
2. The fact that one man in a given room is a third son does not confirm the hypothesis that all men
in the room are third sons, so the hypothesis is just an accidental generalization.
2. In this case, the new riddle of induction is concerned with the set of criteria that allows one to Formatted: Normal, No bullets or numbering
differentiate the validity of the generalizations mentioned above given knowledge about only a single
piece of copper or a single man in the room. Formatted: Font: 11 pt
Goodman then considers the following scenario, which is at the heart of the new riddle of Formatted: Indent: First line: 0.5"
induction. Repeated observations before 2020 of emeralds being green confirm the hypothesis that all
emeralds are green. But these very same observations also support the hypothesis that all emeralds are
“grue”, where something is grue if it is first observed to be green before, say, 2020, or first observed to be
blue after 2020. (Something is “bleen” if it is first observed to be blue before 2020 or first observed to be
green after 2020.) According to Goodman, despite the extremely unnatural nature of the grue/bleen
definitions, the evidence equally confirms the statements “all emeralds are green” and “all emeralds are
grue”, and both statements are equally lawlike.
In this paper, I will argue that Goodman is correct regarding the equal validity of the green and
grue hypotheses, as long as green/blue and grue/bleen are precisely defined and are capable of describing
exactly the same space of all possible emerald color observations. I will then discuss why the fact that
different definitions yield different predictions, given that the definitions cover the same space of possible
observations, is not fundamentally problematic. Finally, I will then argue that one does not come across
grue-like hypotheses outside of philosophy because one naturally makes hypotheses and classifications
based on the constraints of the human mind. Beyond this, one chooses definitions based on agreement
with past experiences and what one deems simplest given the current state of knowledgeand has seen in
the past.
Typically, whenWhen one states that “all emeralds are green”, one typically means that all
emeralds would always be observed to reflect light of within a specific range of wavelengths independent
of whether it is the first, second, third, etc. observation of the color and independent of the specific time it
the color is observed. In other words, all emeralds are green if they reflect light in a specific range of
wavelengths for all time and for all instances in which it isthey are either observed or not observed.
However, the stated definition of grue only pertains to the characteristics of the first observation of the
emerald and says nothing about subsequent observations. In a grue/bleen reference frame, one could
attempt to construct the definitions of green and blue by saying that green means grue before 2020 and
bleen after 2020, and likewise blue means bleen before 2020 and grue after 2020. However, substituting
in the above definitions of grue and bleen, “green” (which I will subsequently refer to as “green-2”)
means that something is first observed to be instantaneously green before 2020 or first observed to be
instantaneously green after 2020. (The world “instantaneously” is used here to isolate the reflected light
wavelength component of the definition of green given for “all emeralds are green” above.) This
equivalently means that something is “green”green-2 if it is seen to reflect light within a particular range
of wavelengths during a first observation, regardless of the date or whether the reflected wavelength falls
out of the range during subsequent observations. The corresponding definition holds for “blue”, which I
will subsequently refer to as “blue-2”. These constructed definitions are seemingly similar but not
identical to the typical definitions used. So it is not possible to directly compare grue and bleen to the
commonly understood blue and green descriptors. For example, say an emerald I discover on November
315, 2017 reflects 500 nanometer wavelength light (usually described as instantaneously green) upon first
observation that day, but then reflects 650 nanometer wavelength light (usually described as
instantaneously red) upon a second observation the following day. The observations would confirm the
hypotheses “all emeralds are grue” and “all emeralds are ‘green’ (constructed from grue and bleen)green-
2” but would not confirm either “all emeralds are green” or “all emeralds are blue” in the way these are
typically describedused.
The above argument shows that the makeup of the total space of possible observations covered by
any two sets of descriptors must be identical for the sets to be directly comparable. Grue and bleen cover
reflected wavelengths on first observation during different years, whereas blue and green as they are
typically used cover reflected wavelengths over time but not over specific observations. This notion is
similar to the linear algebra concepts of vector spaces and different bases that span these vector spaces.
Grue/bleen and “green”green-2/blue-2“blue” are different bases that span the same vector space, so any
observation within the vector space can be described equally well described using either basis. The same
cannot be said of all observations in the vector space formed by green and blue. Thus, while Goodman’s
example supports all three of the green, “green”green-2, and grue hypotheses, the slight modification
described in the previous paragraph only supports the two hypotheses that are part of the single space that
contains all of theboth observations.
However, without a way to distinguish between lawlike and accidental generalizations, the equal
validity of each of the infinite number of bases to describe a particular space of observations appears to
pose a major problem. Given a limited number of observations, each distinct basis would correspond to a
distinct lawlike generalization, leading to an infinite number of different but equally valid predictions. For
example, a discovery of a ruby that is observed to be instantaneously red on November 15, 2017 would
confirm the hypothesis that all rubies are red. But the observation would equally well confirm the
hypothesis that all rubies are “rurple-n”, which means always instantaneously red before some finite time Formatted: Font: Italic
n after November 15, 2017, of which there is an infinite number of possible n, and always instantaneously Formatted: Font: Italic
purple after n. (“Purpred-n” means always instantaneously purple before n and always instantaneously red Formatted: Font: Italic

