Leviathan’s Program

Psychologists make models of the mind in order to explain what we say and do. Some particularly want to explain our abilities: how do we build a tower from toy blocks, or recognize a goldfinch? How do we manage to get across a room without hitting the furniture? Like many naive questions that lead to science, these do not have a meaning that is entirely fixed before people start to answer them. The questions seem, rather vaguely, in place when we reflect that we must learn to do these various things, that characteristic mistakes are made in trying to do them, and that our abilities to do them can be impaired through illness, injury, or age.

Yet at the outset, it is not very clear what you are asking when you ask exactly how we do such things: Is there a way in which you pick up a brick? The models of the mind that are designed to answer such questions also help to set them up, by defining what a “way” of doing something will be. They provide a topography of the mind’s functions, according to which our skills in perception, movement, or thought can be analyzed. We shall have come to understand how we do one of these familiar tasks when we see what must go on within the mind, so described, if we are to do it.

That bland and general description, however, conceals a difficulty—one large enough to make some people think that no explanation in such terms could ultimately succeed. How are we to describe what goes on in the mind? If we say that elements in the mind are able to do various relevant things—that one element recognizes a brick, for instance, and another sends the arm off to reach for it—we have only moved the problem back, from a familiar agent (the person) to a less familiar agent, the departmental chief in the mind. We shall have fallen into the homunculus fallacy, the old mistake of explaining a person’s actions in terms of actions by another (perhaps more specialized) person inside him.

If the aim is to provide a model by which all basic psychological skills can be explained, there is only one way out of this difficulty: the analysis must be carried to a level at which the mental elements themselves have no psychological capacities at all. Each of these ultimate mental items must be the kind of agent that simply, when it receives an input, sends off some output, which typically triggers other elements. The model-builder, aiming to display the structure of the mind of an agent who does have psychological abilities, will construct hierarchies and systems of such elements, specialized in the information that can affect them and in the changes that they can initiate.

It is a system of this kind that Marvin Minsky discusses in his book—or, rather, the possibility of such a system, or speculations around that possibility, since the book is ruminative and discursive, presents no detailed theory, and is not tied down to technicalities. It is large in format, and full of simple diagrams, and every section is presented on a separate page. It is the model-builder’s thought-book, designed to encourage you to wander around in it and go off in interesting directions. I think that Minsky conceives of it as rather like a mental model itself, an application of its own theory. Its reader is in somewhat the situation of the explorer in a recent Doonesbury strip, clambering around inside Reagan’s brain; and as that adventure itself suggests, he or she will find much of it fairly easy going.

Minsky’s model involves numerous “agents” organized hierarchically into “agencies.” An agent is merely a switch that turns on something else when it itself is turned on, and an agency is an organized set of switches. They are linked by complex lines of communication, called “K-lines,” which are set up and modified by what happens to the system (by experience, as one might say). The organization of the agencies is explicitly bureaucratic, and this is one of two analogies that helps to guide this model of the mind. It is a very old analogy, that of the Leviathan. The organization of agencies is the “Society” of the book’s title, but the society in question is conceived not as a political community, but as a rigidly stratified corporation. The other analogy that guides the model is that of a computer, an information-processing device, and here the similarity is thought to go beyond an analogy: the mind, for Minsky, is such a device, and the project of discovering its architecture constitutes a well-known research program, that of psychology as cognitive science.

This is closely related to the subject called “Artificial Intelligence” of which Minsky has been for many years a leader, having set up in 1959, with John McCarthy, the Artificial Intelligence Laboratory at MIT. “Cognitive science” is now the usual name for the subject that tries to explain human psychological capacities in these terms; “artificial intelligence” particularly emphasizes the project of simulating thought processes, which is not necessarily at the heart of the matter. Although computer programs controlling robot devices have played a big part in the development of cognitive science, that science is not committed to thinking that one could simulate much of what a human being can do on the kinds of computer we are familiar with. Minsky freely admits that the brain does not work like such machines, and that moreover we know little of how it does work. The models of various agencies with their linking connections are not meant to relate directly to structures in the brain but rather suggest the structure of the work the brain does. It could perhaps be misleading that some of the diagrams, with their interweaving cords, look rather as though they were intended as neurology.

