Philosophers influenced by W.V. Quine or by continental hermeneuticists maintain that what a subject believes or desires is entirely a matter of how that person is interpreted or translated into someone else’s preferred idiom for one purpose or another, there being no antecedent or inner fact of the matter.  A distinctive though slightly weaker version of this view is that of Dennett.

Dennett’s instrumentalism

    The Identity Theorists and the Functionalists (Machine or Teleological) joined common sense and current cognitive psychology in understanding mental states and events both as internal to human subjects and as causes.  Beliefs and desires in particular are thought to be caused by perceptual or other cognitive events, and as in turn conspiring from within to cause behavior.  If Armstrong’s or Lewis’ theory of mind is correct, this idea is not only commonsensical but a conceptual truth; if Functionalism is correct, it is at least a metaphysical fact.
    In rallying to the inner-causal story, of course the Identity Theorists and Functionalists broke with the Behaviorists, for Behaviorists did not think of mental items as entities, as inner, or as causes.  Behaviorists paraphrased mental ascriptions in terms of putative responses to hypothetical stimuli.  More recently (though under the direct influence of Ryle), Dennett denies that beliefs and desires are causally active inner states of people, and maintains instead that belief- and desire-ascriptions are merely calculational devices, that happen to have predictive usefulness.  Such ascriptions are often objectively true, but not in virtue of describing inner mechanisms.
    Thus Dennett is an instrumentalist about propositional attitudes such as belief and desire.  (An “instrumentalist” about Xs is a theorist who claims that although sentences about ‘Xs’ are often true, they do not really describe entities of a special kind, but only serve to systematize more familiar phenomena.  E.g., we are all instrumentalists about “the average American homeowner,” who has exactly 2.3 children.)  To ascribe a “belief” or a “desire” is not to describe some segment of physical reality, Dennett says, but is more like moving a group of beads in an abacus or doing vector sums by parallelogram of forces in kinematics.

    There are three “stances” from which we can predict the behavior of some behaver—device, creature or human being:  The physical stance (physics or other low-level science), the design stance (program or otherwise functional), and the intentional stance.  Behavior prediction from the intentional stance is most fundamentally a matter of extrapolating rationally from what a subject ought to believe and ought to want in his/her circumstances; we then presume that the subject does believe and want those things, and predict the appropriate behavior.  This is exactly what we do with laptops and pocket calculators, and even with thermostats (when we don’t know how the thermostat works).  And this epistemological strategy works astoundingly well.

    A thing S is a believer, Dennett says, just in case S’s behavior is reliably predictable from the intentional stance, and the intentional stance is indispensable.  (The indispensability clause is required to rule out the lectern example).  S believes that P just in case to attribute that particular belief to S would result in good behavioral predictions.  Notice that Dennett does not assume that anyone actually does do any attributing.

    His case for this view:

    1.  He argues from the foregoing epistemology of the intentional stance, i.e., of belief- and desire-ascriptions.  Viz., he boldly just identifies the truth-makers of those ascriptions with their verification-conditions; why should we think that what constitutes a propositional attitude should be anything over and above the evidence we have for it?  He challenges inner-cause theorists to show why his instrumentalism does not accommodate all the actual evidence.  Reply:  This is just steam-age, Positivist verificationism.  We should just reject any inference from the epistemology of a thing to the metaphysical nature of that thing.  Rejoinder: (See next argument.)

    2.  The epistemology of the intentional stance contains ineliminably normative elements: “ought to believe,” “ought to want,” and the assumption of rationality.  Dennett argues that if purely normative assumptions are required for the licensing of an ascription, then the ascription cannot itself be a purely factual description of a plain state of affairs.  The epistemology of inner causes is well known and standard, and it does not include normative assumptions.

    3.  Dennett’s instrumentalism requires no psychosemantics, and so avoids the biggest obstacle to Representationalism.  (Dennett adds more arguments against the “language of thought,” two of which we have already considered.  Most of the others rely on caricatures.)
    But he also thinks it quite unlikely that any science will ever turn up any distinctive inner-causal mechanism that would be shared by all the possible subjects that had a particular belief.  I agree, though it hardly follows that there is no such mechanism.

