Apart from administrations of morphine, repeated beyond the point of addiction, no motivating event used in animal experiments has as long-term an effect on future behaviour as the delivery of painful electric shocks. Rats allowed to press a lever which causes morphine to be pumped into their bloodstreams become addicted, and even after long abstinence and treatments with methodone, which reduces withdrawal symptoms, never lose the tendency to try out the same lever, even months after their physical dependence. The only other long-term effect that can be compared to this is when (other) rats are given strong shocks for pressing a lever, or dogs learn to jump a bather to escape from shocks. As a consequence, the rats may never again approach the site of their punishing experience, and the dogs may never cease to jump over the barrier, even though no further shocks are experienced (Church, 1969; Solomon et al., 1953; Solomon, 1964).
There are ethical arguments against exposing experimental animals to this sort of procedure, and also against the use of punishment as an educational or legal sanction. And, as human
experience shows even more vividly than animal experimentation, the use of punishing events to control behaviour may have many counter-productive outcomes, such as the build-up of tolerance to punishment, the development of anger and resentment against punishing agencies and other things associated with them, and debilitating physical and emotional disorders in those punished. The desire to reduce all punishing events to a minimum both in the animal laboratory and in human societies is therefore admirable, and we have no wish, I hope, to encourage beatings in school or to return to the Norman and Tudor traditions of hanging our most punishable miscreants in public until they are too feeble to struggle and then burning their detached entrails and genitals in front of them while they are still alive.
But although the desire to downgrade the importance of punishment may be commendable, it is not possible to defend seriously the position taken by Thorndike (1932) and Skinner (I953), that while rewards strengthen behaviours, punishing events have no equivalent deterrent effects on responses which produce them. At least as far as animal experimentation goes, there is more evidence to support a symmetrical view of reward and punishment, such as that of Mackintosh (1974), who says that just as rewarded responses are generally performed because of ‘the expectation of an increase in incentive’, so ‘responses accompanied by an increase in anticipation of aversive reinforcement are thereby suppressed’ (pp. 269, 299).
The suppressing effect of punishing stimuli on previously rewarded activities is illustrated by the experiment quoted by Church (1969). In this, several groups of rats were first trained for a short time to press a lever for food rewards, with a fixed-interval schedule of reinforcement (seep. 53). Then they were given nine sessions of testing in which no further rewards were delivered. For the first fifteen minutes of the first session every lever press resulted in a shock, different groups receiving different levels of shock. The effect of this was that responding was suppressed according to the strength of the punishment given.
Approach and avoidance conflict
If responses have previously been rewarded, and are punished mildly, there is no reason why they should be suppressed com-
pletely. Any symmetrical theory of reward and punishment will have to predict that response output should reflect the balance between previous rewards and punishments. Broadly speaking, choices tend to be systematically influenced by the intensity of previous attractive and aversive experiences, as these theories suggest, although there are many exceptions. Certainly the deterrent effect of shocks on a rat’s approach to a goal is roughly in proportion to the intensity of the shock, and, at a given level, punishment may be overcome by a large enough food reward. Natural defensive responses may blur the issue (see below), but a considerable body of data reported by Miller (1944) was based on the natural tendency of hungry rats to approach a source of food, but avoid a source of danger.
In a typical experiment, rats were allowed to run back and forth along a plank or alley, at one end of which there was a goal box in which they were both fed and shocked. The net attractiveness of the goal box was assessed both by how far they approached it when placed at the other end of the alley, and by how hard they would pull towards it when placed at a certain point while in a harness attached to a calibrated spring. This allowed for independent assessment of approaches to food and avoidances of shock as well as of the combined effects of both. Under these circumstances, with rats, the factor of distance from the goal seems to be more important for aversive than for attractive stimuli. They would pull almost as hard to get towards the food from a long distance away as from close up. However, while exerting their strongest efforts to get further away when placed close to the source of shock, they showed a very sharp decline in these efforts as their distance from the anxiety-provoking stimulus increased. Thus it was said that the avoidance gradient was steeper than the approach gradient. Obviously enough, pulls towards food were proportional to degree of hunger, and pulls away from shock were proportional to the intensity of previously experienced shocks. With an appropriate combination of hunger and shock level, and with both food and shock having been experienced in the same goal, it was possible to confirm conflict between approach and avoidance tendencies; the rats would start running towards the goal, and then slow down and stop, at a distance depending on the balance between hunger and fear. Animals varied a good deal in their exact reactions, but many vacillated in a rather human manner (Miller, 1944, p. 438).
