|International Journal of Psychophysiology, vol. 2 (1984), 68-9.|
Conditioning and Associative Learning
By N.J. Mackintosh
Clarendon Press, Oxford/OUP, New York, 1983. £17.50 (316 pp.)
This is rightly regarded as the last word in learning theory but, as Mackintosh emphasizes on p. 2, learning theory of this particular kind ‘no longer occupies the exalted position it once held among the various fields of psychology’. The theory applies most directly to rats and pigeons receiving signal-reinforcer pairings in Skinner boxes (conditioned emotional response procedures for rats; autoshaping for pigeons): in the preface Mackintosh notes that he has 'eschewed discussion' of anything directly concerned with schedules of reinforcement, and said little or nothing about naturalistic topics such as imprinting, song learning, navigation or intelligence in animal species. That is because the theoretical aim of the book is to establish certain laws of association which make up 'one possible view of the nature of conditioning'. One cannot do justice to this view briefly, but Mackintosh first adopts a version of two-factor theory, that is he accepts an operational and functional distinction between classical and instrumental conditioning (p. 41); then he advocates a stimulus-substitution theory of classical conditioning, in which a CS elicits responses by activating a representation of a UCS, but only according to its own sensory properties (pp. 68-70); puts forward a Tolmanian theory of instrumental conditioning, in which an animal must infer from previous associations between lever pressing and food that it might be a good idea to press the lever again (pp. 110-112); argues for the theoretical symmetry of reward and punishment (pp. 126-131); just about (I think) accepts the two-factor theory of avoidance learning (pp. 155-170); discusses various laws of association in terms of the adequacy or otherwise of the Rescorla-Wagner single-equation model (pp. 171-239); and in a final short chapter sets off the phenomena of discrimination learning as calling upon processes ‘not normally studied in simple conditioning experiments’ (p. 273) and ‘outside the scope of standard theories of conditioning’ (p. 271).
By comparison with his enormously successful The Psychology of Animal Learning, Mackintosh's present book is theoretically tighter and more succinct. The major theoretical change seems to me to be a slight firming up of the classical/instrumental distinction.
Within the areas covered, Mackintosh is so encyclopaedically knowledgeable, and au fait with the merits and failings of all conceivable theoretical positions, as to be quite above criticism. One can only lament, for the purposes of this journal, that the learning theories of the present Cambridge school make so little contact with any kind of physiology—and with psychophysiology in particular. They are cerebral if not ethereal theories. Cambridge rats have recently become capable of translating their thoughts into action but seldom, in these pages, do they translate their thoughts into emotions. They do not pant, defecate or change their heart rate and therefore, within theories of associative learning, connections with psychophysiological measurement are not yet obvious, though that does not mean such connections are necessarily non-existent. Mackintosh's index contains entries for neither nervous system (central or autonomic), nor drive, motivation or stress. It is surely possible, at least in principle, to extend theories of this general type in the direction of flesh and blood. For instance, it would be interesting to bring data on ulceration and weight loss to bear on the issue of whether the concept of an aversive motivational state is really needed to explain how animals learn to avoid electric shocks, when these are not preceded by a special signal (p. 157).
Mackintosh might not agree that such physiological data would be relevant, since he puts forward a purely behavioural account of learning, both in principle and practice. Within this constraint, I have just one query. On p. 222 it is implied that a general law of association, which will ‘underlie any example of successful conditioning’, is ‘that there be a true causal relation between the events to be associated’. A great deal rests on what exactly is meant by ‘a true causal relation’ here. Truth and cause are going to be a difficulty anyway (they entail false causal and true non-causal relations between stimuli), but as the context here is taste-aversion learning, by which rats can very effectively be put off the taste of saccharin if this happens to precede whole body X-irradiation, I suspect that this could be a page on which Mackintosh might be caught out. However, it is undoubtedly a feature of this book that such pages, if any exist at all, are very rare indeed.
Stephen F. Walker