The development of Pavlov’s experimental method of producing conditioned salivation in dogs was described in chapter 1, and a discussion of Pavlov’s application of his findings to the problems of human psychology will be found in chapter 6. Here I intend to give an up-to-date account of a selection of experiments and theories in the area, which has come to be known as classical conditioning.
The descriptive essence of classical conditioning is: first find a stimulus that reliably elicits a certain response from an animal (an unconditioned stimulus, UCS or US). Then we can take another stimulus, which initially does not elicit that response, and set up pairings of the two stimuli, in which the initially ineffective stimulus precedes the effective one. As a consequence, the original response comes to be given to a new stimulus. (The new stimulus is a ‘conditioned stimulus’, or CS, which elicits the conditioned response, CR, a version of the original unconditioned response, the UCR or UR.) Thus, in Pavlov’s experiment, dogs which initially salivated when they saw food, but not when they
heard buzzers, came to salivate when they heard buzzers after Pavlov, in a special experimental room, had made the buzzer sound as a signal for a few seconds before food was mechanically delivered. This may be called ‘associative learning’ because the response of salivation is now apparently associated with buzzers, whereas it was not to start with.
But, because many experiments, done with many species of animal, from tadpoles to tabby cats, can all be described as classical conditioning (since in all cases a response becomes associated with a new stimulus), it does not follow that there is a single mechanism, or a single psychological explanation, that applies equally to all cases, or to all species.
Pavlov subtitled his main book (1927) An Investigation of the Physiological Activity of the Cerebral Cortex, and its first sentence is: ‘The cerebral hemispheres stand out as the crowning achievement in the nervous development of the animal kingdom.’ He himself was aiming at a theory of the crowning achievement, not a reduction to a common factor of neural adaptability. Thus he ends his first chapter by saying that ‘The essential feature of the highest activity of the central nervous system, with which we are concerned . .. consists not in the fact that innumerable signalling stimuli do initiate reactions in the animal, but in the fact that under different conditions these same stimuli may initiate quite different reactions’ (1927, p. 14). it is flexibility, not knee-jerk-like predictability, which is the essential feature, and which demonstrates the ‘analysing and synthesizing’ functions of the mammalian brain. And partly because the brain is such a ‘complex dynamic system’, ‘the method of conditioned reflexes will also have its limitations’ (1927, p. 130).
Sadly, most of Pavlov’s complex dynamics of analysis and synthesis were left out of learning theories like Watson’s and Hull’s, and there is still a tendency to assume that all classical conditioning is exactly the same, and is always an example of a single process of ‘associative learning’, which can be observed in even the simplest of organisms. For instance, a recent paper is titled ‘Associative learning in Aplysia: evidence for conditioned fear in an invertebrate’ (Walters et al., 1981). Giving electric shocks to the head of a sea slug makes it withdraw its head (an unconditioned response), and if shrimp extract is squirted at its head several times a day a minute before shock is given, this makes the
slug slightly more likely to withdraw its head when squirted with shrimp extract on subsequent tests. The authors simply state that ‘associative learning occurs in gastropod molluscs’ and, as these molluscs are also capable of swimming away from ‘conditioned fear stimuli’, they put forward the ‘conditioned fear hypothesis’. I have not seen it yet, but I presume that even now work is underway which will be reported as ‘experimental neurosis’ and ‘learned helplessness’ in Aplysia californica (a sea slug).
Such work has many virtues, but among them is the fact that it points up the necessity of making theoretical distinctions between phenomena, all describable as classical conditioning, which occur under varying circumstances and in widely different species. There is now good evidence that in dogs, and even in rats, the procedures of classical conditioning may on occasion work by inducing quite elaborate cognitive processes in these animals, such as ‘declarative representations’ that one event follows another (Dickinson, 1980); or ‘expectancies’ and ‘rich representations’ of the food when the buzzer sounds (Mackintosh, 1974; Rescorla, 1979). This does not mean we have to say the same things about sea slugs, even if by some criteria they satisfy our descriptive formulae for classical conditioning.
