The rudiments of conditioning are familiar to most people. After experiencing a number of pairings of a signal, for example a tone or light, and a reinforcer, in this case food,
Ivan Pavlov's (1927) dogs came to salivate to the signal much as they did to the food. Similarly, B. F. Skinner's (1938) rats (see
Skinner box) would readily perform some action, such as pressing a lever, that procured food.
Clearly, both these behaviours depend upon
learning, in that their development requires the animal to experience a relationship or association between the signal and food in the case of Pavlovian or classical conditioning, and between the action and food in instrumental or operant conditioning. Although the empirical phenomena themselves are not a matter of dispute, the significance of this type of learning is contentious. Traditionally, conditioning represented the cornerstone of the
behaviourist's analysis of learning, but with the increasing focus on
cognition and information processing, the study of this form of learning has been consigned to a backwater of psychology.
The current neglect seems to be based largely upon the assumption that conditioning is a simple, automatic, and unconscious form of learning that underlies the acquisition of relatively trivial behaviour. Following a thorough survey of human conditioning studies, W. F. Brewer (1974) concluded that there was no good evidence for conditioning in human beings. This surprising claim was based not on our failure to show the appropriate behavioural changes, but on the observation that whenever conditioning-like effects occur, we are aware of the association between the signal or action and the reinforcer. Underlying this argument there appears to be the assumption that, by definition, conditioning must be an unconscious process. But neither the empirical effect itself nor its associated terminology implies such an assumption. Pavlov referred to the signal and its associated response as a 'conditional stimulus' and a 'conditional response' respectively, simply because the acquisition of this reaction was conditional or dependent upon having experienced the relationship between the signal and the food. Similarly, the food can be identified as a reinforcer because, at an empirical level, it appears to be an important agent for strengthening, or reinforcing, the conditional response. However, this terminology has undergone a subtle change in the West, so that one now finds statements to the effect that people and animals can be 'conditioned' and 'reinforced'. Obviously such statements are a travesty of the Pavlovian terminology: a person cannot be conditioned or strengthened, at least not by a conditioning experience. These distortions are important, however, for they reveal a fundamental and widespread belief about the nature of conditioning, namely that it is a passive process to which a person or animal is subjected.
The origin of this belief lies not in the conditioning phenomenon itself, but rather with the initial behaviourist explanations of this type of learning. For instance,
E. L. Thorndike (1911) in his famous 'law of effect' argued that following an action by a reinforcer simply strengthens a connection between the stimuli present when the action is performed and the action itself, so that the action reoccurs as a response to these stimuli when they are again presented. This stimulus-response theory has at least two features that bolster the idea that conditioning is a simple and passive process. First, the conditions for learning, and by implication the process underlying such learning, appear simple: in essence, successful conditioning depends just upon the temporal contiguity between the response and reinforcer. Secondly, a conditional response occurs because it is automatically elicited by the appropriate conditional stimulus. Thus, a trained rat presses the lever not because it knows about the relationship between its action and the occurrence of the reinforcer, but rather because the sight of the lever automatically triggers lever pressing. Both these claims turn out to be incorrect on further analysis.
It is a relatively easy matter to show that instrumental conditioning does not necessarily establish a simple habit released by the appropriate stimulus. Stimulus-response theory, by denying the animal any knowledge about the consequences of its actions, implies that conditional behaviour, once established, should be insensitive to any subsequent changes in the value of the reinforcer. We can test this claim by training an animal to perform an action for a particular food before devaluing the food, for example, by establishing quite separately an aversion to the food. This is readily done by inducing nausea shortly after consuming the food. If we now give the animal the opportunity to perform the action that had previously procured the food, stimulus-response theory would anticipate that the vigour of the action should be unaffected by the devaluation of the reinforcer. Not surprisingly, however, the animal is reluctant to perform this action (Adams and Dickinson 1981), indicating that simple instrumental conditioning reflects the acquisition of knowledge about the relationship between an action and the occurrence of the reinforcer rather than establishing a reflexively elicited habit. The deployment of this knowledge in controlling behaviour can then be modified by other relevant information, such as the current value of the reinforcer. Given this perspective, instrumental conditioning can be seen to represent a relatively simple procedure for investigating learning about the consequences of our actions and general purposive and goal-directed behaviour.
Just as our view of the knowledge underlying conditioning has changed from that specified by Thorndike's 'law of effect', so have our theories about the conditions for the acquisition of this knowledge. The 'law of effect' states that simple temporal contiguity between an action and a reinforcer is sufficient for instrumental conditioning, and Pavlov himself argued that a classical conditional stimulus also acquires its properties as a result of contiguous pairings with a reinforcer. It has long been recognized that a system sensitive only to temporal contiguity would often fail to distinguish real causal and predictive relationships from purely fortuitous and coincidental conjunctions, and thus be prone to the development of superstitious beliefs and behaviour. It is now clear, however, that the conditioning mechanisms embody a number of subtle and complex processes designed to counteract the formation of superstitions. For instance, conditioning depends not only upon the temporal relationship between the conditional stimulus or action and the reinforcer, but also upon whether or not the occurrence of the reinforcer is surprising or unexpected. The role of surprise is demonstrated by L. J. Kamin's (1969) 'blocking effect'. If an animal receives conditioning trials in which a conditional stimulus,
A, is paired with a reinforcer, it will subsequently show little conditional responding to a second stimulus,
X, when a compound of
A and
X is reinforced. In the initial stage the animal learns to expect the reinforcer following stimulus
A, so that its subsequent presentation following the
AX compound is fully predicted by stimulus
A and hence unsurprising. As a result, minimal conditioning accrues to stimulus
X. Conversely, if stimulus
A is initially established as a predictor of the non-occurrence of the reinforcer, the presentation of the reinforcer following the
AX compound is very surprising and leads to enhanced conditioning to stimulus
X (Rescorla 1971). This sensitivity to the surprisingness of the reinforcer will serve to protect us and other animals against the formation of superstitious beliefs and behaviour. Effectively, the conditioning mechanism appears to embody an assumption of minimum sufficient causation; an animal is less likely to attribute the contiguous occurrence of a reinforcer to the presence of a stimulus or the execution of an action if another adequate cause or signal for this reinforcer is also present.
In conclusion, the contemporary view of conditioning is considerably more complex and subtle than that expounded by the early behaviourists. We should not overplay, however, the importance of conditioning as a general model for all learning. There are many examples of non-associative learning that lie outside the scope of conditioning theory. Even so, conditioning studies appear to reveal universal mechanisms by which we and other animals intuitively detect and store information about the causal structure of our environment.
(Published 1987)— A. Dickinson
Bibliography- Adams, C. D., and Dickinson, A. (1981). 'Instrumental responding following reinforcer devaluation'. Quarterly Journal of Experimental Psychology, 33B.
- Brewer, W. F. (1974). 'There is no convincing evidence for operant or classical conditioning in adult humans'. In Weimer, W. B., and Palermo, D. S. (eds.), Cognition and the Symbolic Processes.
- Kamin, L. J. (1969). 'Predictability, surprise, attention and conditioning'. In Campbell, B. A., and Church, R. M. (eds.), Punishment and Aversive Behaviour.
- Pavlov, I. P. (1927). Conditioned Reflexes.
- Rescorla, R. A. (1971). 'Variations in the effectiveness of reinforcement and non-reinforcement following prior inhibitory conditioning'. Learning and Motivation, 2.
- Skinner, B. F. (1938). The Behavior of Organisms.
- Thorndike, E. L. (1911). Animal Intelligence: Experimental Studies.