Escape and avoidance training
In the learning of particular responses to escape from and avoid electric shocks, the importance of natural patterns of behaviour (called ‘species-specific defence responses’: Bolles, 1978) is even more obvious than in other forms of animal learning.
Pigeons will peck at almost anything in surroundings where food has been recently presented, but can be trained to peck keys to escape or avoid shocks only with the utmost difficulty. Avoidance of dangerous places, by flying, running or any available form of locomotion is easy to obtain in most species, but arbitrary responses such as lever-pressing or key- pecking are likely to be hampered by competition from these natural responses or from others such as freezing or keeping very still. Certainly, rats will learn to press levers to turn off shocks, and even to lower the frequency of an intermittent train of shocks, whether at fixed or random intervals (Sidman, 1953; Herrnstein, 1969), but behaviours which involve the avoidance of dangerous places are more revealing. In a Skinner box, rats learn best to press a lever to avoid floor shocks if the lever press leads to the motorized insertion into the box of a shelf onto which they can jump (Baron et al., 1977).
Much of the data on which theories of avoidance learning are based have been obtained using a device known as a ‘shuttle box’. This is simply a box divided into two, sometimes with a barrier or door between the two compartments, shocks being deliverable through the floor of either compartment independently. Sometimes shocks are always delivered in the same compartment, and animals thus learn to get out of this whenever they are put in it (‘one-way avoidance’). In other cases shocks are delivered to either compartment and are preceded by a sound or light signal, and animals learn to shuttle out of whichever compartment they are in when the signal is given (called ‘two-way avoidance’). In all cases of successful avoidance learning, what animals learn is to avoid contact with the motivating stimulus. This is one of many important differences between attractive and aversive stimuli, or reward and punishment, as motivating events for learning. With attractive stimuli, learning leads to increased contact with the motivating event, and consequently to sustained exposure to important information, whereas with aversive stimuli, patterns of
avoidance may be construed as lack of reality testing, and it is inevitable that successful learning leads to decreased exposure to information about the reason for learning (Mower, 1969). To be on the safe side, it would be best to avoid doing things which might be dangerous, and in this way avoidance strategies can become self- sustaining in a way in which reward-seeking strategies cannot.
The only theory of avoidance-learning worth considering here is the ‘two-factor theory’ which supposes that both Pavlovian conditioning and instrumental learning are involved — these are the two factors (or two processes; Gray, 1975). Pavlovian conditioning means that stimuli associated with shock or other forms of unpleasantness come to elicit a motivational state of fear or anxiety, and the instrumental learning consists of the animal’s attempts to lessen the fear and anxiety. This deals reasonably well with successful avoidance-learning, for example when animals run away from dangerous places. Particular test cases attempt to separate out the (classical) learning of fear from the (instrumental) avoidance of fear. Thus Miller (1948) shocked rats in one compartment of a shuttle box, and then locked the rats in that compartment. No further shocks were given, but escape was possible by the turning of a wheel which opened a door, allowing the rats to get out into the other compartment. They quickly learned to do this. It seems plausible to assume that being in the compartment in which they had previously been shocked was in some sense unpleasant, and this provided the motive for learning to turn the wheel to get out of it. Herrnstein (1969) and Mackintosh (1974) suggest that because rats will also learn to press levers to reduce shock frequencies while staying in the same box, the hypothesis of ‘conditioned fear’ is not strictly necessary. But since they have to talk instead about ‘a discriminative stimulus for the avoidance response’ (Herrnstein, 1969, p. 49), or ‘a decrease in proximity to aversive reinforcers’ and ‘an increase in anticipation of aversive reinforcement’ (Mackintosh, 1974, pp. 314, 299), their scepticism seems singularly misguided.
Taste-aversion learning and other defensive behaviours
Since the two-factor theory includes the idea of associative shifts in natural emotions aroused by unpleasant or feared events, it is
not difficult to incorporate within it phenomena which point up the importance of natural, species-specific reactions to aversive stimuli. It is now commonly accepted that associative shifts themselves will be partly determined by the innate propensities of the species involved. Taste- aversion learning is in fact widespread across species, although there are differences in detail. About 70 per cent of people in a psychiatric survey of human food aversions said that their aversion dated from particular experiences of feeling ill after eating food they were now averse to (Garcia et al., 1977). A similar phenomenon, sometimes called the ‘Garcia effect’, has been observed in many animals (Garcia, 1981).
