S. Yablo

third draft: February 3, 2000

A now-standard argument against physicalism attempts to wring a metaphysically possible zombie-world — a world just like ours in physical respects but with no consciousness — out of the fact that consciousness is not a priori deducible from physics. Crucial to this argument is the premise

(A) if it is not a priori that S, then there are worlds which construed as actual make S false.

Some authors (me), thinking that an argument whose premises are easy to reject is to that extent not a convincing argument, have pointed out how easy it is to reject (A) — how easy it is to explain S's non-a-priority without bringing in S-falsifying worlds. It's enough if our grasp of the set of S-verifying worlds leaves it open — gives the impression, even — that not all worlds are contained in it. Since our grasp can leave this open even if the set of S-verifying worlds is the set of all worlds, an S-falsifying world is not forced on us.

The anti-physicalist however has a ready reply: that it is possible to understand non-a-priority your way doesn't mean that's how we have to do it. Why shouldn't we stick with my original strategy of having a world for each a priori coherent supposition? That way non-a-priority is traced to a single source, rather than two sources as on your proposal. There isn't anything question-begging about the strategy since I willingly take it on as a separate obligation to show that worlds thus conceived are metaphysically possible.

How to reply to this reply? One has to find fault with the strategy of postulating a world for every a priori coherent supposition. One has to show that it is not just that non-a-priority can be explained otherwise than by way of witnessing worlds, in some cases it should be explained otherwise than by way of witnessing worlds. That's what I want to attempt now. The plan is as follows. It follows from (A) that

(B) if it fails to be a priori that all Ps are Qs, there is a world that construed as actual contains Ps which aren't also Qs.

Below we'll explore a certain way of constructing counterexamples to (B), and then see if the method can be extended to encompass the case where P = physical stuff and Q = pain. The idea in a nutshell is that you will have counterexamples to (B) when Q is a grokking predicate. By a grokking predicate I mean one that identifies its object in part by aspects of our experience of it that don't purport to be representational, or at least don't purport to be representing features of the object itself.

For example: A certain kind of line drawing will be seen by anyone who looks at it as a human face. I don't mean that everyone will judge it to resemble a face or to represent a face, just that we cannot easily stop ourselves from "seeing a face in it" and forming associated judgments, e.g., the face looks cruel or alarmed or what have you. I will call line drawings like this "facical." What makes "facical" a grokking-predicate is this: the aspect of our experience whereby we "see a face" in this sort of drawing does not purport to represent the drawing as being any particular way. It's not as though one's experience would be non-veridical if the drawing didn't in fact look much like a face or if no one else grokked it that way. (That the face-seeing aspect of our experience is not representational does not mean that there's no definite geometric property of facicality. It's just a question of how the property is picked out; in this case we pick Qness out via a feature of our experience that makes no claims about Qness.)

Why should grokking predicates be a source of counterexamples to (B)? This will take a little while to explain. It's familiar from Kripke that in evaluating a sentence with respect to a world considered as counterfactual, how the inhabitants of that world understand the sentence is irrelevant. If it was the inhabitants' understanding that mattered, then it would be true rather than false that "if "leg" had applied to tails, horses would have had five legs." The idea is to take our sentence, understood our way, and ask whether it is true of the given world considered as counterfactual "True of" here expresses a transworld relation between our sentence and their world.

So much concerns evaluation with respect to worlds taken as counterfactual. Although it's less widely appreciated, similar points apply to worlds taken as actual. The proper method is to take our sentence, understood our way, and ask whether it is true in the given world considered as actual. Thus Chalmers:

Just as we saw with Kripke, Chalmers's theory gives what he would regard as the wrong results if evaluation is construed as intra-world rather than cross-world. He does not want it to come out a priori that, say, there are truth-bearers. By (A), then, he needs a world which taken as actual makes "there are truth-bearers" false. Such a world obviously cannot contain truth-bearers; so the token of "there are truth-bearers" that gets evaluated with respect to that world had better be here, not there.

