Reber, Arthur S.;
Implicit Learning and Tacit Knowledge: An Essay on the Cognitive Unconscious
Oxford University Press, 1996, 200 pages
ISBN 019510658X, 9780195106589
topics: | psychology | learning | sub-conscious
Laments the shift in psychology from a focus on learning to a focus on knowledge modeling, without much consideration of acquisition. That this book is more of a manifesto becomes clear in its opening line: .
I want to make it clear from the outset that this will be, in large measure, a book about learning. A decade or three ago that would not have been particularly unusual; today it is a genuine rarity. Indeed, it is curious, given the pattern of psychological research over the course of this century that the topic of learning should be so poorly represented in contemporary psychology. recent textbooks ... cover learning in the context of conditioning and animal studies and run the standard historical gamut from Pavlov to Rescorla on one (classical) hand and Thorndike to Skinner and Herrnstein on the other (instrumental or operant) hand. ... It is almost as though learning, as an area of independent study, is viewed as a historical topic. In the recent Oxford Companion to the Mind (Gregory, 1987), a general compendium of practically everything in contemporary (cognitive) psychology, the only acknowledgements of active, pure research on the topic since the 1950s are biofeedback and operant conditioning with animals. The impact of what Baars (1986) has called the "cognitive revolution in psychology" on the study of the topics of learning and conditioning has been profound. One of the consequences of this dominance by the cognitive sciences has been that learning and conditioning are now typically interpreted within a cognitive framework. There has been, however, very little influence in the other direction — that is, not much of the work on conditioning has affected workers in the cognitive sciences. Herein lies a problem. ... within an evolutionary framework, one can see learning and cognition as richly intertwining issues and not as two distinct fields with one dominating the other. The argument will be developed around the proposition that the basic principles of cognitive induction and abstraction on one hand and conditioning and associative learning on the other share a common process — the detection of covariation between events. The case for the generality of this process has been articulated by others (notably Holland, Holyoak, Nisbett, & Thagard, 1986, and Lewicki, 1986a). Glaser (1990) specifically cites Newell and Simon's (1972) arguments that the study of performance, in an information-processing format, should take precedence over examinations of learning and development. ... the Simon and Newell research program also gives formalizations a high priority and invites the building of theories of knowledge representation rather than knowledge acquisition. my focus will be on a particular kind of learning, specifically what was dubbed back in 1965 as implicit learning (Reber, 1965). Implicit learning is the acquisition of knowledge that takes place largely independently of conscious attempts to learn and largely in the absence of explicit knowledge about what was acquired. One of the core assumptions of our work has been that implicit learning is a fundamental, "root" process, one that lies at the very heart of the adaptive behavioral repertoire of every complex organism. As such, it is the concept of implicit learning that can, in principle, encompass the early work on learning and conditioning and tuck it into a more general epistemic framework along with the more contemporary analyses of learning, induction, and discovery.
The same standard sources that reveal the shifting pattern of emphasis on learning and cognition also reflect an accompanying tilt away from the classical empiricist's point of view and toward the nativist's. Along with the deemphasizing of learning, there has been a clear resurgence in nativistic theorizing exemplified most notably by the work of people such as Chomsky (1980, 1986) and Fodor (1983). There are interesting and controversial historical patterns in these shifts. It seems reasonably clear that these two paradigms are like the opposite ends of a seesaw. The flowering of empiricism typically accompanies advances in learning theory and results in a lessening of the influence of nativist thinking. Neglect of the processes of knowledge acquisition correlates with resurgences in nativistic theory and increases in the attractiveness of the presumptive a priori. ... I will be presenting arguments in support of a strong environmentalism that developed around the thesis first put forward by Dewey — that the empiricist position is the proper default position and that nativism should only be adopted when the evidence against empiricism becomes overwhelming.