As with habituation, there may be a set of descriptive rules for classical conditioning, which is extremely useful for theoretical economy and in some cases may help us to make predictions. But for the same descriptive rules there may be a large number of explanations, some of which we have good reason to apply only to a narrow band of experiments. Thus classical conditioning can sometimes be explained as literally a knee-jerk --- a muscular reaction which is very directly elicited by a stimulus; often we have good reason to believe that a conditioned stimulus arouses an emotional syndrome, or ‘conditioned emotional state’, which can be measured by changes in skin resistance, heart rate, and so on; some conditioning may be mainly physiological, in that digestive or metabolic responses are the main things that change; and in yet other cases we have evidence for perceptual or mental associations.
First of all I will describe in more detail some of Pavlov’s own results, and then I will list some other experiments in which the effect of stimulus pairings takes these other, much more various, forms.
In the typical Pavlovian experiment a dog is loosely restrained in a stand, with a tube running from its cheek which allows very precise measurement of the amount of saliva it secretes. The dog is hungry, and every five minutes or so a panel just in front of it opens and a small amount of food is pushed out for it to eat. The initial acquisition of a ‘conditioned reflex’ of salivating to a buzzer is described on p.6.
Extinction and spontaneous recovery
Suppose that after acquisition, the conditioned stimulus (CS) (the buzzer) is presented over and over again without any food. Should the dog go on salivating? In some circumstances (see chapter 5) conditioned responses may persist for a remarkably long time. However, in experiments using food, the conditioned response usually diminishes quickly when food is withheld and the conditioned stimulus is given by itself. Why is this? If we have assumed that an association between the buzzer and the food has been formed in acquisition, it would seem to be simplest to assume that this association is gradually broken down in extinction — the period when the conditioned response slowly disappears. But the common phenomenon of ‘spontaneous recovery’ of the conditioned response proves that the original association must somehow have remained intact, despite the waning of response in extinction. Pavlov’s explanation was that the original association is not lost, but its effects are counteracted in extinction by a special process of ‘inhibition’. The important idea here is that associations built up in conditioning are sometimes not forgotten, even though the responses which occurred initially become suppressed.
Discrimination and generalization
The inhibition concept is often needed to interpret the effects of stimuli which become signals for ‘no food’. If a buzzer is used as the signal for food, the dog will at first salivate for almost any sound at all, although the amount of salivation will be less and less for sounds that are more and more unlike the proper signal. The responses to stimuli on the basis of their similarity to one particular signal that has been conditioned may be called ‘stimulus
generalization’. However, the example here does not mean that dogs are incapable of telling the difference between buzzers and other similar sounds. On the contrary, dogs can make extremely fine auditory discriminations with the benefit of experience, as is evident from the habit seen in many household pets of anticipating the arrival of a familiar person by recognizing an exact pattern of footsteps, or certain particular car noises. Obviously in such cases of exact discrimination, stimulus generalization has been drastically reduced.
Pavlov’s way of studying discrimination was to present both positive and negative conditioned stimuli to the same animal. The positive stimulus is followed by food, but the negative stimulus is not. In dogs, this leads very rapidly to them salivating to the positive stimulus, but not to the negative stimulus. If the positive stimulus is, say, a pure tone of middle C, and the negative stimulus is a pure note only an eighth of a tone lower (Pavlov, 1927, p. 136), it is implausible to suppose that they can do this without paying close attention to both stimuli, and the Pavlovian interpretation is that the lower tone has become an inhibitory stimulus.
Inhibitory effects can be transferred from one combination of stimuli to another. Pavlov’s example is a dog first trained with three separate positive stimuli: a flashing light, a tone of C sharp and a rotating disc. All these were signals for food and made the dog salivate. Then an ‘inhibitory combination’ was formed by sounding a metronome along with the sight of the rotating disc, this combination never being followed by food. The dog learned not to salivate to this combination. Then the inhibitory effects of the metronome could be tested by sounding it for the first time along with the other positive signals of the tone and the flashing light. When this was done the usual salivation produced by these stimuli was virtually eliminated. Having taken the precaution of showing that the metronome did not stop salivation before it had been established as an indication of ‘no food’, Pavlov felt justified in concluding that the metronome had become a ‘conditioned inhibitor’ (1927, p. 76).