The idea of cognitive science is that the description of the mind should correspond not to a wiring diagram but to a computer program. In the case of a machine, it is not hard to identify the program, since someone wrote it (or, in some cases, wrote another program that wrote it). In the case of human beings and other animals, however, the Divine Hacker who wrote the system has provided no manual to it, and we have to discover the program by scientific inquiry. That inquiry is constrained by what we know or believe about the nervous system, but within those constraints, there is a vast range of possibilities for what such a program might be. Granted there is no manual or directly identifiable software, what counts as the mind’s actual program? What does it mean to say that a given structure is in fact part of it? Presumably it means that a theory including that structure provides the best fit to the psychological data, such as the mistakes that people typically make, and also fits our understanding of evolutionary history. Thus it is a plus for a particular theory of vision expressed in these terms (derived from the work of the late David Marr) that it helps us to grasp how our visual system can be an elaborate descendant of more primitive visual systems.

Some who criticize the research program of cognitive science think that they can show that in principle it could not work. Some suggest that if it escapes the homunculus fallacy, it will merely land on the other horn of a dilemma: how could an arrangement of elements that do not understand anything issue in a system (as Minsky puts it, a society) that does understand something?¹ This Catch-22 style of argument is too quick to be convincing, and like some other arguments in the history of philosophy designed to show that a particular scientific advance is impossible, can be accused of begging the question: what is at issue in this research, in part, is precisely whether intelligent systems can be compounded of unintelligent parts.

Other critics are merely empirical and skeptical, and say that there is no reason to think that a program of this kind must work, and that the rate of progress achieved after a great deal of labor provides not much reason to think that it is going to do so. This is a weighty objection, and what has been achieved is disappointing, particularly when measured against the large promises still made by some of the apologists of artificial intelligence. In this book, Minsky does not do much to reassure the skeptics, but I do not think that he is trying to. Except in passing, he does not report results, nor does he set out to justify the entire enterprise. He merely sets out various ideas and models, and invites us to consider what might be done with them.

Two authorities quoted on the back of the jacket refer to the “theory” of the mind that the book offers, but I do not see any such theory. A theory would have actually to explain something, and to explain anything, especially in this kind of field, involves explaining a lot. As Noam Chomsky has always emphasized in linguistics, it is not hard to think up some model or principle that will fit a few cases—the difficulties break out when it is confronted with a wider range of phenomena. To get to a point at which anything was actually explained, and there was really an effective theory, the notions sketched in this book would have to be given a demanding application. It may be that existing work in cognitive science has succeeded, to some extent, in giving them such an application, but if so, it is not displayed here, and Minsky, as I read him, does not suppose that it is. His aim is rather to get the reader used to the notions.

Minsky introduces a good many new technical terms, but (despite his dismissing old terms as vague or being attached to discredited theories) the new terms are sometimes just ways of labeling well-tried conceptions. A “polyneme,” for instance, is an old friend from empiricist philosophy, the complex idea (without the demand, made by some empiricist philosophers, that it should present itself as an image). Just as a word was supposed, by that old philosophy, to have its meaning in virtue of a complex idea compounded of simple ideas, so exactly the polyneme associated with the word “apple” “sets your agencies for color, shape, and size into unrelated states that represent the independent properties of being red, round, and ‘apple-sized.”‘ (The word should strictly speaking be “polymneme,” since its suffix is said to indicate a connection with memory; it is not clear whether etymology has been overridden or overlooked.)

Polynemes are connected, indirectly, to items in the mind Minsky calls frames, which are, effectively, sentence-forms. A frame embodies the structure of a certain kind of situation. A kind of frame that Minsky finds particularly important is the “Trans-frame,” which relates two items by some kind of trans-action, as when one causes another, or information is passed from one to another, or a spatial movement occurs. Thus there are “travel-frames,” containing blanks or “pronomes” for each of a number of elements including Actor-Origin-Trajectory-Destination-Vehicle. A particular value of such a frame would be a sentence such as “Jack drove from Boston to New York on the turnpike with Mary.” When you think of this sentence, and you think of the destination of the trip, you think of New York, and this is because

the polyneme for New York is attached to an AND-agent with two inputs; one of them represents the arousal of the travel-frame itself, and the other represents the arousal of the Destination pronome.

