Chapter 4 Summary

Theories of conditioning have made steady progress. Since the early 1970s, they have been able to account for an increasingly wide and sophisticated range of effects that have been discovered in classical conditioning.
The Rescorla-Wagner model is built on the idea that learning depends on the surprisingness of the US. If the US is perfectly predicted on a conditioning trial, it is not surprising, and no learning will occur on that trial. If the US is not predicted accurately, there is surprise, and the CSs present on the trial either gain or lose associative strength accordingly. The associative strengths of all CSs present on a trial add together to determine what is predicted.
The Rescorla-Wagner model assumes that excitors are CSs that acquire a positive associative strength during conditioning. Inhibitors are CSs that acquire a negative strength. To explain the effects of zero and negative contingencies between a CS and a US (which cause no learning and inhibition, respectively), the model assumes that background contextual stimuli acquire associative strength and then influence the CS as any other CS would.
Research stimulated by the Rescorla-Wagner model makes it clear that conditioning is not a simple-minded matter of associating two events that are paired. Pairing a CS and a US is not always good enough to cause learning; under some conditions, pairing a CS with a strong US can actually reduce associative strength or even cause the learning of conditioned inhibition. Conversely, presenting a CS without a US does not necessarily result in extinction or conditioned inhibition. Learning on any conditioning trial depends crucially on the associative strength of all cues present on the trial.
Despite its successes, the Rescorla-Wagner model has several shortcomings. It incorrectly predicts that inhibition will be lost if an inhibitor is presented without a US. It does not explain latent inhibition. And it also fails to predict results that suggest that blocking may result at least partly from the animal learning not to pay attention to redundant predictors of the US.
The Mackintosh and Pearce-Hall models propose that learning is always influenced by the animal’s attention to the CS. The amount of attention paid to a CS changes over trials; in fact, it is controlled by previous learning. According to the Mackintosh model, animals pay attention to relatively good predictors and tune out bad or redundant ones. According to the Pearce-Hall model, animals attend to uncertain predictors; they will still attend to a CS on Trial 2 only if the US was poorly predicted on Trial 1. Hybrid models suggest that both of these attentional processes can operate and play complementary roles.
Wagner’s short-term memory model expanded on the Rescorla-Wagner model by proposing that learning is determined by the surprisingness of both the US and the CS on a conditioning trial. Either stimulus is not surprising if it has been “primed” in short-term memory. Priming occurs (1) if the stimulus has been presented recently (self-generated priming) or (2) if a memory of the stimulus has been retrieved from long-term memory by a stimulus previously associated with it (retrieval-generated priming). Contextual cues associated with the CS or US can cause retrieval-generated priming.
The short-term memory model explains habituation by proposing that a stimulus becomes less surprising with repeated exposure. Self-generated and retrieval-generated priming should both play a role, although the implication that habituation should be lost with a change of context has not been confirmed.
During conditioning, the subject may learn an association between nodes in long-term memory that correspond to the CS and US. Wagner’s SOP model proposes that the nodes can be activated to two levels of intensity: A1, which roughly corresponds to focal awareness; and A2, which corresponds to peripheral awareness. When a stimulus is presented, its node is briefly activated to A1; from there, it decays to A2. An association is formed between two stimuli only if they are both in A1 at the same time. SOP makes most of the predictions made by the short-term memory model, but it accounts for even more phenomena, such as the crucial effects of CS and US timing on conditioning.
AESOP, an extension of SOP, proposes that USs have both “sensory” and “emotive” nodes. Activation of these nodes evokes different kinds of responses. Emotive nodes control emotional responses, and once activated, they take more time to decay than do sensory nodes.
An additional extension of SOP further proposes that activation of the CS node is different when the CS is presented alone or along with other stimuli. This allows the theory to explain certain “configural” effects in which organisms respond differently when the CS is presented alone versus being presented in a compound with other stimuli.
Although conditioning theories will continue to change and advance, SOP theory (with its extensions) is a powerful theory that can account for many of the most important effects in Pavlovian learning.