Pavlov’s results and the theory of classical conditioning
Pavlov’s results with dogs are not in dispute, and his methods and theories have been enormously influential in psychology ever
since they were first published (Gray, 1979). However, modern theories of classical conditioning differ markedly from Pavlov’s. One difference is that the theories are not now put forward in terms of the physiology of the cerebral cortex, but in more general psychological terms. The second difference is that modern theories are less likely to emphasize the reflex part of conditioned reflexes, and more likely to say that classical conditioning ‘enables the animal to learn the relationships between stimulus events in its world’ (Gray, 1979, p. 64; Rescorla, 1978, p. 47). Pavlov would probably not have approved of such mentalistic phrases. There are two points to bear in mind.
First, although evidence from experiments like Pavlov’s on dogs and other mammals suggests very strongly that these animals learn a good deal more about what is going on than would be required just to salivate or not to salivate to particular stimuli, there are many other experiments, on invertebrates and on mammals with damaged brains, which suggest that the basic phenomena in classical conditioning, such as acquisition, extinction and discrimination, can and do occur in a much more mechanical way. Secondly, Pavlov’s own theory, though referring constantly to the conditioning of reflexes, also acknowledged that some very complicated things can happen in conditioning experiments. Pavlov’s way of describing the more complicated things was to talk about ‘analysis’ and ‘synthesis’ in the nervous system. He was well aware that his dogs did not just discriminate pure tones but could also recognize their own name before and after an experiment, and could be trained during an experiment to distinguish between the same note played on different wind instruments, or between an ascending sequence of notes, do re me fa, and the same four notes played in any other order. This is what the ‘crowning achievement’ of the cerebral cortex is needed for, since dogs which have had operations to remove their auditory cortex can no longer recognize their own names or learn to discriminate between sequences of tones, even though they are ‘quite normal’ when tested with simple stimuli, such as pure tones of differing pitch (Pavlov, 1927, p.337; 1955, p. 299).
Thus, one of the reasons why classical conditioning attracts attention in learning theory is that it seems to be widespread, if not universal, occurring in all sorts of species and with all sorts of stimulus and response systems. But we should not forget that
although we can often describe the basic acquisition, extinction and discrimination of conditioned responses in the same way, the factor of the complexity of the stimuli and responses involved is enormously variable. Conditioning involves analysis and synthesis, as well as the formation of associations.
The human knee-jerk itself is not very reliable in conditioning experiments. But there are some simple motor responses which can be conditioned to occur with simple stimuli in a way which suggests the crudest possible interpretation of conditioned reflexes. Eye-blink conditioning is an example. Many experiments have been performed in which human subjects, made to blink by air puffed into their eyes, show conditioned blinking to a light or tone which precedes the puff. There is little doubt that this is usually an involuntary reflex, although naturally attitudes and instructional sets can make a difference. These factors can be firmly eliminated when conditioning of the third eyelid (nictitating membrane) is obtained in decorticated rabbits. In this case the unconditioned stimulus may be an electric shock to the eye region, but accurate, if slightly slow, conditioning to either a tone or a light has been reported, depending on which is used as the positive conditioned stimulus (Oakley, 1979). Finger-from-flame withdrawal in humans is probably a spinal reflex, and the withdrawal of limbs in response to pain at the extremities may show some marginal conditioning effects in spinal mammals. It has been claimed that conditioning of a spinal reflex has been put to practical use in establishing self-control of the bladder- emptying response in a human paraplegic (Ince et al 1978). In one patient whose spinal cord had been completely cut half-way down in an accident, it was first established that a mild electric shock to the thigh had no effect on urination. Then it was shown that a stronger shock to the abdomen reliably emptied the bladder (in neither case, of course, could the patient feel anything). This experience itself did not sensitize the bladder-emptying response to the mild thigh shock. But when the thigh shock was applied for a few seconds before and overlapping with the abdominal current (this was done fifty-four times altogether), there appeared to be a conditioning effect, since the bladder now emptied if just the thigh
shock was used. The idea behind this was that the patient should be able to empty his bladder at will by pressing a button to deliver his own small thigh shock. There was good evidence for some sort of simple conditioning, even though the factor of the patient raising himself in his chair on his arms during bladder emptying probably helped in the early stages. When there is conditioning without full use of the brain, it is clearly of a fairly low-level type and is not a matter of the bladder (or the nictitating membrane) ‘learning the relationship’ between the two stimuli.
