FINAL DRAFT, for RORTY AND HIS CRITICS
July 18, 1996
Daniel C. Dennett
In the late 1960s, I created a joke dictionary of philosophers' names that circulated in samizdat form, picking up new entries as it went. The first few editions were on Ditto masters, in those pre-photocopy days. The 7th edition, entitled The Philosophical Lexicon, was the first properly copyrighted version, published for the benefit of the American Philosophical Association in 1978, and the 8th edition (brought out in 1987), is still available from the APA. I continue to receive submissions of further entries, but I doubt that there will ever be a 9th edition. The 8th edition lists two distinct entries for Dick Rorty:
rort, an incorrigible report, hence rorty, incorrigible.
a rortiori, adj., true for even more fashionable continental reasons.
These were submitted to me years apart, inspired by two distinct epochs of Rorty's work. It may be hard to see the connecting threads between the Princeton professor whose tightly argued "Incorrigibility as the Mark of the Mental" (1970) and "Functionalism, Machines, and Incorrigibility" (1972) were aimed specifically at the smallish clan of analytic philosophers of mind, and the international man of letters described by Harold Bloom as the most interesting philosopher in the world. Can we see the stirrings of Rorty's later ideas in between the lines of his early papers in the philosophy of mind? Perhaps, but that will not be my topic.
I want to go back to Rorty's papers on incorrigibility(1), not for historical clues about how to read the later, more widely influential Rorty, but in order to expose an excellent insight lurking in his claim that incorrigibility is the mark of the mental. It went unrecognized at the time, I think, because the reigning methodology in that brand of analytic philosophy ignored the sorts of questions that would have provoked the relevant discussion. While the incorrigibility papers were sufficiently influential--or at least notorious--to anchor an entry in the Lexicon, they have never been properly appreciated by philosophers of mind, myself included (of all people). I say "of all people" because Dick Rorty has always drawn explicit links between his ideas and mine, and has played a major role in drawing philosophers' attention to my work. If anybody was in a position to see the virtues of his position, it was I, and while I can now retrospectively see that I did indeed subliminally absorb his message and then re-invent some of it in my own terms (without sufficient acknowledgement), I certainly didn't proclaim my allegiance to, or even deign to rebut, clarify or expand upon, those claims.
If my take on this is right, it means that Dick also didn't quite appreciate the strengths of his own idea, and might even have been misled to some of his more fashionable and famous ideas by a slight misappreciation of the import of his claims about incorrigibility, but I won't pursue that surmise here. If I am right, he will have succeeded in spite of himself in making the sort of contribution to science--to our objective knowledge of the way the world, and the mind, is--that he has abjured as a philosophical aspiration. His own philosophical "conversation" turns out to be more than just conversation. He will perhaps reply that all I have shown is that today his ideas about incorrigibility have more political viability, more charismatic oomph in today's conversations than in those of the early 70s. But I want to insist that the reason they do is that they show us something interesting about how reality may be represented.
1. What is the Status of Rorty's Thesis?
His central thesis is as follows:
What makes an entity mental is not whether or not it is something that explains behavior, and what makes a property mental is not whether or not it is a property of a physical entity. The only thing that can make either an entity or a property mental is that certain reports of its existence or occurrence have the special status that is accorded to, e.g., reports of thoughts and sensations--the status of incorrigibility. (1970, p.414)
Incorrigibility is to be distinguished from infallibility. It is not that these reports could not possibly be mistaken, but just that "certain knowledge claims about them cannot be overridden." (p.413) This immediately tilts the playing field, of course, by trading in a host of tempting but indefensible metaphysical claims for an epistemological or even sociological claim. This is just a fact, Rorty suggests, about a "linguistic convention," about the way we treat claims, not a fact about the reality of whatever those claims are about. But at the same time his thesis is not a mere anthropological observation: certain claims cannot be overridden, he suggests, given the role they play in our shared life. (As we shall see, it is this modal claim that never got sufficient attention--from Rorty or his readers--back in the 70s.)
