Rescorla-Wagner Model

The Rescorla-Wagner model is a theory of classical conditioning that formalizes learning as proportional to the discrepancy between what is expected and what occurs. It posits that the strength of an association changes by a fixed fraction of the difference between the maximum possible association and the current associative strength, scaled by the intensity of the conditioned stimulus. The model was revolutionary because it predicted phenomena that earlier theories could not explain — notably blocking, overshadowing, and the conditioned inhibitor effect — by treating learning as error-driven rather than as mere contiguity between stimulus and response.

The model's deepest insight is that associative learning is not about pairing events in time but about updating predictions. It is the psychological ancestor of modern reward prediction error theories in neuroscience and temporal difference learning in artificial intelligence. The mathematics are simple; the implications are not. By making learning a function of surprise, the Rescorla-Wagner model transformed behaviorism into something that looks remarkably like predictive processing.

The Rescorla-Wagner model is not merely a historical curiosity. It is the proof that the brain has been computing prediction errors since before we had words for them — and that any system that learns by association, biological or artificial, must converge on some variant of this rule or fail to learn at all.