Prototype theory proposes that categories are mentally represented by a single summary representation — the prototype — typically the central tendency (mean) of the experienced category members. New items are classified by computing their similarity to each category's prototype, with the most similar category winning. This contrasts with exemplar models, which store all individual instances.
The Prototype Model
Distance: d(x, μ_A) = [Σ wₖ|xₖ − μ_Aₖ|ʳ]^(1/r)
Similarity: s(x, μ_A) = exp(−c · d(x, μ_A))
P(A|x) = s(x, μ_A) / Σ_j s(x, μⱼ)
Prototype vs. Exemplar Debate
The prototype-exemplar debate has been one of the most productive in cognitive psychology. Prototypes are more computationally efficient (storing one point per category vs. all exemplars) and naturally produce typicality gradients. However, exemplar models (like the GCM) can also produce prototype-like behavior as an emergent property of summing over many exemplars, and they additionally capture sensitivity to within-category variability, frequency effects, and exception items that pure prototype models cannot.
Modern research suggests that both prototype and exemplar representations may be used, depending on category structure (rule-based categories may favor prototypes, information-integration categories may favor exemplars) and individual differences.