What goes without saying is that these incorrigible reports are "first person" reports, reports about one's own states and events, to which one is presumed to have one or another sort of "privileged access." This term of art, once so familiar in philosophical writing about the mind, has been eclipsed for some time by other ways of attempting to characterize the crucial asymmetry: Thomas Nagel's (1974) "what it is like" formula or John Searle's (1980, 1983) championing of first person primacy, for instance. It is easy to understand Rorty's lack of sympathy for these later attempts. Far from having overlooked or underestimated the importance of the "first person point of view," he had declared it "the mark of the mental"--but he had also provided a demystifying account of how and why it had emerged, and why it was no bulwark against creeping "third person" materialism. Privileged access is real enough, Rorty was saying, and is indeed the premier feature of mentality, but it is no big deal, metaphysically. It is this deflationary doctrine that I want to re-examine, saving it from some Rortian excesses.
His claim may be expressed with somewhat different emphasis: what makes an entity a "first person," a thing it is like something to be, is that some of its emissions or actions are treated not just as reports, but as incorrigible reports. We vouchsafe an entity a mind by vouchsafing it a certain epistemic privilege with regard to the covert goings-on that control it. A mind is a control system whose self-reports cannot be overridden by third-person evidence.
Could there even be such a control system? One of Rorty's shrewdest observations is that our underlying materialist skepticism about this very possibility is the chief factor that propels us towards dualism and other mysterious doctrines:
Only after the emergence of the convention, the linguistic practice, which dictates that first-person contemporaneous reports of such states are the last word on their existence and features, do we have a notion of the mental as incompatible with the physical (and thus a way of making sense of such positions as parallelism and epiphenomenalism). For only this practice gives us a rationale for saying that thoughts and sensations must be sui generis--the rationale being that any proposed entity with which they could be identified would be such that reports about its features were capable of being overruled by further inquiry. (1970, p.414)
It does seem at first blush as if the states and events in any material or physical control system would have to be exactly as knowable by "third persons" as by their "owner," and if this is so, then no such states and events could be thoughts and sensations. It seems to follow that in any purely material entity, first-person privilege would evaporate, at which point there would be nothing left to anchor the mental at all. Rorty does not shrink from this implication: in fact, he views his 1970 paper as explicitly arguing for a version of eliminative materialism (p.401). He has his materialist concede that
it might turn out that there are no entities about which we are incorrigible, nearly or strictly. This discovery would be made if the use of cerebroscopes (or some similar mechanism) led to a practice of overriding reports about mental entities on the basis of knowledge of brain states. If we should, as a result of correlations between neurological and mental states, begin taking a discovery of a neurological state as better evidence about a subject's mental state than his own report, mental states would lose their incorrigible status and, thus, their status as mental. (p.421).
He contemplates with equanimity the Churchlandish alternative:
If it came to pass that people found that they could explain behavior at least as well by reference to brain states as by reference to beliefs, desires, thoughts, and sensations, then reference to the latter might simply disappear from the language. (p.421).
Here we need to pause and disentangle some issues, for there are apparently more possibilities than Rorty discusses. First, as just noted, there's standard eliminative materialism: the triumph of neuroscience and its "cerebroscopes" would--and should--lead to the demise of mentalistic language, and we would all cease talking as if there were minds and mental events, a clear improvement in our conceptual scheme from the perspective of Occam's Razor. But there is another prospect: it could also happen, for all Rorty has said, that people mistakenly crown neuroscience the victor, overrating the reliability of third-person theory and abandoning their linguistic practice, coming eventually to treat subjects' self-reports as unprivileged, even though they were in fact reliable enough to sustain (to justify?) the linguistic convention that mentality depends on. This would be the evaporation of the concept of mind from that culture, on Rorty's analysis, but would it also mark the death of the minds themselves? Although people's brains, their hardware, would be up to the task, their attitudes towards their own authority would shift, thereby adjusting the software running on that hardware. Could this diminish their real competence, leading to the loss of the very prowess that is the mark of the mental? As the mind-constituting practice waned, would people lose their minds? What would that be like? Could people come to view all their own first-person reports as unprivileged? What would they say--"We used to have minds, but now we just have brains"?
Would their minds cease to exist once this rush to misjudgment took place? For current Rorty, this is surely a paradigm of a misguided question, assuming, as it does, that there is a neutral standpoint from which the Truth of the ontological claim could be assessed. But many of us unre(de)constructed types may think we can take these questions about the justification and confirmation of our representations more seriously than he now allows. (In fact, he tries to soften this blow by granting scientists and other public and private investigators what he has described to me as a "vegetarian" concept of representation--not the whole ontological hog, but some sort of internal realism in which "facts" may be distinguished from "fictions"--but keep those scare-quotes handy. I think, however, that once this vegetarian concept of representation is exploited to the hilt, we will have enough of a "mirror of nature" in our hands to satisfy all but the most hysterical Realists.)
