Learning set

1. A learned tendency to follow a particular cognitive strategy in response to a particular type of *stimulus.

2. Learning to learn.

3. Progressive improvement in the rate of learning of successive object discrimination problems of a given type, culminating in single-trial learning of novel problems of that type.

The roots of 'learning set' can be traced back to the beginning of the twentieth century, to a group of psychologists at Wurzburg University, Germany, known collectively as 'The Wurzburg School'. They pioneered experimental approaches to thought processes (Boring 1950). In the course of their investigation they noted that with proper preparation of a "subject for a mental task, upon the presentation of the stimulus the thoughts of the subject would run off automatically to perform the task. This was taken to indicate that the particular task and the instructions that have preceded it ('Aufgabe' in German) imposed certain constraints on "attention and thought, yielding a tendency to use a particular mental strategy. This learned tendency to respond in a particular manner to a type of stimulus was termed 'Einstellung' (German for 'set', 'attitude').

For a while, the concept of a cognitive or mental 'set' had been used rather liberally by multiple schools in psychology, dealing with topics as diverse as perception, conditioning, volition, or neurosis. This cast some doubts on the usefulness of'set', leading Gibson (1941) to comment that 'The concept of set or attitude is a nearly universal one in psychological thinking despite the fact that the underlying meaning is indefinite, the terminology chaotic, and the usage by psychologists highly individualistic'. It was Harlow (1949) who rein-vigorated the concept by focusing on one aspect of it, which he termed 'learning set', and devising "assays to quantify it. Harlow felt uneasy with the fact that many in the contemporary field of animal learning study their subjects in short, isolated learning episodes only. He called this approach the 'Blitzkrieg' technique. Animals, so ran his argument, do not learn about the world merely by taking isolated snapshots of it; they are expected to benefit from "transfer from one situation to another and ultimately form some "generalizations and predictive 'hypotheses' that facilitate the proper response to familiar types of situations. This, at least, is the way we humans behave, and there is no reason to assume "a priori that other species are radically different.

What Harlow did was to place "monkeys in a special test enclosure ('Wisconsin general test apparatus', "delay task), and present them with a series of visual discrimination food-reward problems. In each problem, the monkey was required to choose on a front tray the rewarded one of two different objects. Different pairs of objects were used for each individual problem, and the left-right position of the rewarded object was varied in an overall balanced manner. Each of the problems was run for multiple trials. At the beginning, it took the monkeys many trials to respond correctly. But then, a remarkable increase was noted in "performance, reaching after a while > 95% successful discrimination on the second trial of each new problem (Harlow 1949). Harlow suggested that the remarkable level of performance was achieved because the monkey 'learned to learn'. He defined this 'metalearning' as the formation of a 'learning set'. Harlow further found that the monkey could acquire a set that allowed for swift alternation of the response tendency. This was done by using 'discrimination reversal' problems. In this type of problem, the subject was first run on a discrimination problem for a number of trials, but then the reward value of the stimuli was reversed for another series of trials, i.e. the stimulus previously correct was made incorrect and vice versa. After experiencing a number of problems of this type, the monkey learned to respond correctly already in the second trial on the discrimination reversal type of problem. It is noteworthy that in spite of the fact that the discrimination reversal task might have been expected to be at least as difficult, the learning set was actually formed more rapidly (Figure 41); this was attributed to interproblem "transfer from the earlier discrimination training.

The experimental procedure developed by Harlow, namely measuring the improvement in performance on successive discrimination problems, gave rise to a restricted operational definition of 'learning set', which is frequently used in the literature (definition 3, 'discrimination learning set'). However, definitions 1 and 2 are more comprehensive and capture the essence of the concept better. For the change in behaviour noted by Harlow and by many investigators after him (Hodos 1970) could be safely construed as manifestation of the ability of the organism to acquire a response strategy, and not only the particular stimulus-"reinforcer association in an isolated task ("instrumental conditioning). It is important to appreciate that the acquisition of a learning set is conventionally measured in terms of the number of problems solved before the subject has acquired the postulated response strategy, just as the formation of "habit is measured in terms of the number


Fig. 41 The acquisition of learning sets. Learning curves of discrimination (closed circles) and discrimination reversal problems (open circles) in the monkey are plotted as responses on trial 2 in each problem as a function of successive groups of problems. The discrimination reversal learning set was formed more rapidly, probably due to interproblem training *transfer (reproduced from Harlow 1949).


