Sellars's Ryleans Revisited

Wilfrid Sellars's essay, "Empiricism and the Philosophy of Mind,"⁽¹⁾ introduced, although it did not exactly endorse, what many philosophers consider the first defense of functionalism in the philosophy of mind and the original "theory" theory of commonsense psychology.⁽²⁾ I will be posing a problem: If Sellars's account were true, then we would not talk the way we do about our own mental states. This problem confronts any otherwise plausible version of the theory theory.

Part I. Sellars's Myth

The overall purpose of Sellars's essay was to move philosophers from Ryle's alternative to Cartesianism to a more plausible alternative. More precisely: from the prevailing behaviorist understanding of mental language, which denied that mental events were privately accessible episodes that caused behavior, to a broader empiricism that captured what was valid in the classical Cartesian view: the intuitions that mental language does indeed refer to events or episodes that cause behavior, that these events are in some sense "inner," and that we have privileged access to them. What could not be salvaged from Cartesianism were the ideas that this access is by inner perception and that it yields infallible knowledge. Those ideas, which belonged to the Myth of the Given, were to be overthrown.

To achieve this aim, Sellars proposes to replace the Rylean dispositional reconstruction of mental language with a theory-like reconstruction. Mental language does indeed refer to unobserved inner episodes - but unobserved and inner only in the way that the submolecular particles and events of physical theory are unobserved and inner - that is, theoretical, and therefore not definable in observational terms. He proposes this, not as an accurate analysis of actual mental language, but merely as yielding new insight into certain philosophical problems.

To exhibit the strengths of the theory account, Sellars invented a developmental myth: one that portrays our distant ancestors as initially talking about people in the way Ryle represented us as talking, then learning to talk about inner episodes and states as theoretical causes of behavior. Sellars's early humans already have a language, but one that enables them to speak only of "public properties of public objects" (309). He supposes this public language to include whatever logical resources, such as the subjunctive conditional, would be needed for dispositional concepts. What the language would not equip its native speakers to do was to talk about an inner mental life.

Sellars assumes that, as a consequence of this deficiency, the speakers would have had a greatly impoverished capacity to explain human behavior. His student, Paul Churchland, describes the limitations such a language would impose:

Their explanatory resources for explaining human behaviour are limited to a few purely dispositional terms, all of which can be operationally defined (like 'is soluble')... For this reason, Sellars refers to these people, pejoratively, as 'our Rylean ancestors' ... They can explain some human behaviours, but only very few.⁽³⁾

Because Sellars thought these people so compromised, he supposed them to be in need of someone to teach them how human speech and action could be causally explained. That innovator, whom he dubs Jones, speculates that overt utterances and "intelligent" nonverbal behavior are caused by certain inner episodes -- episodes that resembled utterances in having semantic properties such as reference. These episodes, which he calls "thoughts," were supposed to be actual occurrences that cause behavior, not mere dispositions to behavior. They were hidden in the way theoretical properties are, yet grounded in public experience, like theoretical entities in the physical sciences, and eventually open to confirmation or disconfirmation. I will return later to the more interesting elaborations of Jones's theory.

Part II. My own visit

Now I offer a brief myth of my own. I too paid a visit to a prehistoric human cohort with a vocabulary limited to public properties of public objects, and with no conception of an inner mental life. Despite this deficiency, I was surprised to find that their capacity to explain human behavior did not seem greatly diminished.

I accompanied them in their daily activities and made something of a pest of myself, constantly testing their ability to explain behavior. Here is an excerpt from my journal:

We are tracking across the savanna, lots of excited talk. After ten minutes, the leader makes a sharp turn, and we all follow. I ask one of them, Why did she turn this way? My Rylean does not seem at all puzzled by the question. He replies: This is the way to the mounds. Testing further, I ask, But why is she leading you to the mounds? The answer: They are the termite mounds! I persist: What is special about termite mounds? A few others overhear me. They turn around in surprise. That is where we can get termites! For you, our honored guest! Rudely I ask: What is so special about termites? They look at one another and break out in general laughter. The leader walks over to me with a solicitous smile and says, It is the way they taste, of course. Nothing in the whole wide world - I include the forest and yonder mountain as well as the savanna - nothing has a more pleasant taste! That is why we are laughing.