Now it is time to bring grokking-predicates back into the picture. Take "facical" introduced above, and consider a world w about which all I'm going to tell you is that it contains Figure F:

Is "facical" true of Figure F in w considered-as-actual? The answer is yes, and it remains yes no matter what I go on to say about the observers in w: that they see nothing in the figure, that it looks to them like a battleship, or whatever. The reason as before is that we evaluate the figure with respect to our word "facical," understood as we understand it. Our dispositions to see faces in presented figures figure crucially in that understanding, so they are part of what we (imaginatively) bring to bear on the figure in w.

Suppose we introduce P as an objective, third-personal, predicate applying to all and only line drawings with such and such geometrical properties, the properties exemplified above by Figure F. Q is the predicate "facical" as already explained. What is the truth-value of "if something is P then it's facical" in actual-world-candidates other than world w? Given that we know nothing about w beyond that Figure F occurs in it, and that the features of Figure F that contribute to its classification as facical are summed up in predicate P, there seems little option but to say that "if something is P then it's facical" is true in any world considered as actual.

But if "Ps are facical" is true in all actual-world-candidates bar none, then principle (B) makes a prediction: it predicts that " Ps are facical" is true priori. This is a problem because intuitively, and I believe for Chalmers also, you have to do some experimenting to realize that Ps are facial: you have to expose yourself to Ps and ask yourself if you see faces in them. Reflecting on the geometrical property will not do it, no more than reflecting on the geometry of Muller-Lyer lines will reveal to you that one will appear longer.

I suggest then that "Ps are facical" is a counterexample to principle (B). More things are true in every candidate-for-actuality than are true a priori. The reason is simple enough. When we evaluate sentences in candidates-for-actuality, we are allowed to exercise any dispositions that inform our understanding of the relevant words; we are allowed in particular to check how various items considered-as-actual strike us or make us feel. When we ask about a sentence's status as a priori or not, we are not as generous about allowable methods of verification. A sentence that we cannot know to be true without self-experimentation is not considered a priori. (Otherwise it would be a priori which geometrical figures make for optical illusion!)

The moral so far is that claims employing grokking predicates can be non-a-priori without a falsifying world. And now a conjecture: "pain" is a grokking predicate. Pain is picked out at least partly in terms of non-representational aspects of our first-personal experience of pain, viz. the hurting aspect. (Compare the seeing-a-face-in-it aspect of facicality.) If the conjecture is granted, then someone who thinks that pain = c-fiber-firings can defend her position as follows.

(1) What is so special about grokking-predicates? Why do you focus on predicates that pick out their referent R by way of non-representational aspects of our experience of R?

Reply: Two questions here really. (i) Why focus on predicates picking out R via our experience of R? (ii) Why focus on predicates doing this via non-representational aspects of our experience?

The answer to (i) is that we want predicates whose associated concepts (narrowly understood) presuppose certain aspects of our actual sensibility. The reason for wanting that is that we want it to be our actual sensibility -- not our as-if-actual sensibility — that comes into play when we compare these predicates to objects in as-if-actual worlds. For a predicate to pick out its referent by way of our experience of it is a special case of a predicate's picking out its referent by way of its impact on some aspect of sensibility. Instead of experience we could just as easily have fixed on, e.g., seeing a thing as loveable or meritorious or silly.

The answer to (ii) is that we want predicates whose associated concepts force us to "sample" a thing before counting it into the predicate's extension. Suppose that Q invokes only representational aspects of our experience of its referent. An example might be the way that "round" picks out roundness. Grant me for argument's sake that a proper understanding of "round" involves the disposition to experience round objects in a certain way; grant me in other words that condition (i) is met. But, of course, there is no need to experientially "sample" an object in w-considered-as-actual to decide whether it is round; it's enough to ask whether it has the property our experience attributes to it, viz. the property of consisting of a bunch of points equally far from a central point. It's only when a predicate operates via non-representational aspects of experience that sampling is needed. There's no possibility with predicates like this of listing the conditions a thing has to satisfy to be Q; we have to expose ourselves to that thing and see if experiences with the relevant non-representational features do indeed ensue.

