Ecological Learning Theory

Graham Davey

• Publisher: Routledge
• Year: 2018
• ISBN:
  9781351371100
• Edition: 1
• Topic: Psychologie
• Subtopic: Histoire et théorie en psychologie

Originally published in 1989, this title presents a view of adaptive behaviour which integrates both evolutionary and psychological perspectives on learning. The study of learning, and in particular conditioning, had evolved in isolation from the rest of the biological sciences, and until the late 1980s had largely ignored the fact that learning processes are adaptive functions subject to the pressures of evolutionary selection. This text is designed to give a thorough insight into contemporary views of learning mechanisms, at the same time incorporating an evolutionary perspective on the function and performance of learning.

Graham Davey gives a detailed introduction to evolutionary approaches to behaviour and basic learning phenomena such as Pavlovian and instrumental conditioning. He also provides a comparative introduction to both learning and performance aspects of conditioning. He covers ecological approaches to adaptive behaviour (e.g. foraging theory), specialized learning processes such as concept formation, spatial learning, and language learning.

Innovative in its integration of ecological and evolutionary approaches with more traditional associative views of learning, the book introduces the reader to learning in a very wide variety of species other than the traditional laboratory rat and pigeon. It will be valuable to anyone with a general interest in animal behaviour, and also to those with a specific interest in learning, adaptive behaviour, and evolutionary approaches to behaviour.

Chapter one

Ecology, evolution, and learning

A hungry hamster will readily learn to dig, rear, or scrabble to acquire food, yet simply cannot learn to face-wash, scent-mark, or groom to achieve this result. An earthworm has no trouble learning to associate a taste with a hot, dry place, but cannot associate a tactile stimulus with electric shock. Cats eagerly approach and investigate localizable sounds that signal food, but such signals simply do not interest rats. Pigeons learn to peck a lighted key to avoid electric shock only with great difficulty, yet they will quickly learn a wing-flapping response to avoid shock. These examples represent some of the enigmas facing anyone attempting to construct a theory of learning; not only do different species differ in their abilities to cope with different learning tasks, but individual species also frequently exhibit paradoxical irregularities in their learning abilities. Demonstration of a learning ability in one situation (such as the ability to associate behavior with its consequences) is no indication that the animal will be able to learn such a thing in other situations. The theorist appears to be left with a heap of facts and phenomena which show no apparent species or task regularities. So what kind of order, if any, can be imposed on this literature? There are a number of often quite different approaches to this problem, some of which set out to extract general principles of learning from this melee; others take a different conceptual approach by placing emphasis on how learning abilities may have evolved as a form of behavioral adaptation. We shall discuss these different approaches in a moment. However, when setting out on a path to the understanding of learning there are clearly two distinct questions that need raising at the outset. The first asks what is learning for? The second asks how do animals learn? The first question asks why animals evolved learning abilities at all, in the sense of having to specify what selection pressures were important in the evolution of learning, and it is also concerned with what animals learn in terms of what problems they are faced with in their natural environments. The second question is concerned with the mechanisms of learning: given that an animal has to learn such-and-such in its natural environment, how does it do this? Different species may solve similar problems in completely different ways: evolution only selects for outcomes, not for specific means or processes, and it will select on the basis of the entire functional set of mechanisms available at the time (for example, some species may find Pavlovian associative mechanisms quite adequate for coping with a specific adaptive problem, whereas others may resort to instrumental learning mechanisms to solve the same functional problem).

The questions of what and how are not as theoretically distinct and separable as they have just been portrayed. In the last forty to fifty years the emphasis in learning theory has been on the question how, as exemplified by the attempts of general process theories to extract principles of learning which transcend species and tasks. These theories were addressing the question how in the sense that they attempted to specify generalities in basic psychological laws of learning such as principles of association (an instance of this is the assumption that a very wide variety of species possessed the ability to process Pavlovian associations, and the rules by which these associations were learned were relatively universal). Many writers agree that this general learning theory approach has failed, as the list of anomalies at the very beginning of this chapter tends to infer. But what should take its place? Most of the contents of this book reflect a recently expounded alternative to the general learning theory approach: namely, an approach to learning based largely on ecological and evolutionary considerations. Nevertheless, a useful and adaptable ecological learning theory should not attempt to throw out the baby with the bathwater. We should retain some of the conceptualizations of generalized learning theory as being useful to what is basically an ecological approach. This, then, is not a polemical exposition of ecological learning theory but what I hope will be a satisfactory integration of the ecological approach with some of the more durable aspects of generalized learning theory.