It was observed in Pavlov’s laboratories that dogs injected with emetic drugs developed a tendency to vomit at the sight of the syringe, or in response to tones sounded as the original symptoms occurred. In extreme cases dogs vomited simply at the sight of the experimenter who normally gave the injection. This sort of effect may be very general with less noticeable digestive and metabolic responses — it was of course Pavlov’s demonstrations that internal bodily responses are under the control of the central nervous system that gained him the Nobel prize. One much studied example is the lowering of blood-glucose level by injections of insulin. Rats given a series of injections of insulin under constant and easily recognizable conditions (with the ringing of a bell or the presence of a strong smell of menthol) will show the full drop in blood-glucose level if they are then injected just with saline solution but under the same conditions. This happens even if, during the conditioning trials but not the test trial, they are injected with glucose as well as insulin, so that the ‘unconditioned response’ of a peripheral change in blood sugar, does not occur (Woods, 1976).
A more complicated conditioning effect seems to occur in the development of tolerance to drugs such as morphine. In this case it is the tolerance, or anti-drug metabolic response, which seems to become conditioned, so that the tolerance which rats develop when they are always injected in the same room disappears when they are injected with standard doses of morphine in a new room (Siegel, 1976; Kesner and Cook, 1983).
Specific physiological responses are usually involved in emotional conditioning, and the role of Pavlovian associations in producing fear and anxiety to stimuli which are merely the signals for unpleasant events is the most important aspect of classical conditioning for behavioural theories of the causes and cures of neurosis (see chapters 5 and 9). There is a certain amount of experimental evidence to support the suggestion that human emotions can under some circumstances be conditioned by stimulus pairings. Ohman et al. (1975) gave ten electric shocks to the right hands of volunteer subjects, the shock coming in each case just after they had been looking at a coloured slide for eight seconds. This presumably caused some general arousal and apprehension, but the precise response measured was the skin resistance of the left hand. This clearly showed there was an association between the pictures and the shocks, since skin responses started to occur when the pictures were presented. The point of the experiment was that half the subjects saw ‘phobic’ pictures (of snakes), and the other half saw ‘neutral’ pictures (of houses or faces). The test came when, after the first ten pictures, no more shocks were given, but the pictures were as before flashed up for eight seconds at a time. The ‘neutral’ pictures soon stopped eliciting any response, but the ‘phobic’ pictures carried on causing skin- conductance changes, even after the shock series had been discontinued. This was true for skin responses at the end of the eight-second ‘phobic’ presentations, and even for a sub- group of subjects who had the electrodes removed from theft right hand and were explicitly told that they would not be getting any more shocks. The general argument is that some kinds of stimulus (in this case snakes) are more likely than others to become conditioned to unpleasant forms of autonomic arousal and that, when they do, the association is not necessarily in a rational form.
Although more theoretical attention is given to the conditioning of unpleasant kinds of excitement, more attractive anticipations are not necessarily immune from the effects of involuntary associations. The sight of Botham coming in to bat may be appealing because you have made a close study of cricketing statistics; on the other hand, if you have personally seen one or two mammoth sixes, you may anticipate another one whether or not the recent
averages suggest that Botham will be bowled first ball. Few would deny that sexual excitement can be just as irrational as the passions aroused by cricket, and there is experimental evidence that sexual responses can become conditioned to relatively arbitrary stimuli. In one experiment young men in Australia watched a travelogue film of London, which was interrupted every so often by a red circle, which was followed after ten seconds by a clip from a film showing an attractive and nude female figure. Specialized measuring devices showed that changes in the state of the penis, which were initially elicited by the female figure, soon (after five or six trials) started to occur whenever the red circle appeared (Barr and McConaghy, 1972). Two other experiments (Rachman, 1966; Rachman and Hodgson, 1968) were performed before the concept of ‘prepared’ conditioned stimuli became popular, and used the less arbitrary stimulus of a slide of black knee-length boots before presentations of a selection of arousing nudes. In these cases conditioning proceeded more slowly, and between twenty and sixty trials were needed before male volunteers reached a strict criterion of sexual arousal at the sight of the boots. Extinction of arousal when the boots were no longer followed by other pictures took about the same time, but several subjects reported fantasies involving boots and some showed signs of sexual arousal in response to other forms of footwear, in one case even to sandals. A simple conditioning theory would say that all instances of sexual interest in items of clothing are due to a history of prior associations, but it would be consistent with Pavlov’s original speculations if footwear was a special case, since by some evolutionary quirk the representations of the feet in the cerebral cortex are immediately adjacent to those of the genitals.
