Learning theory and the cognitive revolution, 1961–1971: a personal perspective
DOI:10.1093/acprof:oso/9780199228768.003.0004
Abstract and Keywords
This chapter describes developments in learning theory and cognitive psychology during the period from 1961 to 1971. When the 1960s began, behaviourism dominated American psychology and this entailed accepting the notion that the study of learning is central to psychology because every important aspect of human psychology is overwhelmingly determined by environmental events during an individual's life. However, these views were not widely held in Great Britain. This chapter describes the cognitive revolution that rejected behaviourism and marginalized the study of learning.
Keywords: leaning theory, cognitive psychology, behaviourism, cognitive revolution, United States, Great Britain
This chapter is about the ten years after I first started to study psychology, a period of unusually rapid change in the fields in which I became involved. When the 1960s began, behaviourism dominated American psychology. This entailed accepting that the study of learning is central to psychology because every important aspect of human psychology is overwhelmingly determined by environmental events during an individual's life. Furthermore, it was very widely believed—and not just by researchers working in rat laboratories—that fundamental principles of learning could be discovered only by carrying out conditioning experiments with animals.
Such views were not widely held in the UK. In particular, in Cambridge, where I first studied psychology, the 1960s continued with excitement over information theory (see Chapters 2 and 19). This was one of several developments that led to a cognitive revolution that rejected behaviourism and marginalized the study of learning. Nevertheless, during this period research in animal learning produced important discoveries and theories that have had a tremendous influence ever since.
As a PhD student at Harvard, I was exposed both to radical behaviourism and to the new US-based cognitive science. The first ‘Cognitive Center’ was founded by Bruner and Miller at Harvard in 1960, and at the other end of town a number of linguists, philosophers, and psychologists at the Massachusetts Institute of Technology (MIT) were exploring the implications of Chomsky's theories of grammar (Boden 2006).
Cambridge
As a teenager I once mentioned to a friend the passing idea of becoming a psychologist. She found this hilarious. Neither of us really had much idea of what psychology was about or what psychologists did—perhaps only hazy ideas of (p.38) listening carefully to what mentally ill people had to say and then ‘analysing them’. (It was probably the thought of me as an attentive listener that was unconvincing.) Until the late 1960s few people in the UK knew much about psychology, and there were few universities where one could obtain a degree in the subject. I was lucky that, after choosing to become an engineer, I went to Cambridge, because—unlike most British universities—it allowed unusual flexibility in the choice of subjects. Discovery that engineering was not for me was followed by an exciting year of philosophy and then a year of psychology.
This last change followed lectures by Alan Watson on experimental psychology given especially for philosophy students. I found the content fresh and exciting, especially his summary of current debates over Hullian learning theory. His lectures inspired me both to transfer to psychology and to obtain a copy of Broadbent's (1961) book, Behaviour. After switching to studying psychology full time I discovered Broadbent's more widely known and extremely influential book, Perception and Communication (Broadbent 1958). He and colleagues from the Applied Psychology Unit gave lectures on short-term memory and selective attention that made the topics fascinating. In contrast, Alan Watson's extremely detailed treatment of learning theory was disappointingly dull—but at least his lectures were well organized.
Most lectures series were a shambles, partly reflecting the Cambridge belief that lectures were not important. What really mattered were the weekly 1-hour one-on-one meetings with a supervisor and the essays to be written for these intensive meetings. Good supervisors would set their students essays on potential exam topics that no lecturer may have even touched on. As a result, I read, for example, Tinbergen's (1951) A Study of Instinct, together with later books on the relationship between learning and instinct (e.g. Thorpe and Zangwill 1961). The nature of imprinting and of birdsong learning were of particular interest at that time, and in Cambridge interaction between psychologists and ethologists meant that questions, for example as to whether Hullian learning theory could account for various types of early learning, were likely to appear in the psychology exam.
During my final year at Cambridge Skinner came into my life. Throughout the 1940s and 1950s the neo-behaviourism of Hull and Spence had maintained a strong intellectual and political grip on American psychology. By 1960, there were growing doubts about both the core theoretical assumptions S-R-reinforcement theory made about rat behaviour and its relevance to many important aspects of human psychology (Leahey 2004). In its place Skinner's radical behaviourism had become increasingly attractive, especially following publication of Science and Human Behavior (Skinner 1953), with its analysis of everyday behaviour in terms of operant conditioning principles (p.39) and its promise of solutions to a wide range of individual and social problems. At Cambridge the general attitude, as reflected, for example, by Watson's lectures and Broadbent's (1961) book, was that Hull-Spence theory was an intellectually exciting, but ultimately inadequate, account of learned behaviour in non-human animals, one that had some, albeit limited, relevance to aspects of human behaviour. Skinner's ideas were less well regarded. I do not believe that his analysis of language and communication in Verbal Behavior (Skinner 1957)—an area where few behaviourists had previously dared venture—was even noted. Instead, I became interested in Skinner's applications of reinforcement principles, particularly in relation to education and his invention of the ‘teaching machine’.
