Goodman's New Riddle'

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 17

Nicod & Hempel

Hempel & Goodman

Goodman & Bayesianism

Triviality?

Summary

References

Goodmans New Riddle


Branden Fitelson
Department of Philosophy
Group in Logic and the Methodology of Science
&
Institute for Cognitive and Brain Sciences
University of CaliforniaBerkeley
branden@fitelson.org
http://fitelson.org/

Branden Fitelson

Goodmans New Riddle

fitelson.org

Nicod & Hempel

Hempel & Goodman

Goodman & Bayesianism

Triviality?

Summary

References

The Plan
This is mainly a historical talk (part of an ongoing book
project on confirmation theory) with the following aims:
First, to place Goodmans Grue paradox in historical
perspective (Keynes, Nicod, Hempel, and friends).
Then, to examine the paradox from both Hempelian and
Bayesian perspectives (with an inductive-logic slant).
Upshot: To try to motivate the following claims:
Grue is not necessary for discrediting Hempels theory of
confirmation. [This can be done in an elementary way.]
Grue is not necessary for establishing underdetermination
cases. [These were known long before Goodman.]
Grue is not sufficient to establish Goodmans desired
triviality result. [This fails even for Hempels theory.]

Grand Conclusion of the talk: Either Goodmans New


Riddle is not new, or it is not much of a riddle (or both).
Branden Fitelson

Goodmans New Riddle

fitelson.org

Nicod & Hempel

Hempel & Goodman

Goodman & Bayesianism

Triviality?

Summary

References

Nicod [12], the grandfather of instantial confirmation, said:


Consider the formula or the law: A entails B. How can a particular
proposition, [i.e.] a fact, affect its probability? If this fact consists of the
presence of B in a case of A, it is favourable to the law . . . on the contrary, if
it consists of the absence of B in a case of A, it is unfavourable to this law.

While Nicod is not crystal clear on the logical details of his


account of instantial confirmation, three fundamental
aspects of it can be groked from his inchoate remarks:
(1) Instantial confirmation is a relation between singular and
general propositions/statements (viz., facts and laws).
(2) Confirmation consists in positive probabilistic relevance, and
disconfirmation consists in negative probabilistic relevance
(for Nicod, the probabilities are Keynesian-inductive [10]).
(3) Universal generalizations are confirmed by their positive
instances and disconfirmed by their negative instances.

Hempel takes Nicod as his point of departure. But, Hempels


reconstruction of Nicod is rather uncharitable. Sadly, the
resulting straw man has had profound historical influence.
Branden Fitelson

Goodmans New Riddle

fitelson.org

Nicod & Hempel

Hempel & Goodman

Goodman & Bayesianism

Triviality?

Summary

References

Hempels reconstruction of Nicods instantial principle is


(NC0 ) For all objects x, and for all properties and :
x confirms [(y)(y y)\ iff [x & x\ is true, and
x disconfirms [(y)(y y)\ iff [x & x\ is true.

(NC0 ) diverges from Nicods conception in two key ways:


(NC0 ) makes the relata of the confirmation relation objects
and statements (as opposed to statements and statements).
(NC0 ) says nothing about (inductive) probability.

As Hempel explains, (NC0 ) is quite unattractive (hence


uncharitable?), since it violates the equivalence condition
(EQC) If x confirms y, then x confirms anything equivalent to y.

Hempel formulates a general, logical theory of confirmation.


His theory satisfies (EQC) and some other desiderata he lays
down. But, it exhibits some undesirable vestiges of (NC0 ).
E.g., Hempels theory implies that, for all objects x, for all
(compatible) properties and , and for all statements H:
(NC) [x & x\ confirms [(y)(y y)\ and
[x & x\ disconfirms [(y)(y y)\.
(M) If [x\ confirms H, then [x & x\ confirms H.
Branden Fitelson

Goodmans New Riddle

fitelson.org

Nicod & Hempel

Hempel & Goodman

Goodman & Bayesianism

Triviality?