In these connections, Minsky emphasizes the role of “stereotypes” or “default assumptions.” A frame embodies a standardized situation. Given the sentence about Jack’s travel, we assume, unless there is an indication to the contrary, that it was a car that he drove, that the car had no more than four wheels, that Mary was with him in it rather than in another car, and so on. There can be little doubt that a structure of stereotypes or presumptions does underlie our linguistic and other capacities, and to this extent the view that Minsky uses is shared by much contemporary philosophy of mind and language.

Just as there are traditional elements in the account, however, so there are traditional problems, which Minsky seems not to recognize. One problem concerns “frames,” and the ways in which they are linked to language. It is easy to postulate a frame corresponding to a sentence about travel, with appropriate “pronomes”: it is simply read off from the structure of an English sentence. But what is the psychological status of this frame? Do we come to acquire it because we have been exposed to English sentences, and have in effect abstracted it from them? How could we do that unless we already understood them? And how could we understand them, at least on Minsky’s account of the matter, unless we already possessed the frame?

If we are to get out of this circle, it looks as though we must suppose that there is an innate structure, a basic architecture of the mind, that is brought into action by learning a given language and by other kinds of experience. But then it is surely quite naive to suppose that this structure should stand in simple correspondence to the surface structure of English sentences. Minsky himself makes the point that a small child can easily acquire any human language. (He also offers the very ingenious, if not quite convincing, hypothesis that this ability decays before the time of puberty—as indeed it does—in order that parents should not acquire imperfect speech habits from their children.) Real work in comparative linguistics is needed if we are to know what structures of the mind underly this ability. The point is not that, in absence of such work, Minsky’s models provide too narrow an explanation of understanding, one that may need generalization in the light of other languages. Rather, without such work there is no psychological explanation at all—for all we know Minsky may merely be rewriting English in psychologese.

Minsky well brings out the enormous complexity of simple tasks, such as picking something up or putting one thing on another. We certainly know, in part from the hard and often disappointed work of AI investigators, that whatever processes underly the capacities to do such things, they are very subtle and complex. But having started off admirably by making us look again at what we take for granted, he is disposed to move to another thought, that what we regularly think of as deep or difficult is merely complicated, and probably no more complicated than what we ordinarily think of as simple.

This step has rather less admirable results. It rests on the assumption that all complexity, cultural or personal, is of the same type, and this assumption leads Minsky to occasional outbreaks of reductivism, particularly near the beginning of the book, that are always unnecessary and sometimes astonishing. Thus he writes that “Selves” are not, as some “ordinary views” are alleged to hold, “magic, self-indulgent luxuries that enable our minds to break the bonds of natural cause and law. Instead, those Selves are practical necessities.” Their function is to enable us to carry out our plans. Without Selves, “we’d never get much done because we could never depend on ourselves”; as he also puts it, “One function of the Self is to keep us from changing too rapidly.” Too rapidly for whom? How can there be plans for the future to worry about, without already a self that has a future and whose plans they are.

In supposing that “we” can adopt ingenious schemes to bring ourselves into existence, as also when he tries to answer the baffling question, “How do we control our minds?” Minsky seems to have been driven to absurdity by a desire to represent the noninstrumental as instrumental. This emphasis does not necessarily follow from taking cognitive science seriously as a research program; but I think that it is a very natural result of exaggerating what it might achieve. Because it basically tries to find out how we do things, it naturally represents what we do in terms of problems and tasks, success and failure, and this, of course, presupposes that we have ends in view. With many of our activities, such as picking up a brick, the end is obvious, and at the same time the question of how, as Minsky emphasizes, is not trivial. But when we get away from these immediate or obvious kinds of ends, it may be less easy to see our projects or concerns in the light of problem-solving, and so less rewarding to try to apply the cognitive science repertoire, at this level, to what we do and experience.

One aspect of our activities to which this obviously applies is what is called “creativity.” Minsky, like other writers in these subjects, discusses this as a matter of problem-solving, considering such things as programs that “try out” strategies that may have been, in part, randomly generated. Such strategies may do for chess; but for most creativity in the arts and, indeed, the sciences, such an account is misconceived. To make a creative step is not simply to produce something new or unpredictable, but to produce something new that we find interesting or significant; and the fact that we do see some innovation in that light is, first of all, a cultural and not merely a psychological matter, and, inasmuch as it is a psychological matter, it is not in the first place a question of the heuristics of problem-solving.

