knowledge argument
An argument from Frank Jackson (1982) purporting to show that physicalism is false on the ground that there exist facts that cannot be known solely in virtue of knowing all the physical facts. See dualism, consciousness.

The knowledge argument goes as follows:
  1. If physicalism is true, then one can know all the facts there are just in virtue of knowing all the physical facts.
  2. What it is like to have experiences as of red things is a fact one cannot know just in virtue of knowing all the physical facts.
  3. Therefore, physicalism is false.

Jackson vividly illustrates his argument with the now famous example of Mary, a brilliant scientist who, by hypothesis, learns all there is to know about the physical world.  But Mary was raised from birth in a black-and-white environment, and so does not know what it is like to have experiences of red.  When she leaves her environment and sees her first ripe tomato, she undoubtedly learns something about the world, viz., what it is like to experience red.  But if physicalism were true, she would have already known that, because physicalism is the thesis that all facts are physical facts--that is, that to know every fact expressible in  physical language is to know everything.  Therefore, physicalism is false.

Further discussion focuses on replies to the argument, and is organized as follows:
  1. Unhelpful replies
  2. The opacity-of-knowledge reply
  3. The ability hypothesis
  4. The a posteriori physicalism reply
Unhelpful replies
(a) Failure of Integration
We can know both that P and that P entails Q, yet not have bothered to draw the logically valid inference.  According to this reply, that is precisely how it is with Mary: she knows everything there is to know in the sense that she knows enough to infer everything, and simply has not drawn the inference concerning what it is like to experience red.
This reply seems to reflect an inability to grasp the nature of the problem Jackson poses for physicalists.  It seems that we can give Mary enough time for reflection as she would like, and give her as much prompting as we can, and she still would be unable to come to know what it is like to experience red just in virtue of knowing all the physical facts.  By contrast, we are inclined to think that when some logically competent person cannot infer that P in virtue of knowing some set of propositions S--despite reflection and prompting--the reason is simply that S does not entail that P.
(b) "Mistaking a Failure of Imagination for an Insight into Necessity" (Dennett 1991, pp. 398-406)
In fact, Mary would know what it is like to experience red in virtue of knowing everything physical: for example, if she were presented with a ripe tomato painted blue, she would not think that it was red, but would know that it was blue.  The reason is that, in virtue of knowing everything physical, she would know what effect seeing blue would have on her nervous system, and thereby would know that she was seeing blue, not seeing red.  If we truly imagined Mary's knowing everything, we wouldn't think she doesn't know what it's like; but imagining Mary's knowing so much is too taxing for our imaginations.
This reply also seems to misconceive the nature of the problem Jackson poses for physicalists.  It is irrelevant that Mary, upon being shown a colored object, can correctly identify its color just by looking at it and drawing upon her vast knowledge; what matters is whether or not Mary learns something upon being shown a colored object--that is, upon having her first color experience.
(c) Experience is the only teacher of color facts
A strongly empiricism view of concepts holds that one can have color concepts only in virtue of having color experiences.  Because Mary has never had color experiences, she lacks color concepts.  Therefore, even if all facts are physical facts, she cannot know everything physical without having had color experiences.
There are at least two serious problems with this reply.  First of all, it stands or falls with the empiricist view of concepts it presupposes, and there is little reason for believing this view.  Secondly, this view is inconsistent with physicalism, and so is of no comfort to the physicalist.  One might think the reply could be saved by defense of a weaker version of physicalism--a brute supervenience thesis, perhaps--but then the appeal to how one learns color facts drops out, and the reply becomes a version of the a posteriori physicalism reply.
(d) "There Must Be a Reply" (Jackson 1995)
Jackson himself now thinks that the knowledge argument is unsound, but acknowledges that he does not know what the proper refutation of it could be.  We cannot really characterize this view as a reply; rather, it is a brute belief that something must be wrong with the argument.
The opacity-of-knowledge reply

This kind of reply is by far the most popular in the literature.  For some of the more classic examples, see Horgan (1984), van Gulick (1985), Churchland (1985), and Lycan (1987, Chapter 7).  For an interesting and subtly argued version of this reply that also addresses Saul Kripke's modal argument, see Loar (1997).

