Classical conditioning

Ivan Pavlov was measuring dog saliva to study digestion when he noticed the dogs drooling at the sound of the lab assistants' footsteps.

A cat runs to the kitchen at the whir of the can opener. A patient feels sick walking into the clinic where chemotherapy made them ill. Neither reaction was taught on purpose; both were learned by pairing. Classical conditioning is the name for learning that one event predicts another, and it was the first kind of learning to be studied with careful measurement.

Four labels

Meat powder in the mouth triggers salivation with no learning, so the powder is the unconditioned stimulus (UCS) and the drooling is the unconditioned response (UCR). The tone starts out as a neutral stimulus (NS): on its own it does nothing. After repeated pairing, the tone becomes a conditioned stimulus (CS), and the salivation it now triggers is a conditioned response (CR). The response looks the same; what has changed is what sets it off. Timing matters. The neutral stimulus usually needs to come just before the unconditioned stimulus, often within seconds.

Learning, unlearning and coming back

The period when the link forms is acquisition. If the tone keeps sounding with no food, the response weakens and stops: extinction. Yet after a rest, the tone can bring the response back, which Pavlov called spontaneous recovery. Extinction does not erase the old learning. Animals also respond to tones similar to the trained one (generalization) and, if only one tone predicts food, learn to tell them apart (discrimination).

Not every pairing is learned equally. John Garcia and Robert Koelling (1966) made rats ill after they drank flavoured water, or after lights and sounds. The rats learned to avoid the flavour, even with hours between taste and illness, but did not learn to avoid lights and sounds. Bodies seem prepared to link food with sickness. Robert Rescorla and Allan Wagner (1972) later showed that what matters is prediction: a cue that reliably signals the stimulus is learned, a cue that comes along at random is not.

In 1920 John B. Watson and Rosalie Rayner conditioned a baby, known as Little Albert, to fear a white rat by striking a steel bar behind him whenever he touched it. Albert came to cry at the rat and then at other furry things, a case of generalization. The study is famous and also weak evidence: it involved one infant, the fear was never removed, and it would be judged unethical today. Textbooks keep it as history, not as a model.