The right conclusion from this is not that cognitive science is a dead duck, but that it should stick to trying to solve the kinds of problems it is adapted to solving, which, as it reminds us itself, are hard enough.² When it goes beyond those problems into larger ambitions of explaining human life, it can fall into a stupid and shallow reductivism that is not essential to it. It does this because it has to look for elementary and obvious ends that are supposedly served by more complex and culturally elaborated activities. A particular, and notably ideological, version of this is that in which the “success” that is introduced by the very idea of problem-solving is identified simply with Success: career advancement, fame, and competitive victory are assumed to be the ends of human activity. There is a strain of this in Minsky’s book, minor but, when it appears, quite dramatic. “Consider,” he says, “an example from everyday life”:

I was trying to concentrate on a certain problem but was getting bored and sleepy. Then I imagined that one of my competitors, Professor Challenger, was about to solve the same problem. An angry wish to frustrate Challenger then kept me working on the problem for a while. The strange thing was, this problem was not of the sort that ever interested Challenger.

Well, there’s one strange thing. Another is that this is an example from everyday life. Yet another is that this is the example from everyday life that would first occur to one in thinking about the self’s construction of itself.

It would be wrong, however, to suggest that this represents the limits of Minsky’s view of things, or even a major part of it. His book is full of other kinds of thoughts, including numerous quotations from religious, philosophical, and literary writers, from Proust to the Buddha to what seem to be calendar mottoes. A few seem to me intolerably cute and corny; others are marvelous, and some of them come from a quite different kind of life, casting a great and revealing shadow over the book. So Dr. Johnson:

And while it shall please thee to continue me in this world, where there is much to be done and little to be known, teach me, by thy Holy Spirit, to withdraw my mind from unprofitable and dangerous enquiries, from difficulties vainly curious, and doubts impossible to be solved.

It is certainly no complaint that the quotations are a mixed bag. Going through Minsky’s book is a bit like going around his house, and he has a perfect right to put up on its walls what he finds familiar, helpful, interesting, or attractive. It is perhaps rather more disappointing that these thoughts are merely stuck on the wall. If he takes seriously some of the things he quotes, it is hard to see how he can say some of the things he says.

For another kind of reason, too, it is hard to see how he can say some of them. At various points in the book, statements turn up that should not be there, not because they are theoretically shaky or deeply misconceived, but merely because they are, in the most immediate and everyday sense, not true. Minsky says:

When Jack says, “Mary knows geometry,” this indicates to us that Jack would probably be satisfied by Mary’s answers to the questions about geometry that he would be disposed to ask.

It does not indicate this, unless we know something special about Jack, for instance, that he is Mary’s teacher: on the contrary, a centrally important occasion of Jack’s saying “she knows geometry” is that on which he might add “…but I don’t.” Again, Minsky points out that we might say, “I just heard a pin drop,” but not “I hear a pin dropping”; and he says that this is because a pin takes so short a time to drop that we cannot get the sentence out. But we could not say it even if the pin were dropping from the top of the Empire State Building. The explanation might rather be that a pin dropping makes no sound.

In both these cases, and others, we might repeat what A. E. Houseman said about an evident but neglected principle of classical scholarship: “Three minutes’ thought would suffice to find this out; but thought is irksome and three minutes is a long time.” Does it matter if Minsky, economizing on time, cannot spare a second to turn up the corner of a sentence he has just written? There are circumstances in which it may help inquiry not to fuss with exactitude. But it is very doubtful that cognitive science is in those circumstances, and granted that it is struggling with the complexities of our thought about the everyday, it will do best if it listens carefully to its own words.

I think that many people will enjoy Minsky’s book. It brings in a lot of ideas and gives a suggestive sketch of a certain kind of psychological model. In addition, it has a kind of disheveled, undemanding, personal quality that contrasts amiably with the aggressive scientistic display that characterizes some of cognitive science and, still more, the philosophical propaganda for it. It is written in an easy and unintimidating manner. It is intriguingly concerned with skills of an everyday kind. But it is sometimes inattentive to everyday truths, and it cannot afford to be: no inquiry that is going to help us understand ourselves can do without that kind of truthfulness, an acute and wary sense of the ordinary.