(2) (Chalmers) Isn't a grokking-concept very like what others have called a response-dependent concept? What makes the concept of facicality so different from that of, say, provocativeness or pleasurability? The concept of pleasurability is the concept of being such as to elicit pleasure-reactions. Isn't facicality similarly the concept of being such as to elicit "seeing-faces-in" reactions? The reason this matters is that when response-dependent concepts are involved we do expect a witnessing world. That eating strawberries is not a priori pleasurable goes with the expectation of a world where strawberries taste terrible.

Reply: First let's distinguish weaker and stronger notions of response-dependence. The weaker says that the concept expressed by P is response-dependent iff it is a priori that something is P iff it is apt to evoke thus and such responses in thus and such subjects. I don't necessarily want to deny that the concept of facicality is response-dependent in the weak sense. (It can be, and has been, argued that all concepts are.) A concept is response-dependent in the stronger sense iff it's the concept of a certain sort of disposition: the disposition to evoke thus and such responses in thus and such subjects. (This is more or less what Mark Johnston in "Objectivity Refigured" calls a response-dispositional concept.)

From this point on let's use "response-dependent" in the strong sense only. With that understood, I'm prepared to concede the following: if grokking-concepts were response-dependent concepts, the objection would be correct.. Whether "pleasurable" applies in w-construed-as-actual depends on the responses of the as-if-actual population. If the concept of facicality were response-dependent, then the applicabiity of "facical" would depend on the responses of the as-if actual population too.

How, you ask, are grokking-concepts not response-dependent? Response-dependent concepts are descriptional concepts which invoke at a certain point to a certain kind of experiential response. The content of a grokking-concept is supposed to be the content of that experiential response. When I experience a figure as facical, I experience it as having a certain geometrical property. I don't experience it as: apt to cause facicality-experiences.

The arguments for this bear comparison to the objections that have been raised to the view that to experience a thing as red is to experience it as disposed to cause experiences of that very sort. (See Boghossian and Velleman, "Color as a Secondary Quality.") For one thing, the attributed content is implausibly sophisticated. Visual experience isn't "about" experience or the causation of experience; it's more naïve than that. Second, if the metaphysicians are wrong and there is no such thing as causation, my experiences of things as red, or as facical, are not automatically falsified. They would be if they had causation written into their truth-conditional content. Third, there seems no way other way to identify the intended response than as the experience of a thing as red, or facical. Which leads to a circularity problem; experiencing X as red would be experiencing X as disposed to cause observers to experience it as red, ie., as disposed to cause observers to experience it as disposed to cause observers to….

(3) (Block) Your argument assumes that our actual seeing-face-in dispositions are part of the narrow or a priori content of "facical." If that were true, then it should make no sense to imagine ourselves discovering that it is another disposition that really defines "facical," a one we had confused for good and specific reasons with our seeing-face-in-dispositions. But it does make sense. So our actual dispositions cannot possibly be part of "facical"'s a priori content.

Reply: One could just as easily argue that unmarriedness forms no part of the a priori content of "bachelor." It can't be ruled out a priori that someone should turn up to convince us that we considered unmarriedness necessary only because secret semantic agents were messing with our minds. The proper conclusion to draw is not that "bachelor' ("facical") has a thinner priori content than we had supposed,

(4) (Block) If we define "facical" to mean "looks face-like," then the world in which Ps looks battleships is a world in which Ps are not facical. "All Ps are facical" is not true in all worlds-considered-as-actual AND not a priori. If we define "facical" so as to make it a priori that Ps are facical, then the world in which Ps look like battleships is world in which Ps are facical (though we don't appreciate it). "All Ps are facical" is true in all worlds-considered-as-actual AND a priori. So where's the counterexample?

Reply: I don't define "facical" in either of those ways. The meaning of "facical" is supposed to ride piggy-back on the content of our experiences of seeing shapes as facical. These experiences have a non-representational part — the part where we see a face in the shape — and a representational part — the part where we see the shape to have a certain geometry. I tried hard not to say that the representational part represents by way of the non-representational part: that would give us Block's second option, on which "facical" means "looks face-like" or "is such as to elicit a seeing-face-in response" or the like. I also tried hard not to say that the representational part floats completely free of the non-representational part: that would give us Block's second option, on which "facical" is defined in geometrical terms and "all Ps are facical" comes out a priori.