Ecology and learning

The tradition in learning theory has always been toward what has been labeled as the “arbitrary general learning theory approach” whose goal was to identify and describe general principles of learning that transcended both species and learning tasks. This approach arguably reached its peak in the theorizing of B. F. Skinner and other neo-behaviorists of the 1950s and 1960s. The pervasive assumption in this approach was that there were general principles of learning to be found, and once identified in one species these principles could simply be extrapolated to other — completely unrelated — species. This led to the intensive study of learning in individual species (namely, the laboratory rat and, later, the laboratory pigeon) and also fostered many of the radical behaviorist assumptions about the control and prediction of human behavior (e.g. Skinner, 1953). Examples of the so-called general principles of learning that this approach “discovered” were the principle of instrumental reinforcement (behavior is modified by its consequences), stimulus generalization, the equipotentiality of stimuli (all stimuli are equally capable of becoming Pavlovian conditioned stimuli), and the principle of temporal contiguity (associations between events will only be learned if they occur in close temporal proximity).

Subsequently, during the late 1960s and early 1970s, animal psychologists began to find exceptions to these general principles, particularly when studying species other than the laboratory rat or pigeon, in learning environments different from the Skinner-box and T-maze, and with reinforcers other than food or water. At first, the learning establishment was reluctant to acknowledge such anomalies, as was evidenced by the difficulty of some animal psychologists in getting the studies that reported these anomalies published (cf. Garcia, 1981). Such anomalies were largely considered to be the result of procedural differences or deficiencies in experimental design and control.

Nevertheless, as the catalog of these anomalies began to grow, it became clear that the traditional general laws of learning were no longer sacrosanct. At first the approach to these anomalies was simply to catalog them (e.g. Shettleworth, 1972a; Hinde and Stevenson-Hinde, 1973; Seligman and Hager, 1972), but it soon became clear that a radical rethink was necessary if these so-called “constraints on learning” were to be integrated into any predictive theoretical framework.

1 The biological boundaries approach

The first theoretical attempt to assimilate these learning anomalies alluded in very general terms to biological constraints on learning. This approach proposed that certain things could not be learned or were learned in selected ways because the animal’s biology determined this. Both ecology and evolution were alluded to in very tangential ways when the notion of “biological constraints” was being unpacked. For instance, it was considered that some kinds of learning were simply not biologically sensible and other forms of adaptation could cope more efficiently with the animal’s needs. This type of biological constraint was used in an attempt to explain the great difficulties involved in teaching rats to avoid electric shock by simply pressing a lever. Bolles (1970) suggested that, in the wild, survival is too urgent for an animal to spend time learning to avoid aversive or life-threatening situations, and it therefore had to have a built-in set of defense reactions that it could utilize in response to threatening situations. This gave rise to Bolles’s species-specific defense reactions (SSDR) theory of avoidance learning (see Chapter 7, pp. 223–4), and provided a rather rough account of how evolutionary pressures might biologically constrain some general learning processes. Some other theorists decided that evolution must make some kinds of learning easier than others and this could go some way to explaining many selective association effects. Seligman (1970) labeled such predispositions prepared learning, and he proposed a continuum of preparedness onto which all kinds of learning could be graded. For instance, taste aversion learning (TAL) in rats is, for Seligman, an example of prepared learning; when rats are subjected to gastric illness after being fed a particular food, they subsequently associate only the taste of the food with the illness and not any other audiovisual properties it possesses (Garcia and Koelling, 1966; see Chapter 6, pp. 183–190). The argument here is that making certain kinds of associations, such as between taste and poisoning, is beneficial to the survival of an omnivorous feeder like the rat and so such learning predispositions are likely to prosper and be selected for during the organism’s evolutionary development.

All of these notions related to biological constraints on learning are intuitively sensible even to someone with very little academic knowledge of animal learning. However, although the biological constraints approach has alerted us to the involvement of evolutionary and ecological factors in learning, it has not been an approach capable of fostering a contemporary integrated theory of learning. There are a number of reasons for this. First, the biological constraints view still clings unwittingly to the general processes approach. As Revusky (1977) has pointed out, the use of terms like “boundaries” and “constraints” implies that there is something (such as a general process) which is being constrained. Thus, what biological constraints approaches appear to be doing is, in the way of an addendum, specifying the factors which lead to the list of learning anomalies represented by the constraints on learning literature. Since the general process view was an unproven assumption in the first place, the biological boundaries view must also share many of its increasingly shaky assumptions.

Second, the biological boundaries approaches of the kind originally outlined by Bolles (1970) and Seligman (1970) are basically post hoc theories with either little predictive value or potentially circular definitions. For instance, according to Seligman’s view of preparedness, an association can only be labeled as “prepared” if it can be shown to be learned more rapidly than other associations. The circularity appears when one then attempts to assert that the association is learned more quickly because it is “prepared.” Clearly, to break this circularity one would need some good independent criteria for determining what associations would have been selected for in the animal’s evolutionary past. Unfortunately, such criteria are difficult to formulate and it is often very easy to conjure up plausible adaptive scenarios for almost any behavioral characteristic of an organism, making intuitive guesses as to what might or might not be selected for rather unhelpful. There are some useful techniques for attempting to identify the selection pressures involved in the evolution of particular learning abilities, but they require a detailed knowledge of the function that the learning serves and comprehensive cross-species comparisons (cf. Cullen, 1957; Hailman, 1965, 1976). Theories such as preparedness have failed to appreciate the need for such knowledge.