In the original experiments rats were allowed to lap at bottles of sweet- or salty-tasting water, laps being accompanied by clicks and flashing lights. Then, either electric shocks accompanied lapping, or the animals were made ill by radiation treatment or by lithium chloride added to the water. On subsequent tests shocked rats avoided lapping from bottles if this caused clicks and flashing lights, but drank freely of quiet but tasty water, whereas the animals which had suffered intestinal distress were unconcerned by clicks and flashes, but avoided drinking water which tasted like the water they had drunk before becoming ill. Illness seems to be naturally attributed to tastes, and peripheral pain to audio-visual signals (Garcia and Koelling, 1966). There is also no need for the taste experience to be contiguous in time with illness — animals made ill seem to be put off whatever it was they have last eaten, however long that was ago (Garcia et al., 1966).
The ‘Garcia effect’ can be a very strong one, apparently counteracting other natural behaviours such as hunting. (Male rats become averse to copulation if this behaviour is repeatedly followed by illness: Peters, 1983.) In an interesting twist, wolves can be quickly deterred from attacking sheep by once being given lithium chloride tablets mixed with chopped mutton and wrapped in sheep’s clothing. After being made ill by this concoction, wolves and coyotes who previously attacked and killed live sheep, first made half- hearted attacks and then gradually became more and more submissive, eventually running away or lying down when lambs approached them. Garcia’s theory to explain these results is that the aversive experience after eating poisoned mutton wrapped in sheepskin leads to a ‘hedonic shift’ with respect to sheep
generally (Garcia et al., 1977). There are also ‘conditioned disgust responses’ since coyotes will urinate on, roll on or bury fresh meat, whether rabbit or lamb, to which they have previously been averted. Similarly, rats which have been shocked by touching a particular object in their home cage will urinate on it or pile sawdust or any other movable materials over it (Terlecki et al., 1979). Reactions to aversive events, whether conditioned or unlearned, are not just a matter of arbitrary reactions to emotional distress, but include a variety of related motives and response tendencies whose provenance is innate.
We can usually appeal to innate factors to help explain the many experiments in which animals appear to expose themselves to painful stimuli unnecessarily, either by failing to learn an avoiding response which is unnatural to them or, more dramatically, by the persistent performance of a natural response which is being punished — self-punishment, or the ‘vicious circle phenomenon’ (Brown, 1969). In an investigation in which dogs were trained to jump over a barrier in a two-way shuttle box in order to avoid very intense shock, most dogs continued to jump the barrier whenever the signal was given, though shocks would no longer occur if they stayed put. In an attempt to decondition the jumping, a short shock was given always in the side the dogs jumped to, not in the side they were jumping away from. Out of 13 dogs, 3 stopped jumping, but the other 10 carried on jumping even more quickly and vigorously. Clearly it would have been difficult, under conditions of high emotional arousal, and with the natural response of jumping, for the animals to distinguish the relation between the location of the shock and their jumping behaviour. There is also the relief, or ‘safety-signal’, theory, which suggests that when a short shock comes to an end, the fact that it has ended somehow reinforces even the responses which brought about the aversive experience. However, the importance of the persistence of natural reactions, perhaps just because they are natural, is supported by the following finding. The dogs were prevented from jumping the barrier by use of a glass screen, so that they experienced the warning signal but no shock. Thus they were exposed to the reality that not jumping was objectively safe. Nevertheless, this did not
prevent them from starting to jump again as soon as the barrier was removed (Solomon et al., 1953).
A fairly similar result was obtained with rats by Melvin and Smith (1967), using a one-way avoidance response. The rats were trained to run down a 4-foot alley. They were put in a starting box, and when a trap door opened they had five seconds to run to a safe goal box at the end before the whole alley was electrified. That was the training experience. Two groups were then compared: in one group no more shocks were given, and these rats moved more and more slowly; in the other, the second foot of the alley was always electrified — these rats would have been better off not running down the alley at all, but they continued to run, even faster than in avoidance training. Moreover, when the conditions were switched, both groups kept running, but the fastest running of all was observed in the rats who had been through the phase of running slowly and not being shocked, and then were given the condition in which just the second foot of the alley was always live. Instead of avoiding shock by staying put, by the time they had had ten trials of this final phase they were running down the alley at a speed of 5 feet a second (very fast for a rat).