Objections

    1.  Dennett’s epistemology is too liberal.  Even if pocket calculators have “beliefs” and “desires,” thermostats and lightning rods simply do not.  For that matter, Dennett’s view firmly applies to the Tinfoil Man, and counts him as a believer.  But “he”’s only a piece of Tinfoil!
    Thus, Dennett’s instrumentalism is not too far from Ryle’s Behaviorism; Objection 3 to Behaviorism applies.  Does Objection 2 (the introspecting of mental episodes)?  Not clear; it’s not obvious that we can introspect beliefs or desires.  We can introspect judgements, but as before Dennett seems to admit that.

    2.  (Johnathon)  Never mind such degenerate examples; there are cases in which it’s true of some human being that, were one to ascribe to that person the belief that P, good predictions would ensue, yet the person does not believe that P.  E.g., the actor in mid-performance, or a spy who for decades plays a role in an enemy nation.  The Rylean rebuts this objection by pointing out that actors and other pretenders have dispositions that ordinary people don’t, such as the disposition to drop the pretend-behavior when the pretense is no longer needed.  But Dennett does not put his view in terms of dispositions, so it is not clear how he could reply.

    3.  Dennett’s formulation, “A person S believes that P iff, were one to ascribe the belief that P to S, one would get good behavioral predictions,” is circular!  It uses the expression “belief that P” in what is supposed to be an explication of that very concept. Replies:  First, Dennett is not defining the term “belief that P,” but only offering a metaphysical hypothesis as to what believing really is; so he is not offering a circular definition as in “Jejune, adj.: Said of things that are jejune.”  Second, he does really mean that ascriptions-of-belief (hyphenated) are conceptually prior to so-called “beliefs”; beliefs are second-class items, constructed out of the more concrete, more real ascribings.  If you want to know what an ascription-of-belief is, look to our practices in the intentional strategy.

    4.  In cognitive psychology, beliefs and desires interact closely with the internal representations posited by the psychologist.  Commonsense beliefs and desires are referred to in experiments, pretty much interchangeably with the internal representations.  In abnormal psychology, design-stance considerations are brought to bear on the patient’s neurotic beliefs and desires.  And as Dennett has sometimes admitted, commonsense belief-desire descriptions of people and cognitive psychology sometimes augment or correct each other.  How is any of this possible, if belief and desire ascriptions are purely instrumental while the cognitive representations are real inner states of the subjects?  And why would the two only polysemously intentional phenomena have anything to do with each other?
 

Methodological solipsism

    Several different computers could be printing out the same numbers in response to the same keystrokes, yet computing different things.  Putnam, Fodor and Stich argue that much the same is true of human beings, even though our intentionality is original rather than derived:  Surprisingly, ordinary propositional attitude contents do not seem to be determined by the states of their subjects’ nervous systems, not even by the total state of their subjects’ entire bodies.  Fodor’s “Yon” (= Putnam’s “Twin Earth”) and indexical examples (“I am overpaid”) and beliefs involving proper names (“George is underpaid”) and “New Earth” are widely taken to show that, surprising as it may seem, two human beings could be molecule-for-molecule alike and still differ in their beliefs and desires, depending on various factors in their spatial and historical environments.   This is called externalism about intentional contents.
    Thus we can distinguish between “narrow” properties, those that are determined by a subject’s intrinsic physical composition, and “wide” properties, those that are not so determined, and representational contents are wide.  So it seems an adequate psychosemantics cannot limit its resources to narrow properties such as internal functional or computational roles; it must specify some scientifically accessible relations between brain and environment.
    Fodor and some other theorists continue to maintain that there are narrow contents in addition to, and perhaps underlying, the wide ones.  For example, although WGL believes that water covers 67% of the earth and Twin WGL believes something different, that Twin-water (XYZ) covers 67% of the (Twin) earth, each believes that the familiar clear odorless tasteless liquid covers 67% of the planet he inhabits.
    (Quick quiz:  That last bit was supposed to be an example of a narrow content that WGL and Twin WGL would share.  Is it really an example of that?)

    You may or may not be convinced that the Putnam-Fodor-Stich thesis is right.  But if it is right, it is highly consequential.  Here are some morals that people have drawn from it.

