Animal learning linked to reward and punishment has been a major concern of psychologists since Thorndike and Pavlov. Unfortunately, research has tended to focus on ever-more standardized experimental “paradigms” in which one or two well-defined responses could be studied under highly controlled conditions. Behavioral variation, an essential feature of learning, has been largely painted out of the picture.

The field has become increasingly divided into two topics: classical (Pavlovian) conditioning and operant conditioning. Classical conditioning research is dominated by the idea of association, hence the label associative learning – although associative processes are presumed to operate also in operant conditioning. In fact, the division conforms to no characteristic of the processes involved. The distinction between classical and instrumental conditioning is a matter not of process but of procedure (open-loop vs. closed-loop, in the terminology of Chapter 6; see also Chapter 5). I have said much about operant conditioning in earlier chapters. Now it is time to look a bit more closely at classical conditioning and the differences between the two.

The division between classical and operant conditioners has led to different theoretical and practical preoccupations. As I pointed out in Chapter 9, operant conditioners have almost abandoned work on the process of learning and focus almost exclusively on steady-state adaptation to behavior that is demonstrably reversible – which at least allows them to work with individual animals rather than groups. The underlying state is certainly not reversible, as I also pointed out earlier. Classical conditioners retain an interest in a fundamentally irreversible learning processes, which has forced them to study and compare groups even though this method also has its problems.

Learning means change, and change has given psychologists the same kind of trouble that motion gave Zeno: How can one study motion, when a body must be in one place or another? How can something move at all? The Procrustean solution is to act as if there is a single learning “process” and then design our experiments so as to preclude any other. Some of the more tightly controlled conditioning procedures have this flavor.

A safer tack may be to look at a range of learning phenomena in different species and see what useful generalizations emerge. In this chapter, I look at the learning of bees, taste-aversion learning, and several experiments on classical conditioning.