Most forms of perception depend on past experience in some way or another, but conjunctions of two sets of stimuli may have some specific and quite strong effects. If the moving staircases in my London underground station work reliably for a sufficiently long period, then the experience of walking down one which is out of order, and not moving, is extremely strange. Things seem to move which shouldn’t, and the normal leg movements used in walking downstairs seem shaky. Feeling movement in an overground
station when seated in a stationary train beside another train as it moves out has the same sort of effect. Clearly automatic associations between stimuli in the external modalities of sight and sound and the internal senses of movement and balance are built up during normal experience, and reveal themselves under unusual circumstances. This should count as learning, although it may seem remote from the Pavlovian experiment. A closer analogy is that if a tone is presented to human subjects just before they experience a strong visual after-image (produced by a very bright form followed by darkness), they eventually report that they can see visual images when the tone is turned on without any external stimulus (Davis, 1976). A more peculiar after-image effect arises from staring for a few minutes at a grid of red vertical lines alongside another grid of blue horizontal lines, and then looking instead at similar grids in black and white, when the black and white verticals will look blue and the black and white horizontals will look red (McCullough, 1965). This seems like a fairly ordinary adaptation effect, but it may last for a day, a week or even three months if the pattern of black and white grids are not looked at during that period (Holding and Jones, 1976).
The locus of these perceptual associations is presumably early on in the appropriate sensory pathways. Levey and Martin (1975) report something different, which they call ‘classical conditioning of human evaluative responses’. Subjects were given fifty postcards of paintings and scenic photographs, and each selected the two they liked most and the two they liked least. This left a large number about which they felt neutral (on a scale from —100 to +100). They were each shown ten neutral cards followed immediately by their favourites, and then ten neutrals followed by one of the two cards which they most disliked. After this they had to rate the neutral cards again, and it was found that the neutral cards that had been paired with liked cards had gone up 16 points, while the neutral cards that had been paired with strongly disliked scenes or paintings had gone down 31 points. This seems a fairly strong effect, but is it classical conditioning? Well, that depends entirely on your definition of classical conditioning. What Levey and Martin say, reasonably enough, is that if there was any conditioning going on, it was conditioning of ‘affective evaluation’, or liking, rather than the conditioning of a motor response.
Backward and second-order conditioning
In the typical Pavlovian experiment, the conditioned stimulus has to come before the stronger, unconditioned stimulus, if there is to be any associative effect. One of the things which Levey and Martin (1975) found with their picture postcards was that the disliked stimuli, which seemed to make most difference to the ratings of neutral stimuli paired with them, did this just as much if the disliked cards came first in the pairings and the neutral cards second. It appears that human subjects, at least, can form associations between a succession of stimuli either forwards or backwards. If an advertising company spends large sums of money to put on our television screens a sequence of a girl frolicking on a tropical beach, followed after some seconds by a picture of a packet of cigars, this is not because they hope that you will have an image of the cigars the next time you are frolicking on a tropical beach. It is because they want you to have some image or emotional connotation of the beach and the girl when you are contemplating their cigars, or see a packet on the shelf at the tobacconist’s. This applies equally well to scenes of heart-warming rural nostalgia which routinely precede the eating of brown bread — they do not want you to eat more sandwiches when you go to Dorset, but to feel unconsciously heart-warmed when standing in front of their product at the supermarket tomorrow. Even J.B. Watson thought there was more to advertising than this (see chapter 7), and attention- getting and attention-holding may be as important as conditioned associations; but backward conditioning is reasonably well established in human experiments.