First, consciousness and phenomenological awareness are recent arrivals on the phylogenetic scene. Hence, consciousness and conscious control over action must have been "built upon," as it were, deeper and more primitive processes and structures that functioned, independently of awareness. On these grounds it is assumed that the processes studied under the rubric implicit learning, operating independently of consciousness, are more primitive and basic that those that are dependent, in some measure, on consciousness and conscious control. Second, one of the standard heuristics in evolutionary theory is that phylogenetically older and more primitive structures will display telltale classes of properties different from the more recently evolved. One of these is that the structures with greater antiquity tend to be more robust and resilient, less prone to disruption of function than the newer. Therefore, we would expect to see implicit cognitive processes show greater resistance to interference from neurological insult and clinical disorder than the explicit processes. There is a large and rapidly growing literature on implicit learning and memory that supports this analysis; (Chapter 3). Third, the evolutionarily more ancient implicit functions of the cognitive unconscious should show a tighter distribution in the population - than the more recently emerging explicit and the conscious — we should expect to find fewer individual differences between people when implicit processes are in use than when explicit processes are. Fourth, there should be a reasonably clear relationship between the point along the phylogenetic tree where a particular property or function evolved and the degree to which we are conscious of its form and content. That is, we would expect to find that the more primitive a function is shown to be, the more refractory to consciousness it will be. How to measure the information available to the unconscious (Erdelyi, 1986)? [Reber assumes] "the primacy of the implicit." ... that it is actually more surprising that any function is conscious than unconscious.
Clark Hull's (1920) early work on the learning of the structure of Chinese-like ideographs identified the process of concept formation by abstraction of common elements, a process still regarded as important (Medin, 1989). Hull's characterization of concept acquisition has some interesting similarities with perceptual learning, a process that takes place largely in the absence of awareness of the rules governing perceptual displays (E. J. Gibson, 1969; J. J. Gibson, 1979). Hull's work, while it implicated variables we now regard to be of considerable importance, had little impact on the field at the time; the Gibsons' research program became extremely influential in the study of perception but had little impact on learning. Hull, of course, abandoned these "cognitive" issues and shifted his focus to the study of motivation and reinforcement — and thereby embarked on a research program that made him one of the most influential behaviorist theorists of the 1940s and 1950s. The earliest studies touching directly on the acquisition of complex information without awareness were those we carried out in the middle and late 1960s using the artificial-grammar learning procedure and a variation on the probabilitylearning experiment (Reber, 1967a, 1967b, 1969; Reber & Millward, 1965, 1968). [reviewed in considerable detail in Chapter 2] learning without awareness and incidental learning (Greenspoon, 1955; Jenkins, 1933; Thorndike & Rock, 1934): much of this clearly cognitive research was initiated by those with a behaviorist orientation. It was carried out as part of the examination of the problems of motivation and reinforcement in learning — the very issues that Hull, having left behind matters of mind, had made the central concerns of a behaviorally oriented experimental psychology. Most of this research involved the use of the repeated trial design, in which subjects were presented with complex stimuli (usually linguistic) and were differentially reinforced for making particular responses. The substantive issues surrounded the question whether subjects could show behavioral evidence of having learned something about the associations between stimuli and responses to which they had been exposed without being aware of the S-R links. (for overviews, see Eriksen, 1960, and Osgood, 1953). The early implicit learning experiments that focused specifically on the issue of unconscious acquisition of complex knowledge were run using complex, rulegoverned stimuli generated by a synthetic, semantic-free, Markovian grammar. In the typical study, subjects memorized strings of letters in the synthetic language and were later tested for their knowledge of the rules of its grammar by being asked to make decisions concerning the well-formedness of novel strings of letters (see Reber, 1967a, 1989a). Unlike the materials in the earlier work on learning without awareness and incidental learning, which consisted typically of word lists and word associations, the stimuli here were composed of unpronounceable sequences of letters whose order was determined by arbitrary rules. The use of arbitrary, semantic-free stimulus domains ensured that their underlying structures would not be known by the subjects prior to entering the laboratory. 