Indeed, their explanations resembled those we give most of the time. True, we might say "She turned this way because she thinks it is the way to the mounds." We do this under special conditions, for example, if we ourselves don't agree that it is the way. I will not take the time to develop the point, but where the choice is between "because p" and "because she thinks (or: believes) that p," it is the former, the factive form, that is the standard or default form of explanation. It is the form that is used unless one has some reason not to use it. The relatively noncommittal "because she thinks" form accordingly carries a conversational implicature to the effect that one does have some reason not to make the stronger commitment implicit in the factive form.

We can easily account for the difference between Sellars's Ryleans and my Outlookers. Sellars's rhetorical purpose was to show how philosophers of a Rylean persuasion can with good reason speak of inner episodes while preserving their self-respect as empiricists. That is why his Rylean ancestors carry a much heavier burden than the restriction to public language. They are restricted to a much more austere idiom, which eschews not only causal explanations of human action in terms of mental states and episodes, but also causal explanations of human action in terms of reasons like, "it is the way to the mounds," which make no mention of mental states and events. My Outlookers have a much lighter load to carry.

Sellars's effort to move from behaviorism to a theory that acknowledges the role of the mental as in some respect an "inner" cause of behavior certainly represented a major step in the right direction. It was a serious attempt to make respectable some of Descartes' better intuitions. Other notable efforts were made in the 1960's to answer Wittgenstein and Ryle with a theory theory that spoke of inner causes of behavior: for example, by Putnam and by Chihara and Fodor. But, focusing, as Wittgenstein and Ryle had, on philosophically problematical language, these critics neglected explanations couched in unproblematical terms, such as, "She turned this way because it is the way to the mounds." The upshot is that they wrote as if psychological competence - the capacity for anticipation, explanation, and social coordination of behavior - begins and ends with the possession and use of mental concepts and the principles that link them. The tendency in the years since the publication of Sellars's essay has been to narrow the focus even further to the concepts of belief and desire. Many use the term belief-desire psychology to name the source of human psychological competence.

The success of the Outlookers is bound to be fragile. We have to imagine them to be a small, homogeneous group without dissension, or we will find them frequently perplexed by one another. They are not replicas of one another. If one of them misses a meal and is ravenous, the others who are not hungry at the time may have no way of understanding his behavior. And if one is privy to information another lacks, the two may find one another's behavior beyond explanation. That is why we, at least, find it useful to speak also, when the need arises, of the mental causes of action.

But it is important to appreciate that mental causes are in general a second best, invoked when there is reason not to locate the explanans out in the world. Not only the direction of the mounds, but also the taste of termites, and the pleasantness of that taste, resides in the world, unless one must for some special reason locate the cause in the agent. The simulation theory then goes further and says that there is no radical departure from this focus on the world even when we do have to introduce mental causes such as thoughts. As Jane Heal says, the capacity to think about the thoughts of others is just an extension of the capacity to think about the objects of these thoughts, their subject matter, "together with some extra sophistication." That is, "the capacity to think about thoughts must be seen as an extension of the capacity to think about their objects."

For example: In the famous false belief test of Wimmer and Perner, you see Maxi put his chocolate away in hiding place A. While he is outside, someone transfers the chocolate to hiding place B. Where will Maxi look for his chocolate? If you are simulating, you don't shift from thinking about where the chocolate is to thinking about that other, more complex object, Maxi. You think about the chocolate, but in a special way. Locating yourself mentally with Maxi while he's out of doors, you blank out what's going on inside. Then you come back in with him and look for the chocolate--but without benefit of the crucial information about the move from A to B. According to the simulation theory, then, even when dealing with the problem cases where the explanans is not out there in the world, we go on explaining or predicting in terms of facts, but at the 4-year-old stage we can mentally relocate ourselves and adjust the facts accordingly.