The knowledge argument admittedly relies on the obscure notion of a fact, and given its obscurity, it is not at all clear that when Mary leaves the room and has her first color experience, she learns something new in the sense of coming to know a new fact.

We might think of facts as (true) propositions--that is, as objects of propositional attitudes.  Alternatively, we might think of them as "actual states of affairs" or some such things that are themselves what "make" true propositions true but are not themselves propositions.  However we understand facts, though, it seems that one may plausibly hold that while every fact about experiences of red is a fact about (say) neural states, it is possible to know the neural facts in virtue of which there is something it is like to experience red without knowing what it is like to experience red.  That is, it may be that the sentences 'It is like such-and-such to experience red' and 'To experience red is to be in such-and-such a neural state' are made true by the same facts, but express different propositions; or it may be that while the two sentences express the same proposition, these propositions themselves can be believed/known under distinct modes of presentation.

In this way, we can explain why Mary learns what it is like to experience red without having to suppose that Mary learns new facts: because 'that' clauses create opaque contexts--we cannot in general substitute co-referential terms inside 'that' clauses salva veritate--we cannot reject the possibility that in coming to know what it is like to experience red, Mary came to know a fact she already knew under a different mode of presentation.

One problem for the opacity-of-knowledge reply is that we typically explain failure of substitution by appeal to a believer's knowing only some of a thing's properties.  We explain why Fred doesn't know that Frances Gumm could sing--even though he knows that the star of The Wizard of Oz could sing--by pointing out that Fred doesn't know that Frances Gumm had the property of being the star of The Wizard of Oz.  Unfortunately, Mary by hypothesis knows all the properties of things if physicalism is true, because if physicalism is true she knows everything.

A way around this problem is to appeal, as Loar (1997) does, to the notion of a phenomenal concept.  (See Lycan (1996, Chapter Five) for a similar strategy.)  Phenomenal concepts are "self-directed" recognitional concepts employed in ordinary perceptual and sensory experience, and it is plausible that such concepts can be had only in virtue of having relevant experiences.  Nevertheless, such concepts pick out phenomenal properties (such as the qualitative properties of red experiences) directly, not via some distinct mode of presentation.  (Alternatively, we might characterize phenomenal properties as "self-presenting".)  For Loar, then, concepts play the role of modes of presentation, and concepts that aren't a priori related to one another may pick out the same object even if they aren't associated with distinct properties of that object.

Even if we accept Loar's view that there are phenomenal concepts, however, it is not clear that the opacity-of-knowledge reply suffices to refute the knowledge argument.  The reason is that the reply entails that Mary can know about the qualitative properties of experiences in two different ways, and that she can plausibly know about these properties in one way without thereby knowing it in the other.  This is borne out by the a posteriority of identifications of qualia with physical properties.  But if identifications of qualia with physical properties are a posteriori, it can only be because it is possible to coherently doubt their truth--that is, to coherently conceive their falsity.  Moreover, it seems that conceivability is generally a guide to possibility, and that identities, if true, are necessary.  Therefore, the defender of the opacity-of-knowledge reply must in some way deny the inference from conceivability to possibility, and it is not obvious that this inference can be plausibly denied in a way that refutes the knowledge argument.  For discussion of this issue, see Yablo (2000).

The ability hypothesis

This reply is far less popular than the opacity-of-knowledge reply, but it is well known largely because it is so ably defended by Lewis (1983, 1990).  It is original to Nemirow (1979); versions of it are defended by Levin (1986) and Churchland (1989).

Defenders of the ability hypothesis deny that knowing what it is like is factual knowledge at all.  Instead, it is mere know-how, the possession of an ability.  (For discussion of the distinction between factual knowledge and know-how, see the entry on tacit knowledge.)  If this is correct, then Mary learns no new fact upon coming to know what it is like, just as she would learn no new fact upon coming to know how to ride a bicycle.

