Metodología

The Rise of Cognitivism

So why did cognitive psychology/science eclipse behaviorism? What happened was that psychologists got leave to “do what comes naturally”—leave denied to them by the unforgiving behaviorism of Watson and his successors for more than sixty years. What comes naturally to psychologists is mentalism. “Psychology is the science of mental life,” pronounced William James in 1890, but his writ ran for only a few pre-Watsonian years. Harvard professor George Miller used James’s sentence as a rallying cry in the fi rst salvo of the mentalistic counter-revolution the title of his 1962 book Psychology: The Science of Mental Life. ³

Not only do psychologists want to study mental life, they want to study the mental life of human beings. This was also impeded by the behaviorists, who argued from evolutionary biology that simpler, more comprehensible, anteced-ents of the human mind are to be found in the behavior of animals. The need for experiments in which reward and punishment may be administered freely also implied that scientifi c psychology must be centered on animal experi-ments. This was all soon to be dismissed as “rat psychology.” Psychologists had been chafi ng against the constraints of biology and behaviorism for two generations when they were liberated by the cognitive revolution. The new cognitive psychology legitimated the study of language, mental representation and consciousness, which (it was argued) had been ruled out or trivialized by the old behaviorism. As one cognitivist commented: “[I]f mental states exist, a complete psychological explanation should give an account of them even if they could be dropped from the laws of behavior.” ⁴

How was this emancipation achieved? Chiefl y through two things: the digi-tal computer, and the act of theoretical seppuku committed by Skinnerian radi-cal behaviorism. I discussed B. F. Skinner’s antipathy to theory in Chapter 5 , and I will say a bit more about it later in this chapter. But fi rst, the computer.

The digital computer meant that theories—of mental life, perception, syn-tax or whatever—could be simulated, made up into a working computer model . This was heady stuff in the 1960s and 1970s. Computer-literate cognitivists boasted in the 1950s that the best chess player in the world would be a com-puter within ten years, automatic language translation was around the corner, and comprehensive models of thought were on the horizon. Already, in 1957, a computer program had discovered novel proofs for theorems in Russell and Whitehead’s Principia Mathematica. ⁵ By the 1960s, the sunlit uplands of cog-nitive hegemony were in full view. Especially confi dent prophecies for the new cognitive psychology were made by psychologist and economics-Nobelist-to-be Herbert Simon—see, for example, his ambitiously titled collection of papers Models of Thought. ⁶

Not only did computation offer power, it also offered respectability. Rigor and quantifi cation had been almost a monopoly of the behaviorists. The vague, verbal theorizing of the Gestalt school and other proto-cognitive psychologists could easily be dismissed. But now, the ability to simulate mentalistic theo-ries with quantitative precision conferred a Good Science Seal of Approval

Cognitivism and the New Behaviorism 151 on cognitive psychology. No longer need cognitive psychologists feel embar-rassed by hardheaded behaviorists. Philosophical objections to the study of subjective experience remained unanswered, but no matter: Computers could remember, perceive, even understand. The possibilities appeared boundless.

Well, not quite boundless, perhaps, but still pretty good. Now, with com-puters on every desktop thousands of times more powerful than the water-cooled monsters available to the prophets of cognitive science, the uplands, while still in sight, seem a little cloudy. Chess champion Gary Kasparov has fi nally been beaten by IBM’s Deep Blue, ⁷ Watson (named not after J. B. but Thomas W., a founder of IBM) has defeated top Jeopardy contestants ⁸ and the Internet (ah, the Internet!) has pretty good language translation on offer. But Deep Blue plays chess very differently from Gary Kasparov—more exhaustive search, less remembered strategy; and the computer Watson excels in large part because of speed. Artifi cial intelligence, brilliant at retrieving, sorting and analyzing “big data,” is still more artifi cial than intelligent.

Radical behaviorists were insulated from the siren song of cognitivism by Skinnerian suspicion of formal theory and by their dedication to experimental results as ends in themselves. Behaviorists trained in the traditions of Hull and Tolman were much more susceptible, however. Several of the leading cogni-tive psychologists of the 1970s and 1980s were re-treaded rat runners, or at least, people reared in the Hullian tradition. The few neo-Tolmanians for the most part embraced with enthusiasm the new animal-cognition movement.