    Methodological solipsism for scientific psychology.  Psychologists are supposed to be studying the causes of behavior.  But molecular duplicates behave in the same way, as a result of the same causes.  That my belief is about water (H2O) while Twin WGL’s belief is about a different stuff (XYZ) is causally irrelevant; we behave the same, because the causes of our behavior are in the head.  Accordingly, psychology should appeal only to what’s in the head, i.e., to narrow properties of subjects and not their wide properties.
    (In his comments on Fodor, Stich warns that there may be a lot fewer narrow contents than Fodor thinks, indeed very few for the psychologists to work with at all.  What Stich really believes, though he doesn’t come right out and say it here, is that psychology should get entirely out of the content business.)

    Functionalism as we have been understanding it is untenable.  Till now we have conceived functional roles narrowly; you and your Twin would have exactly the same functions being realized in you from moment to moment.  Yet you believe and want different things; hence, believing and wanting are not functional states.
    The Functionalist has two possible moves here.  One is to restrict the thesis:  Function is what makes a belief a belief, but some other account must be given of its intentional content.  (We need a psychosemantics in any case.)  The other move is to fashion a Wide Functionalism, and define functions themselves in terms of environmental entities rather than abstractly characterized body parts.

    We think of propositional attitudes as having their characteristic effects in virtue of their contents:  people behave as they do because of what they believe, what they want, what they plan, etc.   But if those contents are wide rather than narrow, that commonsense way of thinking is wrong.  You and your Twin will always behave in just the same way, despite having different beliefs and desires.  Behavior is caused by what’s in the head, not by the relations your head contents bear to things in the environment.
    Luca had an excellent answer to this.  In the spirit of Wide Functionalism, he pointed out that behavior can be characterized widely too (George bangs on the Dean’s door, while Twin George bangs on his (Twin) Dean’s door—different doors, you see, so different behavior despite sameness of bodily motion).

    There is a problem about our knowledge of our own intentional states.  We think we have secure first-person knowledge of what we believe.   Our self-knowledge is not infallible, but it is privileged; you have a kind of direct access to your belief contents, in that in order to know what they are you don’t have to examine any part of the external world.  But if those contents are constituted in part by things and stuffs in your external environment, how is that possible??

Searle vs. computationalism

    “Strong AI” is the thesis that I can understand what someone says, or be in some other intentional state, just in virtue of performing formal computations over uninterpreted “symbols,” really just syntactic shapes.  Searle attacks Strong AI by offering his Chinese Room counterexample (very like Block’s population-of-China alleged counterexample to Functionalism), in which there is formal computations over uninterpreted squiggles, but, Searle says, obviously no understanding or other intentionality.

    Objections considered in class:

    1.  (Johnathon; cf. the “systems reply”)  That no component of the Chinese Room setup has understanding or (relevant Chinese) intentionality does not show that the system as a whole doesn’t have it.  The system is, after all, running the program of a real Chinese understander.   Searle would make two replies (p. 512).  First, let the Searle-homunculus in the example “internalize” all the other elements of the system, so that he becomes the whole system.  He still understands no Chinese.  Second, “It is not easy for me to imagine how someone who was not in the grip of an ideology would find the idea at all plausible.”

    2.  (Kristin)  Once the Searle-homunculus got to be expert at the job, he might find that he did understand Chinese.  It would be like picking up a language by hearing it spoken all around you and by practicing, trial and error.  (But this is hard to imagine, given that Searle would not be out in the open interacting with people and (especially) common physical objects.  He can’t see anything outside the room.)

    3.  Fodor agrees, of course, that Strong AI is false.  Any externalist must (even though, ironically, Searle is one of our few anti-externalists).  We already know that a wide psychosemantics is needed to give content to the representational states inside our heads.  Take whatever will prove to be the correct psychosemantics  say, one that appeals to causal-historical chains that connect brain states to environmental things and stuffs.  Then just apply it to the Chinese Room.  If it does apply, it shows that the system does after all have intentional states.  If it doesn’t, that would show why Searle is right and the system can’t have intentional states.  Either way, there’s no issue left.
    Searle rejoins (pp. 522-23):  Egg foo yung!  Connecting a squiggle inside the room causally-historically with a bowl of egg foo yung outside would not help, because it would not enable the Searle-homunculus to interpret the squiggle—he can’t see outside the room.  At the least, he would need to be aware of the causal-historical chain.
    Fodor would say that the Searle-homunculus doesn’t have to know the squiggle’s meaning, any more than one of your neurons needs to know what a brain state is representing; he’s only a functionary.  What matters is whether the whole system is connected to egg foo yung in the right way, and as before, either it is or it isn’t.  But Searle will rejoin that if the whole Chinese Room didn’t understand Chinese or have other intentional states before the causal-historical chain was added, the addition of the chain does nothing to help.  Stalemate.