Summary

References

(NC) is a more plausible thing to attribute to Nicod. But,


Hempels theory is still non-probabilisic. And, (M) is quite
implausible especially from a Pr-relevance point of view.
Interestingly, Hempel himself seems (quite rightly) to have
thought that (M) was (intuitively) false. In his discussion of
the raven paradox, he describes a counterexample to (M)!
Hempel seems to have been unaware of this tension in his
views [1]. My conjecture is that (M) is a carry-over from
Hempels objectual reconstruction of Nicod. Anyhow, (M) is
responsible for various historical confusions in this area.
E.g., the claim that the raven paradox argument shows that
a is a white shoe confirms that all ravens are black.

With that background in place, now were ready to look at


the New Riddle. There are two parts to the Riddle:
Part I: the underdetermination/indistinguishability part.
Part II: the triviality part.

Ill say something about each part (I think both are crucial).
Branden Fitelson

Goodmans New Riddle

fitelson.org

Nicod & Hempel

Hempel & Goodman

Goodman & Bayesianism

Triviality?

Summary

References

Here is (the all too familiar) Part I of Goodmans Riddle.


Now let me introduce another predicate less familiar than green. It is the
predicate grue and it applies to all things examined before t just in case
they are green but to other things just in case they are blue. Then at time t
we have, for each evidence statement asserting that a given emerald is green,
a parallel evidence statement asserting that that emerald is grue. And the
statements that emerald a is grue, that emerald b is grue . . . will each confirm
the general hypothesis that all emeralds are grue. Thus according to our
definition, the prediction that all emeralds subsequently examined will be
green and the prediction that all will be grue are alike confirmed by evidence
statements describing the same observations. . . . Thus although we are well
aware which of the two incompatible predictions is genuinely confirmed,
they are equally well confirmed according to our present definition. . . .

Next, I will reconstruct this first part of the New Riddle,


and evaluate it from Hempelian and Bayesian perspectives.
Then, I will move on to the triviality part of the Riddle.
Branden Fitelson

Goodmans New Riddle

fitelson.org

Nicod & Hempel

Hempel & Goodman

Goodman & Bayesianism

Triviality?

Summary

References

Let Ox x is examined before t, Gx x is green, Ex x


is an emerald, and Gx Ox Gx (i.e., x is grue).
Now, we can state the two salient hypotheses, as follows:
(H1 ) All emeralds are green. Formally, H1 is: (x)(Ex Gx).
(H2 ) All emeralds are grue. Formally, H2 is: (x)(Ex Gx).
Even more precisely, H2 is: (x)[Ex (Ox Gx)].

And, here are three salient (instantial) evidential statements:


(E1 ) Ea & Ga (E2 ) Ea & (Oa Ga) (E) Ea & Oa & Ga
Goodman must be saying that there is some statement
about a that (on Hempels theory) confirms both H1 and H2 .
Neither E1 nor E2 does the trick. But, E does. This follows
from (NC) and (M). As such, I will reconstruct Part I of the
New Riddle as involving the following two central claims:
(6) E confirms both H1 and H2 .
(6? ) E confirms both H1 and H2 equally.

That is, Part I of the Riddle is that H1 and H2 are


confirmation-theoretically indistinguishable by E.
Branden Fitelson

Goodmans New Riddle

fitelson.org

Nicod & Hempel

Hempel & Goodman

Goodman & Bayesianism

Triviality?

Summary

References

For Bayesians, confirmation is a three-place relation,


defined as follows. E confirms H, relative to background
corpus K iff Pr(H | E & K) > Pr(H | K), for some suitable Pr.
From a Bayesian point of view, (NC) is not generally true
[2, 3, 11]. However, a closely related principle is correct:
(NC ) For all objects x and all (compatible) properties and :
[x x\ confirms [(y)(y y)\, and
[x & x\ disconfirms [(y)(y y)\.