For an analogy, think of those little binocular telescopes we sometimes see at tourist sites. People line up to put in their coin and look into the scope. You put your coin in, look into the eyepiece, and if all goes right, the shutter opens and, lo and behold, there's the other rim of the Grand Canyon, or the fumaroles of Kilauea. So too, for the simulation theory, at least as Heal and I understand it, to find out what goes on "inside" a person (in the relevant sense), we look in, and, if all goes right, we see, in some cases, the very scene before us, though from a different physical perspective, and possibly with changes in the emotional and motivational colorations of things, and possibly serious mutations and transformations of the very things themselves.

This is in contrast to another way to understand looking into the telescope and seeing what is there inside it: taking it apart and examine it, or using some sort of non-invasive imaging, or speculating about the internal systems that allow it to do what it does - the internal structures and mechanisms that account for the optical properties of the telescope and its response to an inserted coin.

What the telescope analogy is meant to suggest among other things is that when you look inside people psychologically, what you see there is nothing but the world -- the world with its emotional and motivational qualities, its affordances, and its various modal properties. The point, I think, is similar to Hume's, about entering most intimately into what he calls himself, always stumbling on some particular perception or other, and never observing any thing but the perception. This brings me to the topic of how one looks into oneself and says what's in there. The answer I want to defend will be, of course, that one does this by looking out to the world and talking about what's out there.

Part III. Self-ascription

Theory theories have had particular trouble with accounting for the apparent differences between third person and first person ascription: the third person dependence on situational and behavioral evidence and the first person independence of evidence, together with the epistemologically privileged character of first person ascription. The extreme position sometimes taken is that there are no real differences. One simply applies the theory to oneself on the basis of situational and behavioral evidence. It is just that the range and amount of evidence available in one's own case is typically far greater than what is available to others. Further, it has been argued that our expertise concerning our own psychological states, based on past applications of folk theory to ourselves, may lead to the illusion that our knowledge of our own states is "perceptually immediate, noninferential, direct."⁽⁴⁾

Other theory theorists, perhaps most notably Sellars, have found this implausible and have searched for an alternative that grants at least some degree of first person authority. The main point of his myth, Sellars tells us, is that we need not give up the privacy of inner episodes such as thoughts in order to acknowledge the essential intersubjectivity of language and language-learning. For

the reporting role of these concepts--the fact that each of us has a privileged access to his thoughts--... is built on and presupposes this intersubjective status.

Sellars's novel idea was that privileged access to the truth about one's own thoughts does not require an experience or awareness of one's thoughts. What alternative is possible? Sellars describes the process in terms of operant conditioning. Once the Ryleans have mastered Jones's theory and can apply it to others, they are so conditioned that, when situational and behavioral evidence indicate certain theoretical states and episodes, they are disposed to make the corresponding self-ascriptions, using the language of the theory, without having to consider the evidence. In Sellars's story,

Jones brings this about, roughly, by applauding utterances by Dick of "I am thinking that p" when the behavioral evidence strongly supports the theoretical statement "Dick is thinking that p"; and by frowning on utterances of "I am thinking that p", when the evidence does not support this theoretical statement. Our ancestors begin to speak of the privileged access each of us has to his own thoughts. What began as a language with a purely theoretical use has gained a reporting role.

Once trained, Dick should not have to make theoretical inferences from his own behavior. If the theory is right and if the trainer applies it correctly, the inner state that caused the behavior that was reinforced will indeed be one that satisfies the functional role assigned to "thinking that p" in Jones' theory. And in the future that state will cause Dick's self-ascription, "I am thinking that p" - provided, of course, that Dick wants to report what he is thinking.