How has the computer-assisted excitement of the original cognitivists panned out? Well, okay, but criticism continues. Attacks are coming from the left and the right. On the left, philosophers are arguing that cognitiv-ism is failing because the computer metaphor is wrong—and because cog-nitivism is not mental enough. On the right, some artifi cial intelligence (AI) researchers and roboticists are arguing that the emphasis on representation and symbolic thought is largely irrelevant to the development of truly intel-ligent systems. I suggest that the philosophical objections to contemporary cognitive psychology merely underscore the importance of a sophisticated behaviorism. The AI objections partly defi ne the new behaviorism. Let’s look at the philosophy fi rst.

Philosophical Objections to (Some) Cognitive Psychology

A central tenet of cognitive psychology is that the brain is like a computer and cognitive operations like a program running on that computer. This is still a very widely accepted view in cognition. ⁹ Its founding inspiration was Cam-bridge mathematician Alan Turing (1912–1954). Turing is famous on several counts: He played a key role in deciphering the German “Enigma” code during World War II. He proposed a test for artifi cial intelligence (AI) that has become a classic. ¹⁰ He did the mathematics that forms the basis for the general-purpose digital computer, describing what has come to be known as the universal Tur-ing Machine. ¹¹ And he died tragically by his own hand at the young age of

forty-two in the aftermath of a criminal conviction for homosexual behavior, ¹² which was then a criminal offense in the United Kingdom.

Although he owed no allegiance to behaviorism as a movement in psychology—indeed, he was not a psychologist at all—Turing proposed a per-fectly behavioristic test for “thought.” The experimenter types questions that are read and responded to either by an appropriately programmed computer or by another human being. The experimenter is not told which: His task is to decide from the responses whether his correspondent is man or machine.

If he cannot tell which, within a “reasonable time,” then “thought” should be conceded to the machine as well as to the man. Passing the Turing test is an avowed objective of modern cognitive science. Most cognitivists go beyond the possibility of mere functional equivalence to accept the idea that source of behavior, the brain, is to be thought of as a “symbol-manipulating device,” just like the digital computer.

The Turing test can never be decisive because it depends on two things that are both necessary for answering the question, “Does the machine think?” The fi rst is simply time. If the human interrogator can ask only one or two ques-tions, even a very simple computer program can fool him. I remember in the very earliest days of microcomputers showing a NorthStar Horizon desktop PC (4K memory!) running a little program called Lisa to a clinical colleague.

Lisa was a much-simplifi ed version of a more elaborate mainframe program called Eliza, written by Joseph Weizenbaum, ¹³ an MIT computer scientist. Lisa simulated a Rogerian psychotherapist and responded to typed input. Question:

Are you happy? Answer: I am upset about my sister. Q: How do you feel about your sister? (cued by the word “sister”), and so on. The program had no content and no real intelligence. But a couple of minutes’ exposure to it convinced my colleague that it was pretty smart.

The second limitation is one with which Skinner was very familiar: the repertoire of the human interrogator: “The truth of a statement of fact is lim-ited . . . by the repertoires of the scientists involved.” ¹⁴ If the questioner doesn’t ask the right questions, he may fail to detect the difference between a machine and a human being. But what exactly are the right questions? If you know that, you really don’t need the computer to know what human intelligence is.

Philosopher John Searle ¹⁵ attacked the Turing test from another angle.

Searle objects that even if the test is passed, the successful program need not be accepted as a valid theory of the human mind. Searle’s main objection is that anything can be conceived of as a computer: A ball rolling down an inclined plane, for example, is an analogue computer that computes Newton’s laws of motion; a nail driven into a plank computes the distance it must travel given the force of the hammer and the friction of the wood. The problem, says Searle, is that computational properties are not intrinsic to any system; they are interpre-tations imposed from outside. ¹⁶

I don’t believe that this particular objection holds water. Quantum theory, which is one of the very best theories in existence, is often associated with comments like, “If quantum mechanics hasn’t profoundly shocked you, you

Cognitivism and the New Behaviorism 153 haven’t understood it yet,” by iconic physicist Niels Bohr. In other words, a theory may have little or no intuitive appeal, yet work very well. If it explains much with little and makes accurate predictions, we should be happy. Whether the theory is “intrinsic to the system” or not is completely irrelevant.