Skinner came to Cambridge to give a lecture. At the end there was an opportunity for a mere undergraduate like myself to ask a question. ‘How do you explain latent learning?’, I asked. It was clearly not a topic of much interest to Skinner and he brushed my question aside. On the other hand, he was attentive after the lecture when I told him about my interest in teaching machines. He described a new research centre on ‘programmed instruction’ and encouraged me to apply to Harvard.
Harvard
Graduate students entering Harvard's Department of Experimental Psychology in September 1963 were made well aware of a tradition stretching back to Wundt. Having studied in Leipzig with Wundt, Titchener set up an early American department of psychology at Cornell University. His student, Boring, a historian of psychology (e.g. Boring 1950), moved to Harvard, where his student, Stevens, known for his analysis of measurement in psychology (Stevens 1939) and for his psychophysical Power Law (Stevens 1961), was very much in control when I arrived. One aspect of this tradition was the ‘pro-seminar’, a weekly meeting that all first-year graduate students had to attend and that was dominated by the tenured faculty, notably Stevens and Herrnstein. Their highly atheoretical approach contrasted with presentations by Bruner and discussions, led by Norman, of the entirely new idea that computers might be used to model psychological processes.
Another requirement was that all graduate students complete a ‘practicum’, an experiment in some area other than that of their intended PhD topic. The admirable intention was that our research focus did not become too narrow too soon. It also had the benefit of providing the faculty with unpaid labour to run their experiments. Stevens was particularly keen on this system, and a large number of students ended up—as I did—completing some kind of psychophysical experiment. Stevens was an enthusiastic skier, obsessed by the (p.40) virtues of short skis. As a result, learning to ski on short skis—a very sensible way to begin—combined with learning to run and report psychophysical experiments formed part of many a graduate student's first-year experience.
A further tradition was to require that students be able to read articles in either German or French. One of my few early contacts with Skinner was when he conducted my oral exam in French. Skinner had a research fellowship that allowed him to spend most of his time at home where he had a recorder attached to his typewriter in order to measure his output rate. He was rarely seen in the department and had no contact that I ever detected with the Center of Programmed Instruction. I worked there as a part-time research assistant until decreasing faith in the future of teaching machines helped to turn me from experiments on learning by undergraduates to learning by pigeons.
Only a few students continued with psychophysics after their practicum. Most returned to the subjects that had attracted them to Harvard in the first place. Many were there because of George Miller. Like Broadbent, Miller had been an early enthusiast for the application of information theory (e.g. Miller 1956). He had written a book on Language and Communication (Miller 1951) that promoted these ideas, and also endorsed the operant analysis of language acquisition and performance developed in Skinner's (1957) Verbal Behavior. Following Chomsky's (1959) devastating review of Skinner's book and caustic dismissal of behaviourist psychology in general, Miller had undergone a St Paul-like conversion, his road to Damascus being in California (Miller et al.1960). On his return to Harvard—now bearded, leading Herrnstein to comment that Miller had left Occam's Razor behind—Miller almost single-handedly brought back to life the Lazarus of psycholinguistics, an area of research that had remained effectively unvisited by a generation of American psychologists.
By 1963 this enterprise was in full swing. Miller's students tested whether Chomsky's syntactic analysis could be applied to understanding and remembering sentences. For example, I helped with an experiment to test whether the difference in deep structure between sentences like ‘John is eager to please’ and ‘Jack is easy to please’ would affect how well they were remembered. (We found that logical subjects, e.g. John, were more effective prompts than logical objects, e.g. Jack; Blumenthal and Boakes 1967.) Meanwhile I attended courses given by Brown, whose studies of the development of syntactic knowledge—for example at what stage did children acquire the rule for generating plurals? (Brown and Bellugi 1964)—stimulated an enormous growth of such research. Another converted Chomskian gave equally challenging lectures on the neuropsychology of language (Lenneberg 1967).
(p.41) There was ongoing debate between the psycholinguistics students and those who had come to Harvard because of Skinner. By the 1960s, several American psychology departments, most notably that at Columbia University, had become strongly influenced by Skinner's behaviourism and in turn imbued many of their undergraduates with an almost religious-like dedication to these ideas. Some entered Harvard and were disappointed to find that Skinner was almost inaccessible. Instead they came under the powerful influence of Herrnstein.