(5) (Block, Nagel) But what space is there between these two options? You seem to be saying that "facical" trades on the seeing-in stuff without representing by way of the seeing-in stuff. What does that mean?

Reply: The third option is that this is how the representational content comes; seeing a face in the figure is how one is representationally appeared to in the relevant way. A helpful analogy might be this. Homer, lacking knowledge of physics, had no way to form beliefs about mean random K.E. without thinking about heat. He didn't think about his heat-idea; he didn't think that whatever out there corresponds to my heat-idea is pleasant. No, he thought about heat, which is to say that he thought about mean random K.E. in the heat-ish way. It was by believing that heat was soothing that Homer believed that mean random K.E. was pleasant.

Back to facicality, the knowledge that I lack is not physical but geometrical. I lack a geometrical specification of the (geometrical) property G that all and only the facical things have. This leaves me with no way to mount an experience of a thing X as G but to see a face in X.

(6) (Chalmers?) The analogy works against you. Where is the heat in a world-considered-as-actual in which it's UVW that feels warm? It's where the UVW is. But then the facicality in a world-considered-as-actual in which people see faces in other sorts of figures is where those other sorts of figure are.

Reply: There's a difference. "Heat" is the sort of word that can change its referent from one actual-world-candidate to another without there being any change in its narrow meaning. This is because the worldly facts whereby "heat" refers to what it does are largely outside the head, hence not the sorts of fact that a competent user needs per se to know about. But "facical" cannot change its referent from one actual-world-candidate to another without a change in its narrow meaning. Someone who appreciates the geometry of figure F, but finds it nevertheless not to be facical, doesn't mean what we do by "facical." Let that be a stipulation if you like. (A defeasible one, since it could turn out that it was not figure F we were looking at but a different figure.)

That someone who fails to "see pain in brain state B" doesn't really understand "pain" is not the sort of thing that can be stipulated, of course. But that's not important. The question is whether we would ever be in a position to regard it as true. I take it that we could in principle stumble on psychophysical correlations of such overwhelming predictive power that we were sorely tempted to say: pain = brain state B. Suppose that such a thing were to occur. Along comes Painless Pat who is incapable of getting into brain state B; she can only observe it from the outside. Someone points out that there are worlds w which considered-as-actual contain brain state B but which (according to Pat) contain no pain. (Maybe w = @, but I won't assume that.) Actual-Pat asks: doesn't as-if-actual Pat's negative judgment show that "where there is brain state B, there is pain" is false some worlds considered as actual?

I think it is clear how we would respond. It doesn't matter what Painless Pat thinks, because she doesn't understand "pain," because she is unable to feel pain, because she is unable to get into the brain state that is pain. Another way to put it is that pain is a grokking-concept and she lacks the ability to grok it.

(7) (Chalmers) What can "a priori content" possibly mean, if not: the aspect or component of a word's content that can be ascertained a priori? You seem to be admitting that our actual reactive dispositions aren't part of "facical"'s a priori content in that sense. What other sense is there?

Reply: The words "a priori content" play two quite different roles in this debate. One role is semantical. It's the aspect of content that separates those competent with the word from those not competent, where competence here is to be as independent as possible of a posteriori worldly knowledge. As a result, it's the aspect of content that has to be held fixed in evaluating Q with respect to an as-if-actual world.

An example: Everybody knows what Twin Earth is. Earth-Twin is a place where externalities are as here on Earth, but "water" means silver. A priori content is needed to mark the line between

the water in Twin-Earth considered-as-actual is XYZ

which we do want to say,

the water in Earth-Twin considered-as-actual is silver,

which we don't want to say. Call this the "a priori competence" role. To say that a priori content plays the a priori competence role is to say that the judgments of as-if-actual speakers who disrespect a word's a priori content have have no real bearing on that world's extension in their as-if-actual world.

So that's one role. The other role a priori content plays is epistemological. It's the aspect of content that speakers are in a position to know about just by virtue of their a priori competence with the word. Call this the "a priori knowledge" role.