A great virtue of this account is that it shows how mental predicates may have exactly the same sense whether ascribed by another on the basis of evidence or self-ascribed by the subject independently of evidence. The predicate is univocal simply because the evidence-based third person sense is the only sense it has. The first person use makes no independent contribution to meaning.

I am much in sympathy with Sellars's idea that there is privileged access to the truth about one's own thoughts and that it does not require an experience or awareness of one's thoughts. It is certainly more plausible phenomenologically than the view that, like everyone else, we have to examine our behavior, even if we can do it in a "perceptually immediate" way. I also find it better than the views of those simulationists who bank on an inner sense dedicated to the perception of mental processes. Yet Sellars's account will not do. If something like it were true, then we would not talk the way we do about our own mental states. The problem is easily illustrated for the ascription of beliefs and thoughts. Consider the following sentences:

It is raining, but I don't believe it is.

I believe it is raining, but it is not.

As G.E. Moore pointed out, to assert either of these conjunctions would be self-stultifying, as the asserting it conversationally defeats what is being asserted. Yet neither sentence is formally inconsistent. In fact, there have been occasions on which I wasn't aware that it was raining and also occasions on which I mistakenly believed it was raining. This is Moore's paradox: Despite the formal consistency of what is uttered, the utterances have what has come to be called Moorean or pragmatic inconsistency.

The same problem occurs with reports that one is currently thinking that p, as distinguished from reports of a standing belief that p; thus, with reports of what Ryle and Sellars call episodes rather than states. Consider, "It is raining, but I am not thinking it is," and, "I am thinking that it is raining, but it is not." Unless "thinking" is merely entertaining a thought, you will hear these as pragmatically inconsistent.

Now suppose you have received Jones's theory-based training in the use of "I believe" in self-reports. Now, you use the "I believe that p" form only when the behavioral evidence strongly supports the theoretical statement that you have the belief in question. Consider, , "It is raining, but I don't believe it is." The first clause is a weather report, the second clause is a self-report. If we accept Sellars's account, there should be no reason to expect the two always to correspond. The weather report requires only training in the public language. The self-report, however, is guided by the special theory-based training you received for giving such reports. Such training is based on your behavior, verbal and non-verbal. Part of the behavioral evidence, namely your asserting, "It is raining," may be outweighed by other behavioral evidence that strongly indicates you do not believe that it is raining. So there are liable to be occasions on which you might justifiably and without inconsistency say, for example, "It is raining, but I don't believe it is." Likewise, there should be no appearance of inconsistency if you say, "I believe it is raining, but it is not." For the two clauses concern two quite different topics. As I said earlier, if something like Sellars's account were true, then we would not talk the way we do about our own mental states.

This is not just a problem for Sellars's account of how a theoretically based self-ascription is possible. The same problem exists for Gopnik's view that we examine our own behavior in the light of the theory, and accordingly report that we believe or are thinking, say, that it is raining. Rather than canvass other possible variants of a theory theory account, I will shortly offer an argument that I believe covers any variant that any philosopher or psychologist would seriously want to put forward.

First I will illustrate what I mean by coordinating ascription with expression. Here is a question I invite you take a moment to answer:

Is the following something you believe? "It is raining outside."

I doubt that in answering the question you examined your recent behavior in the light of a theory. And I doubt that you introspectively searched for a telltale feeling or other experiential mark of belief. You probably just reinterpreted the question as, "Is it raining outside?" Maybe, like me, to answer the question you try to recall your last glimpse of the outside world a few minutes or hours ago. If your answer to this meteorological question was affirmative, then you were prepared to say, "Yes, I do believe that it is raining outside." If your answer was negative, then you were prepared to say, "No, I don't believe that it is raining outside," or more strongly, "No, I believe that it is not raining outside." If you had no answer to the lower-order question, then you might have said, "No, I don't believe one way or the other, I just don't know." (There appear to be constraints that rule out, among other things, taking excessive time to retrieve or derive the answer, or actually taking out that map again.)