Back in 1970, the ethos of analytic philosophy let Rorty glide rather swiftly over the question of what the grounds for adopting this linguistic practice might be. In that paper he doesn't emphasize the fact that this innovation might be motivated, or defended against criticism (rightly or wrongly), but he also doesn't treat it as if it would have to be a surd memic mutation, a random happening that had huge consequences for our conceptual scheme but was itself undesigned and beyond defense or criticism. To describe the change in linguistic practices that would amount to the birth of minds, he exploits an elaboration of Wilfrid Sellars' (1963) justly celebrated just-so story about Jones, "the man who invented the concept of mind" (p411). Jones, Rorty reminds us, organized his shrewd observations of the comings and goings of people into a theory which postulated covert events and states in people's heads, the history of which would account for all their overt behavior. Jones then trained all the people in the fine art of making non-inferential reports about these states and events he had posited. When the training was complete, he had succeeded in transforming people from relatively inscrutable objects of theoretical analysis into reliable divulgers of their own internal workings.
According to Rorty, those who went along with Jones
found that, when the behavioral evidence for what Smith was thinking about conflicted with Smith's own report of what he was thinking about, a more adequate account of the sum of Smith's behavior could be obtained by relying on Smith's report than by relying on the behavioral evidence . (p.416).
This passage needs some emendation. Smith's report is part of the behavioral evidence, surely, and a particularly revealing part (when interpreted as a report, not as mere lip-flapping). What Rorty means is that Smith's report, interpreted as a speech act, is recognized as providing a more adequate account than all the other behavioral evidence. He imagines that once this appreciation--it might be misappreciation--of the power of self-reports to trump other evidence was in place,
it became a regulative principle of behavioral science that first-person contemporaneous reports of these postulated inner states were never to be thrown out on the ground that the behavior or the environment of the person doing the reporting would lead one to suspect that they were having a different thought or sensation from the one reported. (p416)
But why should this become a regulative principle? Why turn the recognition of high reliability--what Armstrong had called "empirically privileged access" (Rorty, 1970, p.417)--into a constitutive declaration of incorrigibility? Is this just an unmotivated overshooting of social practice, a bandwagon effect or other byproduct of enthusiasm for Jones' theory? Or might there be some deeper reason--an actual justification--for thus shifting the very criteria (to speak in 60s-talk) for the occurrence of mental phenomena?
Rorty's linguistic convention is a close kin (a heretofore unacknowledged ancestor) of the ploy I attribute to "heterophenomenologists" (1991): deliberately permitting the subject's word to constitute the subject's "heterophenomenological world," creating by fiat a subjective or first-person perspective whose details then become the explicanda for a materialist, third-person theory of consciousness. I took the existence of a wide-spread belief in the primacy of the first-person point of view as given, and characterized heterophenomenology as the neutral method science could--and does--use to investigate the relations between the subjective and objective. Rorty's papers suggest that the emergence of a first-person point of view is itself an effect of a similar burden-shifting move.
In his 1972 paper, Rorty hints at the point I now want to examine in more detail:
If, with respect to a very sophisticated machine, we found that certain states played roles in its behavioral economy very close to those which being frantically hungry, thinking of Vienna, etc., played in ours, then (given that the machine reported on such states and reported making no inferences to such reports) we might decide to extend the same heuristic rule to the machine's reports of those states. But if we then found that the simplest and most fruitful [emphasis added] explanations of the machine's behavior involved overriding these reports, we should cease to apply this rule. (1972, p.215)
This suggests that simplicity and fruitfulness were the grounds for "extending the heuristic rule" in the first place, but why? How? Let us expand the account of this intuition pump, guiding and supporting our judgments by some facts that could only have been dimly imagined in 1972. There is today an entity in roughly pre-Jonesian position, a plausible candidate (with some optimistic projections of engineering) for elevation to first-person status: Cog, a "very sophisticated machine" indeed.
