The Schema Theorem

The schema theorem formalizes how adaptive populations learn. Schemata are patterns of building blocks — partial specifications that fit many candidates. Short, low-order schemata (those specifying few positions, close together on the genotype) that confer above-average fitness increase in frequency exponentially across generations under selection and recombination. This gives Holland's framework its predictive machinery: the system discovers useful building blocks not by evaluating them individually but by letting the patterns containing them propagate through the population. The theorem has been subject to decades of formal critique — it holds rigorously only in infinite populations and cannot distinguish easy from hard problems — but its underlying insight about probabilistic, context-sensitive, never-quite-converging credit assignment has proven durable across domains from biology to economics to human-AI collaboration.