Some think that there's a single notion of content that can play both of these roles. But that's a substantive assumption, which there is reason to doubt. When I say that our actual reactive dispositions are part of "facical"'s a priori content, it's the competence-driven notion of a priori content I have in mind. Whether the a-priori-competence content of "facical" is a priori knowable is part of what's at issue.

(8) (Chalmers) Granted your distinction, I "deny that competence-apriority is relevant to PI evaluation. in effect, what's "competence-apriori" may include various aspects of mental functioning that are best not counted as a priori, and so are not relevant to PI evaluation. What's a priori isn't just what one can know by virtue of competence with a concept -- it's what one can know justified independently of experience by virtue of one's competence with a concept. And reactive dispositions that proceed via self-experimentation don't yield experience-independent knowledge."

Reply: You say that for X-c.a.a. (considered-as-actual) to get into a word W's PI, agents-c.a.a have got to be justified independently of experience in calling X a W. Then it would seem on the face of it that "facical" never has anything in its primary intension: because there's supposed to be no way of telling independently of experience whether a figure is facical. Given the rest of the framework, that would seem to make it a priori that nothing is facical. And that is the wrong result. (I'm assuming that the lower-level descriptions by which worlds are presented don't make mention of facicality.)

(9) (Chalmers) "I'm not seeing why the PI of "facical" should come out empty. As I understand things, it is a priori (normal sense) that if G causes a certain sort of experience, then G is facical. So if we consider as actual a world in which G causes those experiences, we can determine a priori that G is facical. so the primary intension of "facical" picks out G in that world."

Reply: OK, "empty" is too strong. But the point still stands. Take a thing-c.a.a. that does not cause any experiences, say because no one in the relevant world looks at it. Such a thing might still be facical. Indeed you'd need some such things to go into "facical"'s PI to explain the non-a-priority of "if a figure is facical,then someone experiences it." How in your view does such a thing get into the PI? Offhand this looks like one of the cases where we've got to bring our actual-actual concept to bear on it. And it's not clear how we can/could do that without forming a view about whether the figure is such that, imagining the juxtaposition as somehow or other arranged, we here see a face in it there. To run it the other way around, if we've got to know a priori that the unobserved figure is facical, then it doesn't get into the PI, and the PI comes out not empty but emptier than we would like.

(10) (Chalmers) "I agree that it's not a priori that if something's facical, then someone experiences it. But I do think it's a priori that if something's facical, it's the sort of thing that we'd experience as face-ish (at least that it's relevantly similar to things that we'd experience as face-ish)."

Reply: Since the issue is how to get unobserved figures it into the PI of "facical," I'll take it you're suggesting it's a priori too that if X is the sort of thing we'd experience as face-ish, then X is facical. This would indeed explain how unobserved figures made it into the PI of "facical." But only on a certain assumption: the lower-level facts in terms of which worlds are presented are allowed to include facts to the effect that subjects like ourselves would (if suitably positioned) grok the figure in such and such a way.

I see two related problems for the assumption. One is that, if facts like that have a place in lower-level world-descriptions, then presumably they should be written too into our description of the "zombie"-world. That is, a world physically like ours has not been fully described until it's specified whether subjects like ourselves would (if suitably positioned) grok c-fiber firings in a painful way. Since the identity-theorist presumably thinks that the only permissible specification here is that we would feel pain in the envisaged circumstance, it's not clear why she can't just deny the dualist's claim that there's a world physically like ours but with no pain. Because, to repeat, whether or not there is pain is a world-c.a.a. depends on whether or not "pain" in our mouths stands for anything there; and that depends on whether anything there would be experienced as pain were we given the right sort of first-personal access to it.

The dualist might perhaps reply that there's another sort of physical duplicate of our world whose c-fiber firings we wouldn't experience as pain were we granted first-personal access to them. (There is no pain in that world, hence there's nothing that we would experience as pain.) But, leaving aside the fact that the identity theorist is under no evident pressure to agree with this, a problem now emerges about the epistemic status of claims to the effect that "there are worlds which c.a.a. make it true that S." It is crucial to the modal rationalist's project that claims like this be evaluable a priori. But we have just seen that the availability of a world which c.a.a. makes "things are physically like so but there is no pain" turns on the availability of a world which c.a.a. is physically like ours but contains nothing that we (our actual-actual selves) would experience as pain. And whether a physical duplicate of our world has got to contain something that we would experience as pain does not seem to be a matter that can be settled a priori.