Searle’s second objection is to accuse cognitivism of the familiar homunculus fallacy. Sophisticated cognitivists are of course aware of this problem: There is no “little man in the head” looking at some sort of screen on which the products of sensation are displayed. An infl uential idea in the study of human memory is that the frontal lobes are the location of something with the homunculus-like name of the “central executive”: “The central executive component of working memory is a poorly specifi ed and very powerful system that could be criticized as little more than a homunculus,” writes its co-inventor, who has defended the idea at length. ¹⁷ Nevertheless, the ways in which important problems are presented imply such a little man, says Searle: “Typical homunculus questions in cognitive science are such as the following: ‘How does the visual system compute shape from shading; how does it compute object distance from size of retinal image?’ ” ¹⁸ The savvy cognitivist might respond by saying that what is really meant is, “How is information about object distance represented in the brain?” and “How is this information derived from the retinal input?”

He will need to add, of course, “How do we know these things?” In other words, we need to know something about how states of mind are translated into things we can measure (i.e., into behavior). This response shows that the question of representation cannot be separated from the questions of computa-tion and accomputa-tion—but it often is, as these quescomputa-tions show. When all three are considered together, it is diffi cult to see how they amount to asking more than,

“How does the brain work to make behavior?” which is not necessarily a com-putational or even a cognitive question.

Searle’s case against the computer metaphor for the mind has some merit.

Much of the standard cognitivist boilerplate about “information processing”

and computational “levels” now looks less impressive. No information is being processed; perhaps nothing is being “represented” at all: The external world is changing, neurons are fi ring, chemicals are circulating and physical actions are occurring. The psychologist’s job is to fi nd the connections among these events. Rhetoric about “levels”—functional, hardware, computational and intentional—is empty in the absence of clear evidence, which can come only from a truly comprehensive theoretical system that explains behavior within a level and connects levels to one another. There is as yet no “mental chemistry”

and “mental physics” to provide justifi cation for a division between the psy-chological equivalents of “atomic” and “molecular” levels, for example. Until some convincing, specifi c justifi cation for levels is provided, we are under no compulsion to accept the idea. It’s just another plausible conjecture.

To these criticisms I would add fi ve more. First, there is the distinction between competence and performance. This distinction has both an innocent and a sinister meaning. The innocent meaning is just the one I discussed in Chapter 7 . The student often knows more than she shows on this exam on this

day. The sinister meaning is more theoretical and arose in connection with language. Like the “levels” just discussed, it is another “trust me” distinction that depends for its validity on the reader’s acceptance of the theorist’s view of the subject. The subject in this case is language, and the theorist is mathemati-cal linguist Noam Chomsky, ¹⁹ who we have already encountered as Skinner’s most virulent critic. The idea is that language follows certain syntactic “rules,”

but that actual examples—speech as she is spoken or writing as she is writ—

fall short of perfection because of constraints such as memory limitations. The competence/performance distinction is less conspicuous now than it used to be—it has run into empirical problems, for one thing:

Although there is plenty of evidence that humans use language creatively (saying and understanding things that have never been said before) and well (with very low error rates), there is very little evidence for the claim that “perfect” knowledge underlies our (occasionally) imperfect behavior. ²⁰

Nevertheless, the Platonic thinking represented by the competence-performance distinction still permeates cognitive psychology.

The arbitrariness of this distinction is obvious from a thought experiment.

Suppose someone gives you a device of unknown function with a numerical keypad. You try it out by typing in various combinations, and you get the fol-lowing set of outputs: 2  2 = 4, 20  20 = 400, 25  25 = 625, 30  30 = 900.