Fig. 41 The acquisition of learning sets. Learning curves of discrimination (closed circles) and discrimination reversal problems (open circles) in the monkey are plotted as responses on trial 2 in each problem as a function of successive groups of problems. The discrimination reversal learning set was formed more rapidly, probably due to interproblem training *transfer (reproduced from Harlow 1949).

of trials required to acquire the specific habit. Hence the former yields an interproblem 'learning curve' ("learning), whereas the latter an intraproblem learning curve. It is also noteworthy that the formation of a learning set involves gradual improvement in performance, and this is different from "insight, in which an hypothesis or concept form abruptly after a period of latency. It is not a grave sin, though, to refer to the acquisition of a learning set as sluggish insight.

What is it that is actually acquired in a learning set? Harlow himself entertained the idea that the subject acquires the "skill of eliminating inappropriate response tendencies ('error factors', Harlow 1959). Others have emphasized that in forming a learning set, the subject acquires a conceptual understanding of the type of problem (Restle 1958; Levine 1959; Schusterman 1962; Warren 1966). In the case of the types of discrimination problems described above (but not necessarily in other types of problems), the evidence points to the acquisition of an hypothesis, or abstract "algorithm, of the type 'win-stay, lose-shift': the subject remembers the outcome of the preceding trial as being either rewarded ('win') or unrewarded ('lose'), and selects on the next trial the same "cue if previously rewarded ('win-stay') or the alternative cue if unrewarded ('lose-shift').1 Note that this requires the use of "working memory, and is expected to be sensitive to intertrial interval in each problem (e.g. Kamil and Mauldin 1975).

As the acquisition of a learning set implies mastering some type or another of abstract rules, it was soon adapted as an intelligence test in comparative animal psychology. For the purpose of comparison, a useful convention is to measure the mean per cent correct on trial 2 of a given problem as a function of the number of problems experienced of the same type (Figure 41; success on trial 2 reflects single trial learning because trial 1 is the instruction trial). The idea is that the more problems required to form a set, the duller is the brain. This type of "assay has been applied to estimate the difference in intelligence in phylogeny, ontogeny, and among individual members of a species, even in "Homo sapiens (Hayes et al. 1953; Harlow 1959; Warren 1966; Doty et al. 1967; Hodos 1970). Learning sets have also been used as "model behaviours to explore the role of identified brain organs, such as "cortex, "hippocampus, thalamus and striatum, in advanced learning capabilities in various species (Riopelle et al. 1953; Chow 1954; Staubli et al. 1984; Eichenbaum et al. 1986; Lu and Slotnick 1990; Tremblay et al. 1998). However, several caveats are appropriate. First, it is advisable to keep "Ockham's razor in mind, and scrutinize the data even if they do suggest a learning set. Over the years, some interesting debates have taken place in the scientific literature concerning the question whether a marked improvement in performance by a given species on successive presentation of a given type of problem indeed proves the formation of a bona fide learning set (Menzel and Juno 1982; Schrier and Thompson 1984; Reid and Morris 1992; Slotnick 1994). Second, the above notwithstanding, it is important to remember that species differ in the way they sense the world and in the importance they assign to different kinds of problems. It is therefore unlikely that a single type of discrimination problem will do justice to the intelligence of different species. For example, primates are visual animals, whereas for the rat the world is mostly smell and taste and touch. It is hence not surprising to discover that if learning sets are at all formed in the rat, "performance on series of odour discriminations rather than visual discriminations is the place to look for them (Slotnick and Katz 1974).

Selected associations: A Priori, Classic, Habit, Subject, Transfer

1For more on this and similar types of response strategies, see *habit.

Brain Training Conversion

Brain Training Conversion

Hot Tips on Keeping your Brain in Shape. Do you envy those people who always seem to be intelligent and witty? Do you ever wish you could unlock your own intelligence and wit?

Get My Free Ebook

Post a comment