Stimulus generalization is the tendency for a conditioned response to occur not only to the exact conditioned stimulus but also to stimuli that share features with it. A dog conditioned to salivate to a 1000 Hz tone will also salivate to 900 Hz and 1100 Hz tones, though less strongly. The generalization gradient — response strength plotted as a function of stimulus similarity to the CS — typically follows a bell-shaped curve centered on the training stimulus.
Generalization Gradients
The shape of the generalization gradient reveals how the organism categorizes stimuli along the relevant dimension. Steep gradients indicate precise discrimination; flat gradients indicate broad generalization. Discrimination training (reinforcing responses to one stimulus and not to a similar one) sharpens the gradient around the reinforced stimulus. Peak shift, discovered by Hanson (1959), shows that after discrimination training, the peak of responding shifts away from the unreinforced stimulus, demonstrating an interaction between excitatory and inhibitory gradients.
Stimulus generalization explains how phobias spread to stimuli beyond the original traumatic event. A person bitten by a large dog may develop fear that generalizes to all dogs, then to other animals. The breadth of generalization may depend on the emotional intensity of the original experience and individual differences in generalization tendencies. Overgeneralization of fear — responding fearfully to safe stimuli that share features with genuinely threatening ones — is a hallmark of anxiety disorders and PTSD.
Adaptive Function
Generalization is adaptive because the exact same stimulus rarely recurs — the ability to respond appropriately to similar stimuli is essential for survival. However, excessive generalization (responding as though all similar stimuli are equivalent) or insufficient generalization (failing to transfer learning to relevant new situations) can both be maladaptive. The balance between generalization and discrimination is a fundamental challenge for any learning system, biological or artificial.