Third, biological constraints explanations have generally been couched in the absence of relevant ethological information on the behavior and lifestyle of the species concerned. For instance, the SSDR hypothesis of avoidance learning proposed by Bolles is loosely based around the “flight-fight-fright” defense reactions of the rat. Yet it is not clear how this hypothesis might apply to other species with clearly different defensive strategies to the rat, nor what defense reactions will be elicited in what situations and whether these reactions exist in a hierarchical or parallel framework. This makes the SSDR hypothesis relatively unpredictive beyond its basic assertion that only SSDRs will be learned as effective avoidance responses — and this is only helpful if one has clear criteria for identifying a behavior as an SSDR (for instance, is shaking with fear an SSDR?). These criteria are not outlined in the hypothesis, and clearly rely on an understanding of the functional interaction between the behavior and the animal’s natural environment.

2 The comparative approach

When we explore beyond the bounds of simple nonassociative learning, species clearly differ in the relative complexity of their behavior and, presumably, in their abilities to cope with differing adaptive problems. Traditionally, the study of these differences gave rise to the comparative approach to learning which attempted to compare differing species, differing orders (rats, monkeys, dogs, and so on) and differing classes (such as fish, birds, and mammals) according to some arbitrarily defined scale of learning ability or intelligence. Normally, this involved comparing species’ performance on a variety of learning tasks of differing complexity and constructing a hypothetical scale of intelligence or learning ability on the basis of the results.

One example of this approach was that employed by Bitterman (1965, 1975). He used the performance of species in a variety of learning paradigms to classify each species as either rat-like or fish-like in its learning abilities. However, while it may seem intuitively reasonable to rank species according to their performance on a number of tasks which require varying degrees of adaptive sophistication, this kind of approach is not particularly helpful theoretically. It fosters misleading beliefs about the evolution of learning and about the adaptive nature of learning. First, Hodos and Campbell (1969) pointed out some time ago that comparative approaches had traditionally been based on fallacious assumptions about evolutionary lineage. In comparative approaches to learning such as that espoused by Bitterman, abilities were compared in species which did not have common ancestor-descendant lines. Classifying pigeons or monkeys as either rat-like or fish-like bears no relation to an understanding of the evolution of intelligence or learning abilities, because fish did not give rise to pigeons, pigeons to rats, or rats to monkeys.

Another obvious problem with this kind of hierarchical approach to the structure of learning abilities is that such programs of study have tended to adopt a human-oriented criterion of intelligence. As a result, associative learning has tended to be regarded as lower on this scale than reversal learning, probability learning, concept formation, or language learning. Intelligence as a notion related to learning and adaptive behavior cannot be adequately assessed outside of its biological function, and how it assists the animal in securing resources in its natural environment. Thus, it cannot easily be defined a priori in this context; it needs to be related to the demands that are imposed on a particular species of animal by that species’ need to fulfill biological functions (see also Chapter 10, pp. 274–84).

What, then, should a valid comparative approach to animal learning look like, and what, if anything, can it tell us about the structure and evolution of learning? There are at least two alternatives to the rather wholesale traditional approach to comparative studies that took little notice of evolutionary perspectives.

First, we can investigate within taxonomic groups such as mammals, birds, reptiles, and so on to identify principles of learning within these groups. Since animals within such groups share a common ancestry, we may be able to relate adaptive principles to ancestral constraints on possible evolutionary divergence. Similarly, although species within a group share common ancestry, many of them may exploit quite different niches, reflecting environmental pressures which may have selected for quite different adaptive capabilities. If different species within a group do exhibit different learning skills, then examining the lifestyle of these species should help to some extent in differentiating some of the environmental pressures that selected for those learning attributes. This theme will be elaborated in the following section of this chapter.

The second alternative approach within a comparative study of learning is in terms of anagenesis or grades of organization. The main theme of anagenesis is that successive species in a lineage often exhibit improvement in structures and functions that can be described as a succession of grades (cf. Demarest, 1983; Gottlieb, 1984, 1985; Rensch, 1959; Yarczower and Hazlett, 1977; Plotkin, 1983). This is distinct from the notion of evolutionary relatedness from a genetic viewpoint as embodied in the notion of cladogenesis. Thus, a clade is a delimitable, genetically closely related unit resulting from evolutionary diversification, but a grade is “a particular level in an ascending series of improvements on any given structural or functional unit of analysis in which the animal groups may or may not be closely related from a genetic standpoint (e.g. brain/body ratio; adaptive behavior; level of problem-solving or learning ability or, more specifically, ability to exhibit various forms of learning)” (Gottlieb, 1984: 449). Thus, different species may be members of a similar grade, but not of a similar clade; that is, they may be classed as ...