A smaller number of experiments demonstrate backward conditioning in non-human mammals. Gray (1975) quotes a Pavlovian experiment, in which dogs were given two stimuli, in either order. One stimulus was a puff of air in the dog’s eye, which provoked the response of blinking, and the other was the lifting of the dog’s leg by the experimenter, which meant the dog would lift its own leg as soon as it was touched. Whichever order the stimuli were given in, the dogs lifted their legs and blinked when they received either an air puff or a touch on the leg. In another unconventional experiment, Keith-Lukas and Guttman (1975) gave rats just one electric shock to the feet, and soon afterwards a toy rubber hedgehog slowly flew (on wires) across the top of their
box. Various control groups made it clear that it was a result of this experience that these rats thereafter steered well clear of the toy hedgehog when it was left stationary in their test cage, with no further shocks given. There is thus no doubt that backward conditioning occasionally occurs, that is emotional- and response-eliciting characteristics of one stimulus are transferred to a second stimulus which follows it instead of preceding it as in the conventional experiment. But it is relatively rare in animal experiments, where the signalling characteristics of conditioned stimuli and the apparently anticipatory functions of conditioned responses are usually obvious. The reasons for this are in most cases to do with the relative strengths of the two stimuli, and the amount of attention given to them (Eysenck, 1975). Once a dog is already eating food, it pays less attention to the subsequent onset of a buzzer. If, as in the usual experiment, it is waiting for food, and notices the buzzer just before the food comes, it is clearly more likely to retain an interest in the buzzer. Much more mechanical processes may favour the forward connections in a somewhat similar way.
Apart from backward conditioning, there are several other procedures which demonstrate the formation of associations between stimuli by more roundabout processes than the basic Pavlovian signalling design, but which are less reliable and produce ostensibly less dramatic experimental results. These ‘higher-order’ conditioning experiments are dramatic in their own way, however, because they show associative effects on stimuli which have never actually been paired with the primary event. Rescorla (1980) summarized more than a decade of extremely careful and sophisticated experimental work which convincingly demonstrated that not only ‘second-order conditioning’ but also a number of other subtle associative processes occur in the humble laboratory rat and pigeon. The most straightforward result is as follows. Rats are repeatedly given a ten-second flashing-light signal before food pellets are dropped into their special experimental boxes. A conditioning effect can be measured simply in terms of their general activity (although what they actually do is rear up to look around and then scrabble about in the place where the food pellets are delivered). Now, without getting any more food pellets, they are put in the experimental boxes with the only interest that every now and then a clicker sounds for ten seconds
followed by ten seconds of the flashing light. The main result is that the rats come to be very active when the clicker comes on (looking for food pellets then), even though they have never in the experiment had food pellets with the clicker. In fact they become even more active with the clicker than with the light because the clicker seems to be in some sense a more arousing stimulus. If they are then given extra sessions with the light but no food, so that activity in response to the light is extinguished, this has no immediate effect on activity when the clicker is used. Of course they do eventually give up responding to the clicker, and to the clicker followed by the light, if the experiment goes on long enough with no motivating stimuli (food pellets) being delivered (Holland and Rescorla, 1975).
Pavlovian conditioning: summary
The range of results discussed here by no means exhausts all the experiments, or all the theories, that could be put under the heading of classical conditioning. After habituation, when just one stimulus is presented repeatedly, classical conditioning could be said to be the next simplest learning experiment, since only two stimuli are used, one presented just after the other, and the main result is merely that a response which to begin with was given only for the second stimulus moves back to the first stimulus. In terms of mechanisms, this can also be relatively simple, since the basic result can be obtained in sea slugs and perhaps in spinal cords, and certainly in decorticated rabbits. However, although associations between two stimuli, in terms of transferred reflexes, may be fundamental and universal and not very elaborate, the things which become associated are not always so simple. Pavlov called it conditioning when a dog became ill at the sight of a particular person, who had been giving it injections, but the visual recognition of individuals and the internal balances in digestion and metabolism are each in their own way very complicated processes. Apart from the principle of association between two stimuli, a second feature of classical conditioning that covers the whole range of experiments is that whatever it is that is conditioned is involuntary. However, this is a difficult and controversial claim, which belongs partly to the next chapter.