2. The Birth of Cog's Mind: a Just-So Story
At the AI Lab at MIT, Rodney Brooks and Lynn Andrea Stein are leading a team (of which I am a member) that is currently attempting to create a humanoid robot called Cog. Its name has a double etymology: on the one hand, Cog is intended to instantiate the fruits of cognitive science, and on the other, it is a concrete machine situated in the real, non-virtual world, with motors, bearings, springs, wires, pulleys--and cogs. Cog is just about life-size--that is, about the size of a human adult. Cog has no legs, but lives bolted at the hips, you might say, to its stand. This paraplegia was dictated by intensely practical considerations: if Cog had legs and could walk, it would have to trail a colossally unwieldy umbilical cord, carrying power to the body and input-output to its brain, which is about the size of a telephone booth and stands to the side, along with large banks of oscilloscopes and other monitoring devices. No batteries exist that could power Cog's motors for hours on end, and radioing the wide-bandwidth traffic between body and brain--a task I took for granted in "Where am I?" (1978)--is still well beyond the technology available.
Cog has no legs, but it has two human-length arms, with hands (three fingers and a thumb, like Mickey Mouse) on the wrists. It can bend at the waist and swing its torso, and its head moves with three degrees of freedom just about the way a human head does. It has two eyes, each equipped with both a foveal high-resolution vision area and a low-resolution wide-angle parafoveal vision area, and these eyes saccade at almost human speed. That is, the two eyes can complete approximately three fixations a second, while you and I can manage four or five. Your foveas are at the center of your retinas, surrounded by the grainier low-resolution parafoveal areas; for reasons of engineering simplicity, Cog's eyes have their foveas mounted above their wide-angle vision areas, so they won't give it visual information exactly like that provided to human vision by human eyes (in fact, of course, it will be vastly degraded), but the wager is that the information provided will be plenty to give Cog the opportunity to perform impressive feats of hand-eye coordination, identification, and search.
Since its eyes are video cameras mounted on delicate, fast-moving gimbals, it might be disastrous if Cog were inadvertently to punch itself in the eye, so part of the hard-wiring that must be provided in advance is an "innate" if rudimentary "pain" system to serve roughly the same protective functions as the reflex eye-blink and pain-avoidance systems hard-wired into human infants. Cog will not be an adult at first, in spite of its adult size. It is being designed to pass through an extended period of artificial infancy, during which it will have to learn from experience, experience it will gain in the rough-and-tumble environment of the real world. Like a human infant, however, it will need a great deal of protection at the outset, in spite of the fact that it will be equipped with many of the most crucial safety-systems of a living being. It has limit switches, heat sensors, current sensors, strain gauges and alarm signals in all the right places to prevent it from destroying its many motors and joints. The surfaces of its hands and other important parts are covered with touch-sensitive piezo-electric membrane "skin," which will trigger signals when they make contact with anything. These can be "alarm" or "pain" signals in the case of such fragile parts as its "funny bones"--electric motors protruding from its elbows--but the same sensitive membranes are used on its fingertips and elsewhere, and, as with human tactile nerves, the "meaning" of the signals sent along their attached wires depends on what the central control system "makes of them" rather than on their "intrinsic" characteristics. A gentle touch, signalling sought-for contact with an object to be grasped, will not differ, as an information packet, from a sharp pain, signalling a need for rapid countermeasures. It all depends on what the central system is designed to do with the packet, and this design is itself indefinitely revisable--something that can be adjusted either by Cog's own experience or by the tinkering of Cog's artificers.
Decisions have not yet been reached about many of the candidates for hard-wiring or innate features. Anything that can learn must be initially equipped with a great deal of unlearned design. That is no longer an issue; no tabula rasa could ever be impressed with knowledge from experience. But it is also not much of an issue which features ought to be innately fixed, for there is a convenient trade-off. Any feature that is not innately fixed at the outset, but rather gets itself designed into Cog's control system through learning, can then often be lifted whole (with some revision, perhaps) into Cog-II, as a new bit of innate endowment designed by Cog itself--or rather by Cog's history of interactions with its environment. So even in cases in which we have the best of reasons for thinking that human infants actually come innately equipped with pre-designed gear, we may choose to try to get Cog to learn the design in question, rather than be born with it. In some instances, this is laziness or opportunism--we don't really know what might work well, but maybe Cog can train itself up. In others, curiosity is the motive: we have already hand-designed an "innate" version, but wonder if a connectionist network could train itself up to do the task as well or better. Sometimes the answer has been yes. This insouciance about the putative nature/nurture boundary is already a familiar attitude among neural net modelers, of course. Although Cog is not specifically intended to demonstrate any particular neural net thesis, it should come as no surprise that Cog's nervous system is a massively parallel architecture capable of simultaneously training up an indefinite number of special-purpose networks or circuits, under various regimes.