1 There is an intriguing paradox here. As the behaviorist influence faded so did the interest in these essentially cognitive issues. Yet, the recent concerns with the cognitive unconscious actually reflect in subtle ways many of the points first introduced by behaviorists, particularly the notion that learning occurs independent of the learner's awareness of the process. This early work was also motivated by a desire to examine empirically some of the classic philosophical questions of epistemology, including the acquisition and representation of complex knowledge. Among those thinkers whose interest in epistemological issues influenced these early grammar learning studies were two social philosophers: Michael Polanyi (whose original training, interestingly, was in medicine and physical chemistry), who had argued effectively for the importance of tacit knowledge, knowledge whose origins and essential epistemic contents were simply not part of one's ordinary consciousness (Polanyi, 1958), and Friedrich von Hayek (1962), the conservative economist who had put forward some elegant and controversial speculations concerning the necessity for deep rules and other rich mental representations to be held in a kind of "supraconscious" that was not available for ordinary conscious inspection. As the research evolved, implicit learning came to be viewed as a rather general information acquisition process. By the middle 1970s it was being characterized as a situation-neutral induction process whereby complex information about any stimulus environment may be acquired largely independently of the subjects' awareness of either the process of acquisition or the knowledge base ultimately acquired. Most of this work was still being carried using the artificial grammar (AG) learning procedure, although the paradigm was now more widely used (Brooks, 1978; Gordon & Holyoak, 1983; Howard & Ballas, 1982; Morgan & Newport, 1981; McAndrews & Moscovitch, 1985; Reber & Allen, 1978; Reber, Kassin, Lewis, & Cantor, 1980; Reber & Lewis, 1977). In addition, a number of other studies were run that employed a variation on the classic probability-learning (PL) procedure in which subjects had to predict which of several events would occur when the events followed any of a variety of probabilistic sequences (Millward & Reber, 1968, 1972; Reber & Millward, 1965, 1968, 1971). [Chapter 2]
One of the unspoken (implicit?) elements of the period during which the early implicit learning work was being carried out was that humans are rational and logical and they reach conclusions and make decisions based on coherent patterns of reflection and analysis. At least this was the general point of view within the early decades of the nascent field of cognitive psychology (see Baars, 1986, for a cogent history of this era). During the 1970s, however, it became increasingly apparent that people do not typically solve problems, make decisions, or reach conclusions using the kinds of standard, conscious, and rational processes that they were more-or-less assumed to be using. People appear to be, generally speaking, arational. It is not so much that we act in ways that do violence to rationality — although history shows no shortage of examples of such. The important insight was that, when people were observed making choices and solving problems of interesting complexity, the rational and the logical elements were often missing. It was not so much that decisions were being made that were irrational, it was rather that decisions were being made on the basis of processes that simply failed to take into consideration rational elements. Moreover, importantly, people often did not seem to know what they knew nor what information it was that they had based their problem solving or decision making on. This theme developed from a number of interdependent approaches to the study of human judgment, most significantly the work of Kahnemann and Tversky, of Richard Nisbett and his co-workers, and of Ellen Langer and her colleagues. Kahneman and Tversky, in a now-classic series of studies, showed that issues of rationality and logic were largely independent of decision making and were often "replaced" by less than optimal heuristics. Often, these nonoptimal cognitive operations were displayed in the very contexts where one would imagine them to be most compellingly employed — for example, the statistician making a decision that violated Bayesian principles or a physician making inappropriate choices in triage-type settings (see the various contributions in Kahneman, Slovic, & Tversky, 1982, for an overview of this work). An additional avenue of research emerged from social psychologists, who were examining how people made and justified "real-world" decisions. The work of Nisbett and his colleagues at the University of Michigan (Nisbett & Ross, 1980; Nisbett & Wilson, 1977) was particularly important. They directly addressed the notion that there were important cognitive lacunae between the (explicit) knowledge that we thought we used to make decisions and control choices, the (implicit) knowledge we actually used, and our differential capacities to articulate these kinds of knowledge. In a related series of studies, Ellen Langer and her colleagues at Harvard showed that people frequently functioned in ways that were, to use their term, mindless (Langer, 1978; Langer, Blank, & Chanowitz, 1978). In situations where people appeared to be acting according to explicit and consciously developed inferences they were, in fact, drawing on implicit knowledge systems about which they had little or no awareness. Under such circumstances, people provided justifications for their behavior that were clearly at variance with what they had actually done. A substantial literature has grown in this allied area of social cognition over the past two decades. See Wegner and Vallacher (1977) for an overview of the earlier work, and the various contributions in Uleman and Bargh (1989) for the more recent advances. ...the phenomenon of implicit learning [was then tested] in an increasingly widening range of empirical settings. Lewicki and his colleagues at the University of Warsaw and later at University of Tulsa, working initially from the point of view of social psychology and personality theory, reported results in a series of