(NC ) makes sense because (y)(y y) ` a a,


and a & a ` (y)(y y). On the other hand,
(NC) is less clear because (y)(y y) a & a.
Of course, if one assumes (M), then (NC ) entails (NC)! But,
(M) is rather obviously false from a Bayesian point of view.
Let Bx x is a black card, let Ax x is the ace of spades,
and let Jx x is the jack of clubs. Assuming (K) that we
sample a card a at random from a standard deck, we have:
(7) Ba confirms Aa, relative to K, but
(8) Ba & Ja disconfirms (indeed, refutes!) Aa, relative to K.
Branden Fitelson

Goodmans New Riddle

fitelson.org

Nicod & Hempel

Hempel & Goodman

Goodman & Bayesianism

Triviality?

Summary

References

In Hempels theory, (6) trades on (M). But, (M) is clearly false


from a Bayesian point of view. So, weve seen no good
reason for a Bayesian to accept (6). [In fact, (M) is a reason
to reject Hempels theory no grue needed for that!]
Moreover, heres a Bayesian counterexample to (6) & (6? ) (a
la Good [2, 3]) such that E confirms H1 but disconfirms H2 .
(K) Either: (H1 ) there are 1000 green emeralds 900 of which have
been examined before t, no non-green emeralds, and 1 million
other things in the universe, or (H2 ) there are 100 green
emeralds that have been examined before t, no green emeralds
that have not been examined before t, 900 non-green emeralds
that have not been examined before t, and 1 million other things.

Now, imagine an urn containing true descriptions of each


object in the universe, and let E Ea & Oa & Ga be
drawn. E confirms H1 but E disconfirms H2 , relative to K:
Pr(E | H1 & K) =
Branden Fitelson

900
100
>
= Pr(E | H2 & K)
1001000
1001000
Goodmans New Riddle

fitelson.org

Nicod & Hempel

Hempel & Goodman

Goodman & Bayesianism

Triviality?

Summary

References

Of course, all this shows is that neither (6) nor (6? ) is a


logical consequence of Bayesian confirmation theory. But, (6)
and (6? ) are consistent with Bayesian confirmation theory.
As in the literature on the raven paradox [3, 7], we now turn
to the appropriateness of our probabilistic assumptions
for modeling the confirmation relations of the paradox.
Hempel [6, 7] explains that the raven paradox is about Ks
which do not entail that a has (or lacks) any of the
properties involved in the Hs (i.e., ravenhood/blackness).
Presumably, we should say the same thing about grue.
But, it is often assumed (e.g., [13]) that K entails Oa. This
directly contradicts the Hempelian line on paradoxes.
Following Hempel, I have not assumed the object in
question (a) has or lacks any of the properties E, G, or O.
The key question is whether Pr(E | H2 & K) < Pr(E | H1 & K)
can be motivated as a reasonable (appropriate for grue)
probabilistic confirmation-theoretic modeling assumption.
Branden Fitelson

Goodmans New Riddle

fitelson.org

Nicod & Hempel

Hempel & Goodman

Goodman & Bayesianism

Triviality?

Summary

References

In a recent manuscript, Roger White asks us to think about


what a grue (viz., anti-inductive) world might be like:
To achieve our goal of making standard inductive methods unreliable in a
world, we are going to have to take a more aggressive . . . approach. Rather
than just assigning properties to objects and leaving it to chance how our
inductivist victims will obtain their data, we must somehow give them a
tendency to obtain misleading data. . . . We put only green emeralds in the
vicinity of the surface of the earth, burying the blue ones deep below, or
tucking them away in hard to find places. Then once our victims have
sampled sufficiently many green emeralds that they can inductively predict
that the next one will be green, we cause a blue one to rise to the surface
nearby. . . . Coordinating the . . . show will require either a very smart demon
overseeing it all, or the design of some intricate and peculiar physical laws.