How plausible is the hope that Cog can retrace the steps of millions of years of evolution in a few months or years of laboratory exploration? Notice first that what I have just described is a variety of Lamarckian inheritance that no organic lineage has been able to avail itself of. The acquired design innovations of Cog-I can be immediately transferred to Cog-II, an evolutionary speed-up of tremendous, if incalculable, magnitude. Moreover, if one bears in mind that, unlike the natural case, there will be a team of overseers ready to make patches whenever obvious shortcomings reveal themselves, and to jog the systems out of ruts whenever they enter them, it is not so outrageous a hope, in our opinion. (But then, we are all rather outrageous people.)
One talent that we have hopes of teaching to Cog is at least a rudimentary capacity for human language. And here we run into the fabled innate language organ or Language Acquisition Device made famous by Noam Chomsky. Is there going to be an attempt to build an innate LAD for our Cog? No. We are going to try to get Cog to build language the hard way, the way our ancestors must have done, over thousands of generations. Cog has ears (four, because it's easier to get good localization with four microphones than with carefully shaped ears like ours!) and some special-purpose signal-analyzing software is being developed to give Cog a fairly good chance of discriminating human speech sounds, and probably the capacity to distinguish different human voices. Cog will also have to have speech synthesis hardware and software, of course, but decisions have not yet been reached about the details. It is important to have Cog as well-equipped as possible for rich and natural interactions with human beings, for the team intends to take advantage of as much free labor as it can. Untrained people ought to be able to spend time--hours if they like, and we rather hope they do--trying to get Cog to learn this or that. Growing into an adult is a long, time-consuming business, and Cog--and the team that is building Cog--will need all the help it can get.
Obviously this will not work unless the team manages somehow to give Cog a motivational structure that can be at least dimly recognized, responded to, and exploited by naive observers. In short, Cog should be as human as possible in its wants and fears, likes and dislikes. If those anthropomorphic terms strike you as unwarranted, put them in scare-quotes or drop them altogether and replace them with tedious neologisms of your own choosing: Cog, you may prefer to say, must have goal-registrations and preference-functions that map in rough isomorphism to human desires. This is so for many reasons, of course. Cog won't work at all unless it has its act together in a daunting number of different regards. It must somehow delight in learning, abhor error, strive for novelty, recognize progress. It must be vigilant in some regards, curious in others, and deeply unwilling to engage in self-destructive activity. While we are at it, we might as well try to make it crave human praise and company, and even exhibit a sense of humor.
The computer-complex that has been built to serve as the development platform for Cog's artificial nervous system consists of four backplanes, each with 16 nodes; each node is basically a Mac-II computer--a 68332 processor with a megabyte of RAM. In other words, one can think of Cog's brain as roughly equivalent to sixty-four Mac-IIs yoked in a custom parallel architecture. Each node is itself a multiprocessor, and instead of running Mac software, they all run a special version of parallel Lisp developed by Rodney Brooks, and called, simply, L. Each node has an interpreter for L in its ROM, so it can execute L files independently of every other node.(2) The space of possible virtual machines made available and readily explorable by this underlying architecture is huge, of course, and it covers a volume in the space of all computations that has not yet been seriously explored by artificial intelligence researchers. Moreover, the space of possibilities it represents is manifestly much more realistic as a space to build brains in than is the space heretofore explored, either by the largely serial architectures of GOFAI ("Good Old Fashioned AI", Haugeland, 1985), or by parallel architectures simulated by serial machines. Nevertheless, it is arguable that every one of the possible virtual machines executable by Cog is minute in comparison to a real human brain. In short, Cog has a tiny brain. There is a big wager being made: the parallelism made possible by this arrangement will be sufficient to provide real-time control of importantly humanoid activities occurring on a human time scale. If this proves to be too optimistic by as little as an order of magnitude, the whole project will be forlorn, for the motivating insight for the project is that by confronting and solving actual, real time problems of self-protection, hand-eye coordination, and interaction with other animate beings, Cog's artificers will discover the sufficient conditions for higher cognitive functions in general--and maybe even for a variety of consciousness that would satisfy the skeptics.