experiments that strongly paralleled those from the early synthetic language and probability learning studies. They found implicit acquisition of often extremely complex forms of information in experiments ranging from those on the perception of rule-governed spatial locations of stimuli (Lewicki, Czyzewska, & Hoffman, 1987; Lewicki, Hill, & Bizot, 1988; see also here, Nissen & Bullemer, 1987; Stadler, 1989) to the processing of social information and personality characteristics (Lewicki, 1986a), and the development of self-perpetuating biases for coding information about social situations and personality characteristics of target persons (Lewicki, Hill, & Sasaki, 1989). At Oxford University, Berry, Broadbent, and their colleagues discovered similar patterns of acquisition of covert knowledge in an extended series of experiments that explored how individuals developed the capacity to control complex environments, such as a simulated production plant or a socially interactive "computer person" (Berry & Broadbent, 1984, 1987, 1988; Broadbent & Aston, 1978; Broadbent, FitzGerald, & Broadbent, 1986). Similar findings by Mathews and his co-workers (Mathews, Buss, Chinn, & Stanley, 1988; Mathews, Buss, Stanley, Blanchard-Fields, Cho, & Druhan, 1989; Stanley, Mathews, Buss, & Kotler-Cope, 1989) extended and refined these conclusions.
By the middle of the nineteenth century, it was well understood that many complex perceptual processes were dependent on operations that lay outside of consciousness. The most effective proponent of this position was Helmholtz, as represented in his doctrine that perception was dependent on the process of "unconscious inference" (Helmholtz, 1867). Helmholtz's arguments, along with similar points of view espoused by Carpenter (1874), Hering (1920), and Ebbinghaus (1885), were sufficiently strong that even such recalcitrants as James, who had referred disparagingly to such proposals as "mind-stuff theory," were ready to admit that automatic and unconscious encoding existed (James, 1890, chapter 11). After a hiatus of some decades, the relevance of these varieties of nonconscious encoding processes became apparent with the development of interest in the related problems of automaticity and procedural knowledge. The interest in automaticity grew out of the work of Hasher and Zacks and their colleagues (see Hasher & Zacks, 1984, for a review) that showed that such fundamental operations as encoding the frequency and location of objects and events in the environment took place automatically and largely without awareness of the encoding process. Hasher and Zacks also argued that this encoding process was a primitive and fundamental cognitive process and, as such, was relatively unaffected by variables such as age, developmental level, IQ, and affective state, which normally have considerable impact on cognitive processing. Although there has been some dispute over just how robust the specific process of frequency encod16 Implicit learning and tacit knowledge ing is, the general proposition that automatic processes are different in fundamental ways from the consciously controlled, effortful processes is generally accepted (see Kahneman & Triesman, 1984). The other line of highly influential research emerged from the work of John R. Anderson and his colleagues on procedural knowledge (see Anderson, 1976, for the early approach, and 1983, for a more developed theory). Anderson's key distinction is that between declarative knowledge, which is knowledge that we are aware of and can articulate, and procedural knowledge, which is knowledge that guides action and decision making but typically lies outside of the scope of consciousness. Anderson's view is that virtually all interesting complex human skills are acquired in a characteristic fashion. They begin with the labored, conscious, and overtly controlled (declarative) processes of the novice that gradually give way to the smooth, unconscious, and covertly controlled (procedural) processes of the expert. This ordering of processes appears, at first, to be at variance with that presented by the standard theory of implicit learning in which the initial phases of acquisition are marked by a lack of consciousness. The two approaches actually can be shown to be complementary in that the domains of applicability of the two models are different. Implicit learning theory says little or nothing about skill learning. [Chapter 4] ... like the early research on implicit learning, the work on automaticity and declarative and procedural knowledge also provided empirical support for some of the classical philosophical problems in epistemology, again strengthening our ties with philosophy. The concept of automaticity is closely aligned with Polanyi's (1958) notion of tacit knowledge, and the distinction between declarative and procedural knowledge is neatly analogous with Ryle's (1949) distinction between "knowing that" and "knowing how." This bringing of an empirical data base to bear on problems of epistemology was an important element in the development of a truly interdisciplinary cognitive science during the past decade. As I argued earlier, it forced cognitive psychologists to pay attention to issues of mind that philosophers had long regarded as critical to sensible theories of cognition. Moreover, it encouraged philosophers to begin to monitor more closely the work in cognitive psychology. As Bechtel (1988) has pointed out, philosophical inquiry has traditionally been suspicious of experimentalism, preferring rich and detailed argument about plausibility to "hard" data, which of necessity must be collected under resource limited, controlled conditions. Yet this blending of experimentalists and epistemologists has proved exciting and productive and broadened cognitive psychology's domain of investigation.