Psychosemantics

    Dretske bases his nascent psychosemantics on the notion of “information” that figures in the Shannon-Weaver “Mathematical Theory of Communication.”  A state of a device carries the “information” that P iff by law of nature, the device could not be in that state unless P.  Of course, this sort of relationship is ubiquitous in the universe, because there are lots of laws of nature and scads of nomic relationships.  So Dretske’s task is to say what must be added to “information” to get a genuine cognitive intentional content.
    He focuses on the most specific “information” carried by a state.  His second organism (p. 360) can know that it is touching acid without knowing that it is touching HCl because it has a repertoire of two different states, one of which carries the information that it is touching HCl (but nothing more specific) while the other carries only the information that it is touching acid.  He calls this a difference in two ways in which the creature can “code” the information that it is touching acid.
    That suggestion does not help much.  Two thermostats or galvanometers could have such pairs of states.  It follows from what Dretske says here that they would be genuine cognizers.
    But this is only the beginning of a long tale of woe.  The next problem is that which Dretske confesses on p. 361:  How is he to accommodate false intentional contents, such as those of false beliefs?  (If we were to say that a belief that P is even in part a state that carries the information that P, then the subject by definition could not be in the state unless it were true that P.)  Dretske gestures toward the organism’s learning history, but that idea never panned out.

    In later work, Dretske injected an element of teleology: A state carries the content that P iff it has the function of carrying the information that P.  That explains how a belief can be false; the belief is supposed to carry the information that P, but fails to.  (Examples: A malfunctioning gas gauge, a watch that’s fast.)  That’s helpful, but again is only a baby step.  Objection (Johnathon and Luca):  The new teleologized theory entails that every time I have a false belief, I’m broken or at least malfunctioning.  But some of my false beliefs are not the result of malfunction.  Sometimes I am in perfect working order and hold a belief that is perfectly well justified (indeed, I’d be malfunctioning if I didn’t hold it), yet the belief is false owing to a fluke.
    A worse objection:  How does the gas-gauge, galvanometer model apply to the human brain?  Are we to think that various individual brain states have as their biofunctions to carry information about the speed of light, about the Australian bushranger Ned Kelly, about God and my sister?   That is a weird suggestion.  Compare:  The biofunction of teeth is to pulverize food so that it can be swallowed.  If you like, the biofunction of eyes is to extract information from the immediate environment and to output a representation of what sorts of physical objects lie before one.  But how could it be a biofunction of a state deep inside cerebral cortex to indicate something about the speed of light or Ned Kelly?

    A further and much more serious obstacle to psychosemantics is that the objects of thought need not be in the environment at all.  They may be abstract; one can think about a number, or about an abstruse theological property.  And as always they may be entirely unreal.  (N.b., the same things are true of representations posited by cognitive psychology.)  An adequate psychosemantics must deal just as thoroughly with arithmetical beliefs, and beliefs in quantum field theory, and Arthur’s illiterate belief that the number of the Fates was six, and Hegel’s belief that the Absolute is in a constant process of self-realization.  I have no idea how present-day psychosemantics could be extended to deliver contents like those.  (Though Mark made a pretty good suggestion: that perhaps the abstract belief contents are somehow composed of smaller informational contents that are less far removed from the perceptual.)

    Finally (I did not get to this in class):  Dretske and other psychosemanticists focus almost entirely on belief.  (I say “almost” because in his 1988 book Dretske does also address desire.)  Of course belief is information-carrying in some sense; it represents the world as being a certain way, even if we don’t agree with Fodor that it is a kind of internal, computational representation, and it aims at correctness.  But what about other propositional attitudes whose function is not to be correct representations?  S wishes that P; S wonders whether P; S hopes that P; S fears that P.  It is not the function of any such state to carry the information that P.  So here is another dimension along which Dretske’s account will have to be extrapolated, and I don’t see bright prospects for that project.