After all this maybe I can make my point more directly. At bottom it concerns the ability of the 2D framework to handle certain possible-looking semantic phenomena. The phenomenon I have primarily in mind has to do with two categories of sentence.

Consider first the meaning-suffices-for-truth sentences: for short the analytic sentences. These are sentences S whose "narrow" or competence-apriori meaning is such that every w-c.a.a. makes S (with that narrow meaning held fixed) true. A sentence meaning that has got to be true no matter what the actual world may be like.

Note what I haven't said about these sentences. I haven't said that we can tell just by reflection on our a priori semantic knowledge that they are true come what may. I haven't said that it could never be rational for a competent user to wonder about their truth-value, or indeed even to deny them. Sentences with those sorts of properties can be called grasp-of-meaning-suffices-for-knowledge-of-truth sentences: for short, a priori sentences.

From the way I have described the two categories, it is not hard to see how an analytic sentence could be a priori. It would be enough (or almost enough) if grasp of S's meaning carried with it an ability to intellectually contemplate worlds c.a.a. and to decide on the basis of that contemplation whether S comes out true or false.

By the same token, it not hard to see either how an analytic sentence could fail to be a priori. It would be enough (or almost enough) if grasp of S's meaning — meaning by this whatever it takes to be competent with S — did not carry with an ability to intellectually contemplate worlds-c.a.a. and to decide on the basis of that contemplation whether S comes out true or false. One way for this to happen is if grasp of S's meaning takes the form of an ability to tell whether S is true when confronted in the right sort of way, not with a representation of the relevant circumstances, but with the circumstances themselves.

The point of the facical example was to illustrate this possibility. "All Fs are facical" is put forward as an example of a sentence that is analytic without being a priori. Because it is not a priori we can imagine it coming out false. Because it is analytic there are no worlds where it does come out false: any worlds where it seems to do so are worlds where it turns out to be another sentence, homonymous but with a different narrow content, that is false. It is easy for us to confuse this with a world where S, meaning what it actually does, to comes out false, just because our grasp of S's meaning does not give us a priori knowledge of what S means, not even in the attenuated sense of giving us the ability to classify worlds as S-verifying or not on an a priori basis.

Another sentence that might, for all I can see, be analytic but not a priori is "all c-fiber firings are pains." Our grasp of "pain"'s meaning consists in large measure in our ability to recognize pains when presented with them in a first-personal way. Unless one assumes from the start that that it is not an admissible sort of meaning — that grasp of meaning has got to take some other form — there seems no reason to expect that my competence with "pain" should put me in a position to tell whether there can be c-fiber firings that aren't pains.

As far as I understand Dave's response to all this, he wants to deny the possibility of analytic truths that aren't a priori. This requires him to deny as well that there can be words that operate the way I say "facical" does. He'd prefer to interpret "facical" as expressing the response-dependent concept: x is of the right sort of shape for people like me to see faces in x. I reply that while it is indeed synthetic that Fs are facical_D (facical in Dave's sense), that just shows that "facical_D" has a different sort of narrow content than "facical." Someone who sees a battleship and not a face in the frowny-figure doesn't understand "facical" the way I do.

Perhaps there's a question of whether this is the sort of thing you can stipulate; but if so I'm not seeing it. A word with the narrow meaning I'm proposing for "facical" seems highly possible. There would seem indeed to be well-understood words of English with the right general features. "flowing," "tranquil," "bracing," "mellifluous." The worry, and at this point that's all it is, is that the 2D-theorist will be tempted to distort the intuitive meanings of words like this to make them fit the framework. That's fine; let him/her have the particular words. What would still be lacking is a general reason why words cannot be made to behave the intended way, or hence a reason why analytic truths should be expected to be a priori.