Now we are ready to consider Rorty's thesis. At the Royal Society meeting at which I presented the first description of the Cog project, J.R. Lucas embarked on what he took to be the first step of a reductio ad absurdum: if a robot were really conscious, we would have to be prepared to believe it about its own internal states. This move delighted me, for not only did Lucas thereby implicitly endorse Rorty's thesis that incorrigibility was the mark of the mental; it also provided an instance in support of his canny observation that it is skepticism about incorrigibility in machines that strikes many observers as grounds for dualism. My response to Lucas was to give the invited implication a warm welcome; we would indeed be prepared to grant this incorrigibility to Cog. How so?
Cog is equipped from the outset with a well-nigh perfect suite of monitoring devices that can reveal all the details of its inner workings to the observing team. In other words, it will be born with chronically implanted "cerebroscopes" that could hardly be improved upon. Add to this the fact that these observers are not just Johnny-come-latelies but Cog's designers and creators. One might well think then that Cog's observers would have an insurmountable lead in the competition for authority about what is going on inside Cog. The prospect of their finding it "simple and fruitful" to cede authority to Cog's own pronouncements may seem dim indeed.
But all the information visible on the banks of monitors, or gathered by the gigabyte on hard disks, will be from the outset almost as hard to interpret, even by Cog's own designers, as the information obtainable by such "third-person" methods as MRI and CT scanning in the neurosciences. As the observers refine their models, and their understanding of their models, their authority as interpreters of the data may grow, but it may also suffer eclipse. Especially since Cog will be designed from the outset to redesign itself as much as possible, there is a high probability that the designers will simply lose the standard hegemony of the artificer ("I made it, so I know what it is supposed to do, and what it is doing now!").
This is a serious epistemological problem even for traditional serial computer programs when they grow large enough. As every programmer learns, it is essential to "comment" your "source code." Comments are lines of ordinary language, not programming language, inserted into the program between special brackets that tell the computer not to attempt to "execute" them as if they were part of the program. By labeling and explaining each subassembly via helpful comments (e.g., "This part searches the lexicon for the nearest fit, and deposits it in the workspace"), programmers can remind themselves and other observers what the point or function of each such part is supposed to be. (There is no guarantee that the assembly in question actually executes its intended function, of course; nothing is more common than false advertising in the comments.) Without the handy hints about how the programmer intended the process or state to function, the very identity of the state entered when a computer executes a line of code is often for all intents and purposes inscrutable. The intrinsic or just local features of the state are almost useless guides, given the global organization on which the proper functioning of the system--whatever it is--depends.
Even in traditional programs, the actual function--and hence actual identity--of a state or event may well evolve away from what is advertised in the accompanying comment, which may remain unchanged in the source code long after it has been rendered obsolete by undocumented debugging efforts. Large programs never work as intended at first--this is a regularity so unexceptioned that one might almost consider it a law of nature, or the epistemological version of Original Sin. By the time they are actually made to work, the adjustments to their original design specifications are so many, and so inscrutable in combination, that nobody can say with confidence and accuracy what the "intended" function of many of the states is. And the only identity that matters in computer programs is functional identity (a point Rorty makes surprisingly well in his 1972 paper, p.212, in the course of pursuing rather different aims).
In the case of a system like Cog, which is intended from the outset to be self-redesigning on a massive scale, the loss of epistemological hegemony on the part of its "third person" designers is even more assured. Connectionist training regimes, and genetic algorithms, for instance, create competences--and hence states embodying those competences--whose means are only indirectly shaped by human hands. (For that reason, programmers working in these methodologies are more like plant and animal breeders than machine-makers.)
Since, as I noted above, the meaning of signals in Cog's brain is not a function of their intrinsic properties but of their "intended" functions, and since Cog is designed to be indefinitely self-revisable in those functions, Cog's original designers have no secure hold on what the relevant boundaries are between states. What a bit of the system is "supposed to do" is the only anchor for what its meaning is, and when the designers' initial comments about those functions become obsolete, it is open for some new party to become authoritative about those boundaries. Who? Cog itself, the (unwitting) re-designer of its own states. Unlike the genius Jones of Sellars' fable, Cog need have no theory of its own operations (though in due course it might well develop such an auto-psychological interest as a hobby). Cog need only be sensitive to the pressures of training that it encounters "growing up" in a human milieu. In principle it can learn, as a child does, to generate speech acts that do divulge the saliencies of its internal states, but these are saliencies that are created by the very process of learning to talk about them. That, at any rate, is the theory and the hope.