In a couple of recent papers⁽⁵⁾ I argued that the self-ascription of mental states, more precisely, any self-ascription that is not based on evidence, is accomplished by the use of what I call ascent routines. These are procedures - routines requiring no reasoning - that allow one to answer a question about oneself, and specifically about one's mental states, by answering a question that is not about oneself, nor about mental states at all: an outward-looking question. They answer a question about a mental state that is about x by answering another question pitched at a lower semantic level, a question directly about x. That is why I call them ascent routines. The general idea is, roughly, that to ascribe a belief, desire, intention, or other mental state to ourselves in a reasonably reliable way, all we need is the ability to express the belief, desire, intention, and so forth. For example, when I tell the clerk, "I want two of these and three of those," it is very likely that I do want possession of the items I point to. To attain reliability with regard to what I want, I don't look inward, I look outward at the display case.

In the two papers in which I discuss ascent routines, I described the ascent routines for just a few mental states, perhaps most convincingly for belief. So I am not certain that for every mental state there is an ascent routine, and that the use of this routine underlies all self-ascriptions of that state that are not evidence-based. But there is reason for optimism if you consider the following to be pragmatically inconsistent:

That is frightening! But I am not at all frightened by it.

That is absolutely delightful! But I am not at all pleased by it.

My toe hurts! But I am in no pain.

The taste of termites is just fabulous! But I don't like the way they taste.

In each case, the self-ascription must be so guided as to avoid the possibility of any of these utterances - except in a special disclaimer context like the one in which I am mentioning them.

The general idea is that we coordinate one type of verbal behavior, self-reports of a mental state or episode, with another, the outward-looking "expression" of the state or episode. A self-report should give others a glimpse into a subject looking out onto the world. Hence it is tied to speaking about the world, including speaking expressively about the world, finding something scary, or finding a part of the body painful.

The only way for a theory theorist to avoid Moore's paradox is to concede that our self-ascriptions, at least regarding what we believe or are thinking, are coordinated in the way I suggest with verbal expressions of the state - -- the outward-looking "expression" of a belief or thought in the public non-mental language⁽⁶⁾- neglecting all nonverbal evidence and any other verbal evidence.

Of course, this accommodation to Moore's paradox does not come cheap. It leaves the theory theorist with a dilemma. If on the one hand she foregoes the theory in the case of first person ascriptions but does not forego the theory in the case of third person ascriptions, she gives up the univocality of first person and third person. The meaning of "x believes it is raining" will differ between first person and third person uses - a problem Sellars's proposal had avoided by assuming that there is only a third person theoretical meaning. If on the other hand she decides to forego the theory for the third person as well as the first person, then of course she gives up the theory theory.

Suppose we accept my conclusion that self-ascription is coordinated just with self-expression, and not, as the various theory theory accounts would have it, also with nonverbal evidence and verbal evidence other than expression. How can we also allow that third person ascription is and should be based on all of these kinds of behavior? Wouldn't we be giving up univocality?

Not at all, if my version of the simulation theory is right. According to my version, it does not matter whether we are ascribing mental states to ourselves or to another: In either case we answer the question about belief by directing our attention to the subject matter of the belief: meteorological conditions, in the case of the belief that it is raining. Whether in the role of another or not, we find out whether the subject has a belief that p simply by surveying the world to determine if it is the case that p, and then we turn the answer to this over to an ascent routine. The ascent routine insures that the coordination is simply with the verbal expression of belief, and not with any other sort of behavior. And yet, if I am making the ascription within the scope of a simulation -- say, I see someone about to step out the door carrying an umbrella over his head -- I take this nonverbal evidence very much into consideration when I first simulate him. There is a two-step process in ascribing a mental state to another: First simulate, using any available evidence; then self-ascribe, using an ascent routine that connects only with verbal expression.

