W. Arthur

Paper #: 90-026

This paper explores the idea of calibrating a learning algorithm to match the way and rate at which human agents learn in a general, multi-choice setting. It develops a parametrized stochastic-learning algorithm, analyzes its dynamics, and calibrates its parameters against human learning data from psychological experiments. The resulting calibrated algorithm appears to replicate human learning behavior to a high degree. It can therefore be used to replace the idealized, perfectly rational agents in appropriate neoclassical models with “calibrated agents” that represent “actual” human behavior. The paper discusses the possibilities using the algorithm to represent human learning in normal-form stage games and in more general neoclassical models in economics. It explores the likelihood of convergence to long-run optimality, and to Nash behavior, and the “characteristic learning time” implicit in human adaptation in the economy.