Classical Conditioning: Contingency

CONTINGENCY

TheCS’s role as a signal also has a crucial implication for what produces classical conditioning—that is, what the relationship between the CS and the US must be for learning to occur. To understand the issue, consider a dog in a conditioning experi-ment. Several times, it has heard a metronome and, a moment later, received some food powder. But many other stimuli were also present. At the same time it heard the metronome, the dog heard some doors slamming and some voices in the background. It saw the laboratory walls and the light fixtures hanging from the ceiling. At that moment, it could also feel various bodily sensations. What, therefore, should the dog learn? If it relies on mere contiguity, it will learn to associate the food powder with all of these stimuli—metronomes, light fixtures, and everything else on the scene—since they were all present when the US was introduced.

Notice, though, that many of these stimuli—even if contiguous with the US—give no information about the US. The light fixtures, for example, were on the scene just before the food powder arrived; but they were also on the scene during the many min-utes when no food was on its way. So the sight of the light fixtures can’t signal that food is coming soon, because the presence of the light fixtures has just as often conveyed the opposite message. Likewise for most of the sounds in the laboratory; they were present just before the food arrived, but they were also present during minutes without food. Therefore, none of these stimuli will help the animal predict when food is coming and when it’s not.

To predict the US’s arrival, the dog needs some event that reliably occurs when food is about to appear and doesn’t occur otherwise. And, of course, the metronome beat in our example is the only stimulus that satisfies this requirement, since it never beats in the intervals between trials when food is not presented. Therefore, if the animal hears the metronome, it’s a safe bet that food is on its way. If the animal cares about signal-ing, it should learn about the metronome and not about these other stimuli, even though they were all contiguous with the target event.

Are animals sensitive to these patterns? Said differently, what is it that leads to clas-sical conditioning? Is it contiguity—the fact that the CS and US arrive close to each other in time? Or is it contingency—the fact that the CS provides information about the US’s arrival? It turns out that contingency is the key, and in fact a CR is acquired only when the CS is informative about things to come.

In one experiment, rats were exposed to various combinations of a bell (CS) and a shock (US) (Figure 7.11). The bell was never a perfect predictor of the shock, but it did signal that shock was likely to arrive soon. Specifically, presentation of the bell signaled a 40% chance that a shock was about to arrive.

For some of the rats in this experiment (Group A in the figure), shocks also arrived 40% of the time without any warning. For these rats, therefore, the bell pro-vided no information. The likelihood of a shock following the bell was exactly the same as the likelihood of shock in general. And in fact this situation led to no condi-tioning; instead, the rats simply learned to ignore the tone.

For another group of rats (Group B in the figure), the bell still signaled a 40% chance of shock, and shocks still arrived occasionally without warning. For these ani-mals, though, the likelihood of a shock was only 20% when there was no bell. So in this setting, the bell was an imperfect predictor but it did provide some information, because shock was more likely after the bell than otherwise. And, in this situation, the rats did develop a conditioned response—they became fearful whenever the bell was sounded.

Let’s be clear that in this experiment, the two groups of rats experienced the same number of bell-shock pairings, and so the degree of contiguity between bell and shock was the same for both groups. What differed between the groups, though, was whether the bell was informative or not—and it’s this information value, not the contiguity, that matters for conditioning. Notice also that the bell was never a perfect predictor of shock: Bells were not followed by shock 60% of the time. Even so, conditioning was observed; apparently, an imperfect predictor is better than no predictor at all (Rescorla, 1967, 1988; Figure 7.12).