During the past decade, an influential, parallel line of research developed, one concerned with the examination of implicit memory. This approach differed from that taken to implicit learning in that questions about the acquisition of knowledge were seldom raised; the focus was on the processes of storage and retrieval of knowledge. Generally speaking, implicit memory is taken to have been displayed whenever a subject evidences, by some indirect or implicit measure of performance, that there was a memorial residue of an earlier experience in the absence of any comparable phenomenological sense of the previous experience. Interest in unconscious memory has, perhaps not surprisingly, a long history (see Schacter, 1987). http://www.cog.brown.edu/courses/cg195/pdf_files/fall06/Schacter1987.pdf It ranges from various early philosophical treatments of memory such as Descartes' pre-Freudian speculation (cited by Perry & Laurence, 1984) that unpleasant early life experiences affected one's adult life, even though there were no conscious memories of the episodes, to numerous explorations of psychodynamically oriented scholars such as Janet, Bergson, and Freud (see Ellenberger, 1970, and Erdelyi, 1985).
Among the first experiments ... in subliminal perception studies the stimuli were presented using any of a number of techniques designed to ensure that the material was not consciously encoded, including tachistoscopic [very fast] presentation, masking, degrading, shadowing, and parafoveal presentation. For example, in what has become a classic experimental procedure, subjects are shown rapidly presented, masked visual displays designed to make it unlikely that they will be able to determine their identity. Under such conditions subjects have been shown to exhibit the memorial residue of presented material by their choices in semantic or preference tasks even though they are not aware of having been presented with the material and cannot select the "old" stimuli on two-alternative, forced-choice recognition tasks (Kunst-Wilson & Zajonc, 1980; Marcel, 1983a; Seamon, Brody, & Kauff, 1983; Seamon, Marsh, & Brady, 1984).
Many of these experiments involved the use of the repetition-priming effect in which there is increased facilitation in the processing of stimulus material presented previously, independent of explicit memory for that material. This general finding has been reported in a large number of settings, including the lexical-decision task (Scarborough, Gerard, & Cortese, 1979), perceptual identification (Jacoby & Dallas, 1981), and word-stem completion tasks (Graf, Mandler, & Haden, 1982; Tulving, Schacter, & Stark, 1982). The interest in priming and implicit memory has become rather intense in recent years because the evidence for implicit memory using these procedures is so strong (see any number of the contributions in Lewandowsky, Dunn, & Kirsner, 1989).
sergei korsakoff 1889 - in one session gives S mild shocks. patient has no conscious memory, but when he comes again with the shock apparatus, S accuses him of trying to give a electric shock edouard claparede 1911: patient refuses to shake his hand, saying that people have been known to carry a pin.
H.M.: densely amnesic patient with bilateral excision of the medial temporal region, including removal of the hippocampal gyrus, the amygdala, and two-thirds of the hippocampus. His anterograde amnesia was so profound that a half hour after eating lunch he could not recall what he had eaten or even if he had eaten at all. Yet H. M. has showed nearly normal abilities in sensorimotor skills such as mirror drawing and tactile maze learning (Milner, Corkin, & Teuber, 1968). Improved performance on such tasks clearly requires some memorial residue of the previous experiences. However, he had no skill learning when the task requires conscious cognitive processes, such as hypothesis testing. On the Tower of Hanoi problem, if care is taken not to prompt him (Gabrieli, Keane, & Corkin, 1987), H. M. shows little or no improvement in performance over trials. Gabrieli, J. D. E., Keane, M. M., & Corkin, S. (1987). Acquisition of problem-solving skills in global amnesia. Society for Neurosciences Abstracts, 13 1455.