Pavlov’s dogs are the famous story, but the real engine of learning is surprise. Robert Rescorla and Allen Wagner wrote it as one line: a cue’s link to an outcome grows by α·β·(λ − ΣV) — the gap between what happened (λ) and what was already expected (ΣV).
Pair the bell with food and the gap starts wide, so learning is fast; as the bell comes to fully predict the food the gap closes and learning stops. Ring the bell alone and the gap goes negative — the link unwinds. That is extinction.
The strangest prediction is blocking (Leon Kamin, 1968): train the bell first, then add a light that shines on every food trial. The light is perfectly visible — yet the subject never learns it, because the bell already predicted the food and left no surprise to explain.
That same prediction-error signal is now read straight off dopamine neurons and is the “reward” in the reinforcement learning behind modern AI.
Something in the simulation stopped unexpectedly — the lesson continues without it. Nothing you did was wrong; you can move on.