I agree with Sellars that even the first person use of mental predicates is essentially an intersubjective achievement. But Sellars locates the intersubjectivity in the wrong place. We don't need other people to tell us what we think, as in the old behaviorist joke, "You are fine, how am I?" Self-reports borrow from the intersubjectivity of the public language itself, the language of "public properties of public objects." What is essential to regulating the first person use of believe and think, as in "I believe it is raining," is intersubjective training in making assertions about the world, such as, "It is raining." To regulate our first person reports, we do not require any new intersubjective training - none, that is, beyond training in the appropriate ascent routines, which teach us how to piggyback on the public language.⁽⁷⁾

I said that my argument covers any version of the theory theory likely to be put forward. That is because no one would want to endorse a theory theory that denies the relevance of nonverbal behavior. But if the theory theorist allows that nonverbal behavior is relevant to ascription, and thus that first person ascription is not uniquely tied to the verbal expression of mental states, then she invites utterances like,

It is raining, but I don't believe it is.

That is scary! But I am not at all scared by it.

The taste of termites is fabulous! But I don't like the taste.

Thus, were any version of the theory theory that admits the relevance of nonverbal behavior true, we would talk about our own mental states in a way that in fact we do not.

It is generally agreed that the issues between the simulation theory and the theory theory are fundamentally empirical and therefore to be settled on empirical grounds. The usual assumption, however, is that the relevant empirical investigations lie within the domains of psychology, particularly developmental and cognitive psychology, and neuroscience. I interpret "empirical" more broadly. Part of the needed empirical work, perhaps even the decisive part, is "armchair" testing of a sort that calls more on the skills of philosophers than on those of experimentalists: the testing of competing views against philosophical touchstones such as intentionality, referential opacity, the peculiarities of self-awareness, the apparent asymmetry of first person and third person attribution, and Moore's paradox. These are features that have been thought to set mental state concepts off from nonmental concepts, and thus to constitute test cases for resolution of the mind-body problem. They are also test cases for resolving the question of what our psychological competence consists in. An adequate account of psychological competence must at least be consistent with the existence of these phenomena; an ideal account will predict the existence of these phenomena. What I have argued is that no plausible version of the theory theory is consistent with, much less predicts, Moore's paradox and, more generally, the tight linkage between our "inward-looking" self-ascriptions and our "outward-looking" reports about the world.

Notes

1. In The Foundations of Science and the Concepts of Psychoanalysis, Minnesota Studies in the Philosophy of Science, Vol. I, ed. by H. Feigl and M. Scriven (Minneapolis: University of Minnesota Press, 1956), pp. 253-329. Reprinted in Sellars, W., Empiricism and the Philosophy of Mind, with an introduction by Richard Rorty, and a study guide by Robert Brandom (Cambridge, Mass: Harvard University Press, 1997), 13-118. Page references are to the original publication.

2. It was Brad Cohen who persuaded me that Sellars did not subscribe to these views himself.

3. In S. Guttenplan, ed., A Companion to the Philosophy of Mind, Oxford: Blackwell, 1994), p. 308.

4. Alison Gopnik, "How we know our own minds: The illusion of first-person knowledge of intentionality," Behavioral and Brain Sciences 16:1 (March 1993), Pages 1-14.

5. "Simulation Without Introspection or Inference From Me to You," in M. Davies & T. Stone, eds., Mental Simulation: Evaluations and Applications (Oxford: Blackwell, 1995); and, "'Radical' Simulationism," in P. Carruthers and P. Smith, eds., Theories of Theories of Mind (Cambridge University Press, 1996)

6. I am leaving aside the special case in which the content of the belief concerns mental states.

7. One might think this is true only of propositional attitudes, and not, for example, of self-reports of sensations and feelings. One might think there must be special intersubjective training for, "I am in pain," or "I have a toothache." Is not it when the child puts her hand on her cheek and winces and cries that we say, "You have a toothache"? Yes, indeed, but this need only piggyback on pain expression, "That hurts!" or, "Tooth hurts!"