There are three important objections to the ability hypothesis.  The first has to do with the intuition that we can draw inferences from our knowledge of what it is like: if this intuition is correct, then because we can't draw inferences from know-how, knowing what it is like can't be a mere ability.  This is a problem for explaining how knowledge of what it is like can be embedded in conditional reasoning, so let us call it the embedding problem.  (The embedding problem was first raised in the context of the meta-ethical theory of emotivism, by Geach 1960.)  The second is that the best general analysis of knows wh- locutions entails that knowing what it is like is factual knowledge; Lycan (1995), who first offered this objection, calls it the argument from meaning and syntax.  The third objection is that when we know what it is like, we not only possess know how--which is knowledge of ourselves--but also know something about other people; let us call this the third-person objection.

Lycan (1996) and Loar (1997) each think the embedding problem serious enough to count as a reason for denying the ability hypothesis.  In fact, however, the embedding problem isn't really a problem at all.  The reason is that phrases of the form what it is like to F are noun phrases, and so can't be embedded in conditionals alone.  Of course, sentences of the form x knows that what it is like to F is G may be embedded in conditionals, but defenders of the ability hypothesis would agree that such sentences express factual knowledge without denying that knowing what it is like is an ability.  (No one denies that that what it is like to taste chocolate is delightful is a proposition; what defenders of the ability hypothesis deny is that knowing what it is like to taste chocolate is knowing a proposition.)  Those who think the embedding problem is a problem, in fact, cannot describe what is known in knowing what it is like merely by using the phrase "what it is like", but must instead use 'that' clauses such as "that pains feel like such and such".  Unfortunately, defenders of the ability hypothesis would not accept that knowing what an experience is like should be understood in terms of knowing that the experience is like such and such.  For further discussion of this point, see Lewis (1990), pp. 501-2.

Lycan's (1996) argument from meaning and syntax appears to pose a serious problem for the ability hypothesis, because it seems to show that we cannot analyze "knows what it is like to F" merely as "knows how to imagine F" or in terms of any other ability: instead, we must analyze it as "knows that it is like Q to F".  But it may not be as serious as Lycan thinks it is.  Defenders of the ability hypothesis do not put it forward as a linguistic analysis, but instead as an analysis of the phenomenon of knowing what an experience is like.  Because what it is like to have an experience is supposed to be in some important way ineffable and incommunicable, we should not expect a general analysis of knows wh- locutions to help us understand knowing what it is like: in fact, we may expect that knowing that it is like such and such to have an experience is knowledge that does not fully capture knowing what it is like.  And this is precisely what Lewis (1990, pp. 501-2) claims.

The third-person objection, first offered by Jackson (1986), also can seem serious: knowing what it is like is also knowing what it is like for other people, but knowing how to imagine one's own experiences is not the same thing as knowing what it is like for other people.  But this objection, too, is less serious than Jackson supposes.  The qualitative character of a given experience can be understood not only as a token quale, but also as a type.  Thus, in knowing how to imagine an experience, one can imagine a type of experience, a type others would instantiate if they had experiences relevantly like it.  So in knowing what it is like, one knows what it is like for anyone who has experiences of that type--that is, one simply knows how to imagine experiences of that type.

Given the discussion above, it would seem that the three main objections to the ability hypothesis are inconclusive at best.  There is, however, another objection to the hypothesis that cannot be so easily set aside.  It can seem that, in introspection, we can demonstratively refer to our own qualia.  Jackson can argue that when Mary has her first experience of red, she comes to know that that is what the experience of red is like--that is, she can demonstratively refer to the qualia associated with having such experiences.  But if she comes to know that, she could not have already known it, and once again the falsity of physicalism follows.  It would seem that the defender of the ability hypothesis, at this point, has no choice but to deny that knowing that that is what the experience of red is like is factual knowledge.  But it seems obvious that we can demonstratively refer to the qualities of our own experiences.  For this argument to be rebutted, we must show that in knowing that that is what it is like to experience red, Mary knows no new fact; that is, we must show that the opacity-of-knowledge reply is valid.