Following this line, it may be reasonable to use models in


which Pr(Oa | H2 & K) < Pr(Oa | H1 & K) to evaluate the
salient confirmation relations. If so, this opens the door to
Pr(E | H2 & K) < Pr(E | H1 & K). Think inductive Pr [10, 12]!
Branden Fitelson

Goodmans New Riddle

fitelson.org

Nicod & Hempel

Hempel & Goodman

Goodman & Bayesianism

Triviality?

Summary

References

Some would-be Bayesians modelers [13] assume something


even stronger than Pr(Oa | K) = 1: that K entails Ea & Oa.
Relative to Ks that entail both Ea and Oa, both H1 and H2
entail E. Then, E must confirm both H1 and H2 , relative to
K, and Pr(E | H1 & K) = Pr(E | H2 & K) = 1. Thus, (6) & (6?).
Another (more subtle) tactic used by Bayesians to ensure (6)
& (6?) is to argue that K entails that if as color is observed
(by S prior to t), then Oa will inevitably be learned (by S) [9].
And, something thats inevitably learned whenever salient
observations are made cant (from a Bayesian confirmation
theory point of view) be used to discriminate hypotheses.
Similar arguments have been made concerning observation
selection effects (e.g., Swinburnes firing squad example [14]).
Such arguments would have had little force for people like
Keynes or Nicod, who were talking about logical relations. It
is important here to distinguish confirmation and learning.
[Analogy: entailment vs inference in the case of deduction.]
Branden Fitelson

Goodmans New Riddle

fitelson.org

Nicod & Hempel

Hempel & Goodman

Goodman & Bayesianism

Triviality?

Summary

References

Goodman thought that the new riddle led (at least in the
context of Hempels theory of confirmation) to the triviality
result that everything confirms everything! Here is Part II:
. . . Moreover, it is clear that if we simply choose an appropriate predicate,
then on the basis of these same observations we shall have equal
confirmation, by our definition, for any prediction whatever about other
emeraldsor indeed about anything else. . . . We are left . . . with the
intolerable result that anything confirms anything.

So, Goodman wasnt merely describing a clever and hard


underdetermination problem. He was saying that his grue
problem reduces confirmation theory to utter triviality!
Goodmans triviality argument fails, even when aimed at
Hempel.1 And, from a Bayesian POV, its a non-starter.
Next, Ill run through the argument in the context of
Hempels theory. Then, Ill make some Bayesian remarks.
1

On this score, Hempel seems to have conceded too much to Goodman.


He seems to have accepted the triviality part of Goodmans argument [8].
Branden Fitelson

Goodmans New Riddle

fitelson.org

Nicod & Hempel

Hempel & Goodman

Goodman & Bayesianism

Triviality?

Summary

References

Heres my best guess as to how this Goodmanian triviality


argument against Hempels theory is supposed to run [8].
(i) E confirms both H1 and H2 .
[This is just (6), which we have already discussed at length.]
(ii) E confirms both Gb and Bb.
[where b = the first emerald examined after t (assuming
one exists), and G and B are incompatible properties.]
(iii) E confirms anything.
(iv) Mutatis mutandis, anything confirms anything.

While (i) is implied by Hempels theory, (ii) cannot be.


Hempels theory implies the consistency condition [5]:
(CC) If E confirms H1 and E confirms H2 , then H1 and H2 are
logically consistent (provided only that E is self-consistent).

All that follows from Hempels theory is that E confirms


both material conditionals: Ob Gb and Ob Bb.
But, even E & Ob cannot Hempel-confirm both Gb and Bb
[8]. So, Goodmans triviality argument fails at step (ii).
The step from (ii) to (iii) is OK in Hempels theory. . .
Branden Fitelson

Goodmans New Riddle

fitelson.org

Nicod & Hempel

Hempel & Goodman

Goodman & Bayesianism

Triviality?

Summary

References

. . . because Hempels theory implies the following pair:


(&C) If E confirms H1 and H2 , then E confirms H1 & H2 .
(SCC) If E confirms H1 , and H1 entails H2 , then E confirms H2 .