Passive responses as well as active ones may appear to be self-punitive. Rats required to press a lever in a Skinner box in order to turn off shock have a natural and understandable tendency to hold the lever pressed firmly in the down position for long periods. If, at the same time as they are required to escape occasional shocks by depressing the lever, they are given additional short sharp shocks every a or 3 seconds for as long as the lever is held down, this does not induce the most effective behaviour, which would be brief lever presses. Rather, rats hold the lever down for even longer periods, thus receiving thousands of additional punishments (Migler, 1963).
A more peculiar result, originally obtained by Muenzinger (I934), is that rats who learn to run always towards light in a one-choice T-maze to obtain food actually learn faster if they are given shocks along with the food, or almost anywhere else in the maze, than if they are trained only with rewards. Although there are differing explanations for this effect, all bear some relation to the idea that aversive stimuli may enhance attention or add distinctiveness to the environment (Fowler and Wischner, 1969). All such generalizations depend a great deal on the intensity of the
stimuli involved, but for relatively mild motivating stimuli it is worth bearing in mind that one way in which rewarding and punishing stimuli can be said to have identical effects is that they both command attention.
Stress, learned helplessness and depression
Again taking due allowance for emotional intensity, it is incontrovertible that one way in which attractive and aversive stimuli can be said to have opposite effects is that rewards engender health and happiness while punishments endanger these desirable states. There are, however, a number of confounding factors to do with predictability, uncertainty, and the nature of the activities which are rewarded or punished. Although prolonged exposure to pain or other intensely unpleasant states of affairs can generally be expected to produce stress at the physiological level, measurable in animals by stomach ulceration, weight loss and mortality, the most influential hypothesis in this area is that the psychological factor of being unable to react positively to aversive events contributes both to the physiological effects of stress and to the future emotional and motivational characteristics of the individuals concerned. This is Seligman’s ‘learned helplessness’ theory. Animal experiments may be quoted both for and against it, and the application of the theory to the phenomena of human depressive states has received much attention (Seligman et al., 1968; Seligman, 1975; Abramson et al., 1978; Miller and Norman, 1979).
If dogs are strapped in a harness and given inescapable shocks, they perform poorly in subsequent tests of their ability to learn active responses to avoid shock. Similarly rats or gerbils, given tasks of escaping from or avoiding shocks which they are unable to master, will thereafter fail on relatively easy tasks of shock avoidance which other animals without the history of failure are able to learn (Seligman and Beagley, 1975; Brown and Dixon, 1983). The argument is not so much over these data, although the results of the experiments are often difficult to interpret in terms of the relevant experimental variables, but over the details of the explanation. In some cases a factor of temporarily emotional or physical exhaustion may be responsible for the results obtained in animal experiments. The generalizations made to human
depression may therefore be questioned, although emotional exhaustion is not of negligible influence for people (Miller and Norman, 1979). Clearly, however, there may be simpler explanations of the animal results than ‘giving-up’, self-blame, inadequate self-concepts and other cognitive factors which may present themselves in human psychiatric cases.
An intermediate level of explanation, which appears to be appropriate to many of the animal experiments, is that during the phase of inescapable punishment, or of insoluble escape tasks, animals learn to be inactive and passive, by innate crouching or freezing or as a more general response strategy. The main evidence for this is that ‘helpless’ rats, unable to learn the active avoidance responses of running or lever pressing, were not impaired at the task of avoiding shock by gently pushing a small panel placed just in front of their nose (Glazer and Weiss, 1976). it seems probable that passive response strategies, instead of (or perhaps as well as) more general emotional effects, are responsible for the learned helplessness phenomenon in animals. But it can be pointed out (in favour of Seligman’s application of the theory of learned helplessness to human emotions) that inactivity and even extreme slowness of normal movements are sometimes associated with psychiatric depression.
The imposition of aversive stimuli on animals is generally to be frowned on, but results from such experiments have led to developments in learning theory which must be noted. According to the two-factor theory, aversive stimuli give rise to states of fear and anxiety, which maybe conditioned to other signals when there is some natural connection involved. Conditioned fear and anxiety may then impel the learning of new responses which give relief from these unpleasant inner states, and also arouse other natural emotions and innate responses. This theory has been particularly important as an influence on behavioural explanations and recommendations concerning the origin and treatments of neurosis, which are discussed in later chapters.