Interestingly, Lewis (1990) has given up the ability hypothesis, as understood above.  Instead, he favors what we might call an impure ability hypothesis, according to which knowing what it is like may well be both an ability and factual knowledge.  It is worth pointing out that the impure ability hypothesis, if it is to be a response to the knowledge argument, must reject the view that the factual knowledge one gains in knowing what it is like is non-physical: hence, this version of the hypothesis reduces to the opacity-of-knowledge reply.

The a posteriori physicalism reply

According to a posteriori physicalists, physicalism is the thesis that everything is necessitated by the physical.  This thesis is not obviously equivalent to physicalism as it is often understood: it is often held that reduction requires property identifications, and also that property identifications must hold a priori if they are to hold at all.  (For an example of this style of thinking, see White 1985.)

According to a posteriori physicalists, then, not everything necessitated by the physical need be a priori necessitated by the physical.  That water is H2O, for example, is necessitated by the microphysical facts, but not a priori necessitated by those facts.  Therefore, it is plausible to think that facts about what it is like to experience red are necessitated by the physical, but not a priori necessitated; and this suggests that Mary need not be expected to know what it is like to experience red just in virtue of knowing all the facts expressible in physical language.

Of course, the a posteriori physicalist cannot hold that there are non-physical facts that are necessitated by the physical facts--that would be tantamount to dualism--but must distinguish between propositions and facts, and hold that all propositions are necessitated, but not all a priori necessitated, by the totality of propositions expressed in physical language.  Therefore, this reply can only be a version of the opacity-of-knowledge reply.  Even so, Jackson (1994) offers an interesting argument against a posteriori physicalism, which is the focus of this section.

Jackson argues that the water/H2O example does not support a posteriori physicalism, but instead helps him argue for a priori physicalism.  His argument depends on the premise that all the propositions expressible in physical language, in conjunction with certain propositions about the references of our words 'water' and 'H2O', together a priori entail that water is H2O.  The propositions about reference Jackson has in mind are propositions saying how it is that the truth of utterances involving genuine occurrences 'water' and 'H2O' depend on worldly context: for example, the truth an utterance of 'water is wet' in a world depends on the stuff satisfying the description 'stuff that actually falls from the sky, fills our rivers and oceans, etc.' being in the actual world, because that description fixes the reference of 'water' in any given world.  But we know from knowing enough about the physical constitution of the actual world that the actual stuff that satisfies that description is H2O.  Therefore, if we know enough about reference, and we know the physical facts, then we can a priori deduce that water is H2O.  But this knowledge about reference is entirely physical, or is according to physicalists.  Therefore, simply in virtue of knowing the physical facts, we can a priori deduce that water is H2O--that is, the example of water and H2O only adds a wrinkle to a priori physicalism, and in no way refutes it.

Let us see the argument more formally.  Let P be the conjunction of all the physical propositions, and let C be the conjunction of propositions about how truth is determined by worldly context.  Then Jackson holds that the proposition

If P & C, then water is H2O

is knowable a priori.  Of course, C may not be one of the conjuncts of P; but Jackson alleges that any physicalist must hold that the way in which truth depends on worldly context is physical--indeed, what else could the physicalist say?  So it would seem that C is one the conjuncts of P.  But then physicalists must hold that the proposition

If P, then water is H2O

is knowable a priori--that is, they must deny a posteriori physicalism.

If Jackson's argument is correct, then the physicalist cannot rebut the knowledge argument merely by appealing to the a posteriori of psychophysical identifications.  But, in a way, this is precisely what those who would offer the opacity-of-knowledge reply appeal to.  Therefore, those who would offer this reply must in some way rebut Jackson's argument.  For further discussion of this argument, see Byrne (1999) and Yablo (2000).


Adam Vinueza