And that is why I gladly defend this conditional prediction: if Cog develops to the point where it can conduct what appear to be robust and well-controlled conversations in something like a natural language, it will certainly be in a position to rival its own monitors (and the theorists who interpret them) as a source of knowledge about what it is doing and feeling, and why. And if and when it reaches this stage of development, outside observers will have the best of reasons for welcoming it into the class of subjects, or first persons, for it will be an emitter of speech acts that can be interpreted as reliable reports on various "external" topics, and constitutively reliable reports on a particular range of topics closer to home: its own internal states. Not all of them, but only the "mental" ones--the ones which, by definition, it is incorrigible about because nobody else could be in a better position than it was to say.
So it is not mere convention that guarantees (while it lasts) that there are minds in this world. There is, as Rorty claims, a convention or something like a convention in the etiology of mind, but it has a natural justification. Ceding authority to a subject-in-the-making is a way of getting it to become a subject, by putting it in a conversational milieu in which its own software, its own virtual Joycean machine (as I called it in 1991), can develop the competence to make self-reports about which it is the best authority because the states and events those self-reports are about get their function, and hence meaning, from the subject's own "take" on them.(3)
"If you called a horse's tail a leg, how many legs would the horse have?" Answer: "Four: calling a tail a leg doesn't make it a leg." True, and calling a machine conscious doesn't make it conscious. Many are deeply skeptical of anti-metaphysical moves such as Rorty's suggestion that a linguistic convention of incorrigibility accounts for the existence of minds, but what they tend to overlook--and what Rorty himself has overlooked, if I am right--is that the existence of such a convention can have effects over time that make it non-trivially self-fulfilling. This is really not such an unfamiliar idea--let's face it: it's Norman Vincent Peale's idea of the power of positive thinking. Or think of Dumbo, the giant-eared little elephant in the Disney cartoon. His friends the crows convince him he can fly by making up a tale about a magic feather that can give him the power of flight just as long as he clutches it in his trunk. By changing Dumbo's attitude, they give Dumbo a power that depends on attitude. Attitudes are real states of people (and elephants--at least in fables--and robots, if all goes well). Changes in conventions can bring about changes in attitudes that bring about changes in competence that are definitive of having a mind.
Could the attitudes lapse? Perhaps they could, but I have shown that Rorty overestimated the power of "cerebroscopes" to trump first-person reports, so there is no good reason to anticipate that the triumph of neuroscience or robotics would bring about the death of the mind.
Dennett, Daniel, 1978, "Where am I?" in Brainstorms, Bradford Books/MIT Press.
---- 1991, Consciousness Explained, New York: Little, Brown.
Haugeland, John, 1985, Artificial Intelligence: The Very Idea, Cambridge MA: MIT Press.
McGeer, Victoria, "Is Self-Knowledge an Empirical Problem? Renegotiating the Space of Philosophical Explanation," J.Phil. XCIII, 1996, pp.483-515.
Nagel, T., 1974, "What is it like to be a bat?" Philosophical Review, 83, pp.435-50.
Rorty, Richard, 1965, "Mind-body Identity, Privacy, and Categories," Review of Metaphysics, 19, p24-54.
----- 1970, "Incorrigibility as the Mark of the Mental." J.Phil.,67, pp.399-424.
------ 1972, "Functionalism, Machines, and Incorrigibility," J.Phil., 69, pp.203-220.
Searle, J., 1980, "Minds, Brains, and Programs," Behavioral and Brain Sciences, 3, p.417-58.
Searle, J., 1983, Intentionality: An Essay in the Philosophy of Mind, Cambridge Univ. Press.
Sellars, Wilfrid, 1963, "Empiricism and the Philosophy of Mind," in Science, Perception and Reality, New York: Humanities Press, pp.127-96.
1. See also Rorty, 1965, which I will not discuss, although it is an important paper in the history of the philosophy of mind.
2. For more details on the computational architecture of Cog, see my "The Practical Requirements for Building a Conscious Robot," in the Philosophical Transactions of the Royal Society, (1994), 349,pp.133-46, from which this brief description is excerpted, or for more up-to-date information, consult the World Wide Web site for the Cog Shop at MIT.edu/projects.
3. These points grew out of discussion with Victoria McGeer, of a paper she presented at the Society for Philosophy and Psychology meeting in Vancouver in 1993; the successor to that paper, McGeer,1996, carries these points further.