With hindsight I can see that I arrived at a turning point in the history of operant conditioning. The research programme that Skinner had initiated in the 1930s (Skinner 1938), one that explored the properties of operant conditioning in an atheoretical way, had run its course. It had been very fruitful in the 1950s, but arguably the last important studies of this kind were those on discrimination learning undertaken at Harvard before I arrived (Honig 1966). For example, Terrace (1966) had just left after completing for his PhD an influential set of experiments on errorless learning in pigeons. After Skinner, Herrnstein was the only researcher in operant conditioning who remained at Harvard and eventually obtained a tenured position. During my time there he ran the operant laboratory and was adviser to the large group of graduate students working with rats and pigeons. The mid 1960s happened to be a particular creative period for Herrnstein; his research from that period has had an enormous influence on subsequent generations of Skinnerian researchers (Boakes 2002). Most notably, he and his students ran experiments that gave pigeons two response keys to peck at and measured their choice between different reinforcement schedules. This research gave rise to the ‘Matching Law’ which, in its original simple form, claimed that the relative rates with which animals made different responses matched the relative rates of reinforcement obtained by those responses (Herrnstein 1970).
Herrnstein maintained Skinner's dismissal of all attempts to infer any underlying processes, but in other respects his approach was quite different. Unusually for a behaviourist, he advocated strong hereditarian views on human differences. In addition, unlike Skinner, he displayed little interest in applications of operant conditioning and believed that science was about establishing quantitative relationships between observable events. In this last respect he resembled Stevens, who had been at least as important a mentor to Herrnstein as had Skinner. In the 1960s the term ‘Law of Science’ already appeared old fashioned. The Matching Law seemed to me an attempt at a behavioural equivalent to Stevens' Power Law and, probably because I had been immunized against the appeal of purely quantitative descriptions by my Cambridge training, I was never convinced that either provided much further understanding of either behaviour or perception.
(p.42) Two visitors stand out from my final year at Harvard. One was a student from the University of Pennsylvania named Rescorla, who gave a brilliant talk about his research on Pavlovian conditioning. His whole approach was so alien to that of the prevailing intellectual climate at Harvard that it was no surprise that he ended up at Yale instead. The other was an Englishman, Sutherland, who had recently left Oxford to establish experimental psychology at the new University of Sussex, but was currently on sabbatical at MIT. We spent an afternoon discussing my research and then he invited me to a party that evening. Unusually powerful Manhattans were served and I spent most of the evening vomiting in a toilet. Later I learned that this day had served as an interview process and that I was appointed as an assistant lecturer at Sussex, despite my display of low tolerance to alcohol.
Sussex
Psychology departments began to proliferate in the UK during the 1960s. Appointment of academic staff was generally made in a similar manner to the choice of animals for a small seaside zoo, namely, picking representative animals from widely disparate species (P. Rabbitt, open letter to Jerry Bruner). In contrast, Sutherland's policy for experimental psychology at Sussex was to concentrate on the few areas of psychology he considered to have intellectual merit. These included his core interest, developed at Oxford, in animal discrimination learning, together with mechanisms of selective attention ranging from the level of behaviour to physiological processes in perception; recent discoveries on cortical organization (e.g. Hubel and Wiesel 1962) were particularly exciting in this respect. Sutherland's visit to MIT added psycholinguis-tics and artificial intelligence to his set of enthusiasms. As a result, I arrived at Sussex in October 1966 to find a very active animal laboratory in which all three of my new colleagues were carrying out research and, as the Laboratory of Experimental Psychology expanded over the next few years, we were joined by other biologically minded researchers and an increasing number of cognitive scientists.
Discrimination learning was the major topic for our rat and pigeon experiments. Research assistants trained rats in a jumping stand to test the attentional theory of learning that Sutherland was developing in long range collaboration with his former student, Mackintosh (Sutherland and Mackintosh 1971). Others of us used Skinner boxes to analyse effects such as behavioural contrast—an increase in responding to a reinforced stimulus produced by introducing a second, non-reinforced stimulus—that had been uncovered by Terrace and other Harvard students just before my time there.
(p.43) The latter research was transformed by the discovery of autoshaping in Jenkins' laboratory at McMaster University. After decades of experiments in which the delivery of grain had been made contingent on pigeons pecking at an illuminated plastic response key, it was found that such pecking could be maintained even when it had no effect on the delivery of grain. All that was needed was a classical conditioning contingency whereby a light appearing for ten seconds or so served as a reliable signal for the delivery of grain (Brown and Jenkins 1968). A dramatic demonstration that autoshaped pecking was a classically conditioned response equivalent to salivation by Pavlov's dogs was that, under an ‘omission’ contingency, pigeons would persist in pecking even when such responses prevented the arrival of grain (Williams and Williams 1969). These results opened up the possibility that a large range of puzzling phenomena could be explained in terms of interactions between instrumental and classical conditioning. Halliday and I began to test the idea that behavioural contrast could be explained in this way (e.g. Halliday and Boakes 1972).