“Great!” you think. “I’ve fi gured it out. It’s a calculator that multiplies.” But then you try: 40  40 = 999—whoops! It turns out that the calculator multi-plies fi ne, so long as the answer is less than 1,000. Anything greater, and it just gives the largest number of which it is capable: 999. How shall we describe the behavior of this device? The competence-performance people would say,

“Well, the deep rule underlying this behavior is multiplication, but because of memory limitations, answers greater than 999 cannot be produced.” There is an exactly parallel argument in linguistic theory to account for the fact that people are limited in their ability to understand and produce deeply embed-ded, but perfectly grammatical, sentences. “Fine,” you may well say, “what’s wrong with the competence-performance account of the behavior of this ‘con-strained’ calculator?” The answer is: “nothing, so far.” The problem is that there are alternative accounts, and deciding among them demands additional information.

For example, one alternative is that the device doesn’t multiply at all, but rather just consults a fi xed lookup table, which comprises all the three-digit (or fewer) products of two-digit numbers. A “brain lesion” (i.e., damage to the device memory) would support this view if it simply knocked out cer-tain answers, but left others involving the two multiplicands unaffected. More general effects might support the competence-performance alternative. The point is that the competence-performance idea, which implies an organization involving a rule-following engine limited by memory and other constraints,

Cognitivism and the New Behaviorism 155 represents just one kind of model for what is happening in the linguistic black box. We are under no obligation whatever to accept it without knowing both the exact nature of the rules that are being enforced and the precise constraints that limit their output. Without the details, we are simply invited to accept as a matter of fact what is nothing more than a self-imposed limitation on theoreti-cal inquiry.

Second, cognitive psychology is all steering and no motor. As Edwin Guth-rie pointed out in his criticism of cognitive behaviorist Edward Tolman many years ago, Tolman’s purely representational approach left the maze-running rat

“buried in thought” at the choice point. Things have not improved much since:

“Psychology is the science that investigates the representation and processing of information by complex organisms,” says the supposedly authoritative MIT Encyclopedia of Cognitive Sciences, ²¹ dismissing in an act of silent imperi-alism not just behaviorism but any nonrepresentational, non-information-processing view of the human mind.

Motivation, the motor, is not intrinsic to cognitive psychology. It is an add-on. Yet for most organisms, the motor is by far the most important com-ponent. The cognitive, “computational” component is for many simple organ-isms almost trivial. As the great dramatists show us, motivation and emotion are major contributors to human behavior also. Every politician and pollster knows that the emotional freight of words and phrases like “extreme,” “risky,”

“working families” (Americans, people), “children,” “our future,” “family val-ues,” etc., is much more important than their syntactic arrangement. In other words, what is important is that the voters hear these words. What is almost irrelevant is the meaning of the sentence in which they are used (so long as it doesn’t provoke hostility or provide ammunition for critics). Much behavior even of human beings is not very “cognitive” at all.

A related objection is that the term cognition derives from an old view of psychology that includes two other divisions of the soul: conation (the will—

motivation), and emotion. It is far from certain that this division corresponds to a natural line of behavioral or physiological fracture. Motivational and cogni-tive elements are intertwined in the behavior of simple organisms and are not certainly separable even in human behavior. And the brain does not divide up neatly into three areas either. But in any case, if one is willing to settle for a Kantian psychology, why not buy the whole package?

Fourth, in early days, the more thoughtful cognitivists seemed to offer a robust view of “mental life” that differs not one whit from a pragmatic behaviorism: ²²

If Watson had not been so inept as a philosopher, he might have offered behaviorism as a pragmatic theory of mind, comparable to Peirce’s prag-matic theory of meaning, James’s pragprag-matic theory of truth, and Dewey’s pragmatic theory of value. The mind—the other chap’s mind, at any rate—

is something whose existence is inferred entirely from the behavior we observe. “John is conscious” must be translated into the hypothesis, “If I

call to John, he will answer,” or, “If I stand in John’s way, he will detour round me,” and so on. In short, if I present him with such-and-such stimuli

call to John, he will answer,” or, “If I stand in John’s way, he will detour round me,” and so on. In short, if I present him with such-and-such stimuli