Finally, we can generalize (i)(iii) to yield something close to


(iv). For any observation statement E [Oa & a\, E
(Hempel-)confirms both [(x)x\ and [(x)(x Ox)\.
Then, by (i)(iii) mutatis mutandis, E confirms anything.
So, while Goodmans triviality argument against Hempels
theory fails at step (ii), it does have some prima facie
plausibility to it. From a Bayesian perspective, it does not.
First, (i) is not implied by Bayesian CT. And, even if (i) were
true, (ii) would not follow from it. Interestingly, (ii) could be
true in BCT [since BCT does not imply (CC)!], but (iii) would
not follow, since (&C), (SCC) arent implied by BCT either.
Using the properties from our counterexample to (M):
(CC): Ba confirms both Aa and Ja, which are inconsistent.
(&C): However, Ba doesnt confirm Aa & Ja.
(SCC): Nor does Ba confirm (Sa) a is some ace [Aa ` Sa].
Branden Fitelson

Goodmans New Riddle

fitelson.org

Nicod & Hempel

Hempel & Goodman

Goodman & Bayesianism

Triviality?

Summary

References

Main points:
Grue is not necessary for discrediting Hempels theory of
confirmation. That Hempels theory entails (M) is sufficient.
Grue is not necessary for establishing underdetermination
cases. Simple curve-fitiing cases with natural kinds will do.
Grue is not sufficient to establish Goodmans desired
triviality result. His triviality argument almost works on
Hempels theory, but its a non-starter for Bayesian theory.

Grand Conclusion of the talk:


EITHER
Goodmans New Riddle is not new, because its just an
underdetermination/indistinguishability example (these
had been constructed before without grue-like predicates).
OR
Goodmans New Riddle is not much of a riddle, because
its an unsound (or non-starter) triviality argument.
Branden Fitelson

Goodmans New Riddle

fitelson.org

Nicod & Hempel

Hempel & Goodman

Goodman & Bayesianism

Triviality?

Summary

References

[1] B. Fitelson, The paradox of confirmation, Philosophy Compass (B. Weatherson


and C. Callender, eds.), Blacwkell (online publication), Oxford, November 2005,
URL: http://dx.doi.org/10.1111/j.1747-9991.2005.00011.x.
[2] I.J. Good, The white shoe is a red herring, BJPS 17 (1967), 322.
[3]

, The white shoe qua red herring is pink, BJPS 19 (1968), 156157.

[4] N. Goodman, Fact, Fiction, and Forecast, Harvard University Press, 1955.
[5] C. Hempel, A purely syntactical definition of confirmation, JSL 8 (1943), 122143.
[6]

, Studies in the logic of confirmation, Mind 54 (1945), 126, 97121.

[7]

, The white shoe: no red herring, BJPS 18 (1967), 239240.

[8] C. Hooker, Goodman, grue and Hempel, Phil. of Sci., 35 (1968), 232247.
[9] J. Joyce, APA Sympoisum comments on an earlier version of this talk, 2006.
[10] J. Keynes, A Treatise on Probability, Macmillan, 1921.
[11] P. Maher, Probability captures the logic of scientific confirmation, Contemporary
Debates in the Philosophy of Science (C. Hitchcock, ed.), Blackwell, 2004.
[12] J. Nicod, The logical problem of induction, (1923) in Geometry and Induction,
University of California Press, 1970.
[13] E. Sober, No model, no inference: A Bayesian primer on the grue problem, in
Grue! The New Riddle of Induction (D. Stalker ed.), Open Court, Chicago, 1994.
[14]

, The Design Argument, in The Blackwell Guide to Phil. of Religion, 2004.

[15] R. White, A New Look at the Old Problem of Induction, unpublished manuscript.
URL: http://philosophy.fas.nyu.edu/docs/IO/1180/induction.pdf.
Branden Fitelson

Goodmans New Riddle

fitelson.org

You might also like