Autoshaping challenged Pavlov's principle of ‘equipotentiality’; put simply, it does not matter what kind of stimulus, response, reinforcer, or species a researcher uses in an experiment, the same basic principles of learning will apply. This assumption was implicit in the agreement between Hull and Tolman—who differed in most other respects—that understanding the behaviour of a rat at the choice point in a maze would illuminate most of the interesting problems in psychology. It was behind Skinner's simple generalizations, for example from the behaviour of pigeons to that of children in a classroom. Equipotentiality was increasingly attacked in the 1960s by ethologically influenced critics, who cited findings such as autoshaping as examples of‘constraints on learning’.
The most substantial body of research of this kind, and the one that had most lasting impact, began in the 1950s at a US Navy research station in Hawaii (e.g. Garcia et al. 1955). Garcia's experiments examined the ways in which exposure to gamma radiation led rats to avoid their food or flavoured solutions. As such, they were quite different from the standard maze or Skinner box experiments of that era. Moreover, Garcia's conclusions and data were treated with great scepticism by journal editors, so that his early articles did not appear in mainstream journals (Garcia 1981; Revusky 1977). Consequently this research was ignored for a decade or more. Despite my relatively intensive exposure to learning theory at Cambridge, Harvard, and Sussex, I did not come across taste-aversion learning until the late 1960s. By then the data clearly indicated that such learning had very different properties from standard conditioning. Most notably, nausea-based learning displays a high degree of stimulus selectivity so that a rat can learn in a single trial to (p.44) associate a taste with sickness, but not to associate a sound or visual signal with sickness, whereas this last kind of stimulus can be easily associated with a shock, while tastes are associated with shocks only with great difficulty (Garcia & Koelling 1966).
Equipotentiality could be taken to imply that, given appropriate training, an ape might acquire some kind of language. A few early attempts to provide such training had been unsuccessful. This provided support for the Chomskian claim that language acquisition is a uniquely human trait and no amount of research on learning in non-human animals can help towards understanding this ability. In this context, the visit to Sussex of a husband and wife team, the Gardners, who had trained a chimpanzee, Washoe, to produce and respond appropriately to a hundred signs or more, was particularly exciting. This success was sensational and, even more so, was their claim that Washoe had started to string signs together in a meaningful way and so display the ‘creativity’ that Chomsky had emphasized as a hallmark of human language (Gardner and Gardner 1969). Washoe inspired other projects of this kind over the next few years, but later the enterprise suffered a major blow from Terrace's (1979) discovery that his team's chimpanzee, Nim Chimpsky, produced interesting strings of signs only when unintentionally cued by a trainer.
Terrace was on sabbatical at Sussex, along with Jenkins and Neisser, who had recently published the first book called Cognitive Psychology (Neisser 1967), at a time when Sutherland organized a weekly seminar to discuss a preprint he had just received, the first paper by Rescorla and Wagner (1972) to describe their theory. All of the now six academic staff attended, along with the visitors and the particularly talented group of postgraduate students; it was clear to everyone that the Rescorla-Wagner theory was of major importance. It provided a model that made sense of data indicating that, in classical conditioning, cues compete for ‘strength’ of association with some outcome, and how well a cue competes depends on its relative validity as a signal of the outcome. Impressively, the model made some interesting new predictions that Rescorla and Wagner had started to confirm. We had already heard about Kamin's (1968) blocking effect and his conclusion that conditioning took place to the extent that an outcome was surprising. (The use of terms such as ‘surprise’ and ‘expectancy’ in an article on conditioning was still a little shocking; they would never have been allowed by the behaviourists who so very recently had dominated learning theory.) What Rescorla and Wagner did was to turn these loose ideas into a precise model that also captured various results from their own separate laboratories, including data providing the first real insights into conditioned inhibition since Pavlov's time (e.g. Rescorla and LoLordo 1965).
(p.45) This seminar and a conference on inhibition and learning that Halliday and I organized later in 1971 (Boakes and Halliday 1972) were followed by nine years of exciting learning research at Sussex. It was a period when cognitive research also flourished there. Things were different elsewhere, as I began to learn. Research within the Skinnerian tradition, mainly following Herrnstein's lead, became even more isolated from the rest of psychology. Skinner had never replied to Chomsky's attack. The latter was widely perceived as having demolished the intellectual foundations of radical behaviourism, and the several practical achievements of behaviourist psychology were modest compared with the early ambitious claims. As belief in the importance of animal research declined, so did the number of psychology departments that included increasingly expensive animal laboratories. How learning theory later revived is another story (Boakes, in preparation).
No comments:
Post a Comment