after n. Both the red/purple and rurple-n/purpred-n bases span the same space of instantaneous color and Formatted: Font: Italic

time properties.) Using the red/purple basis, one would predict that the ruby will always remain Formatted: Font: Italic

instantaneously red, whereas using the rurple-n/purpred-n basis, one would make the very different Formatted: Font: Italic
Formatted: Font: Italic
prediction that the ruby will spontaneously turn instantaneously purple at time n. The reason that this
Formatted: Font: Italic
discrepancy is not actually problematic is that there is always a potential set of observations that will
Formatted: Font: Italic
determine which of two predictions based on two different bases is correct, thus giving preference to one
Formatted: Font: Italic
of the bases. For instance, a second observation of the ruby from before on May 23, 2024 that shows it to
be instantaneously red would confirm the hypothesis that all rubies are red but contradict the hypothesis
that all rubies are rurple-2023, thus lending preference to the use of the red/purple basis over the use of
the rurple-2023/purpred-2023 basis. Of course, when choosing a preferable basis, one must make the
assumption that the simplicity of the description of the ruby must remain constant (or practically, at a
maximum). Otherwise, it would seem equally preferable to say that the observations of the ruby in the
above example support both the hypothesis that all rubies are red and the hypothesis that all rubies are
rurple-2023 before 2023, purpred-2023 between 2023 and 2029, and again rurple-2023 after 2029.
Therefore, the varying predictions given by different choices of basis do not pose a fundamental
problem in the new riddle of induction. Under the constant simplicity assumption, as the number of
observations increases, it is possible to discard predictions made according to certain bases and thus come
closer to finding an ideal set of descriptors for making lawlike generalizations. As I will explain next, this
process does occur in the real world. Also, due to the constraints of the human mind, science manages to
converge on a handful of preferred bases or even a single privileged basis using this process, even though
in the above ruby example a finite number of observations would still leave an infinite number of
preferable bases remaining.
One does not encounter grue hypotheses in particular outside of philosophy because one usually
observes the color of an object more than once and bases conclusions off of multiple observations. In this
way, it is generally simpler and more useful to consider an observation-independent space of color
measurements. Furthermore, because one does not typically observe the colors of objects change for no
external reason, one naturally expects color descriptions to hold for all times and chooses definitions
accordingly. If colors were regularly observed to change over time, then descriptions such as “rurple-
2020” meaning “red before 2020 and purple after 2020” and “pured-2020” meaning “purple before 2020
and red after 2020” would seem more appropriate than one’s notions of purple and red, which are pured-
2020 before 2020 and rurple-2020 after 2020, and vice-versa. Both bases span the same space, so
hypotheses made from them are equally valid. That said, there are some instances where different sets of
descriptors have been used interchangeably to describe a given span of observations, and where
hypotheses made from these descriptors have been agreed to be equally valid. One such instance is in
psychology, where various classifications of disorders have been devised over the years. As more studies
are undertaken, some classifications are becoming more popular than others due to the simplicity in which
they can differentiate personalities and potential disorders, although logically all are equivalent.
Thus, Goodman is generally correct when he argues that different descriptors, even seemingly
unnatural ones, that span a given space of observations lead to equally valid generalizations, even though
his green and grue emerald example is ill-defined as stated. One encounters some descriptors more often
than others because they are based on past observations and are tailored for simplicity in different
applications.

Vous aimerez peut-être aussi