Chaos in Learning a Simple Two-Person Game

Yuzuru Sato, Eizo Akiyama, and J. Doyne Farmer

01-09-049

We investigate the problem of learning to play a generalized rock-paper-scissors game. Each player attempts to improve her average score by adjusting the frequency of the three possible responses. For the zero-sum case the learning process displays Hamiltonian chaos. The learning trajectory can be simple or complex, depending on initial conditions. For the nonzero-sum case it shows chaotic transients. This is the first demonstration of chaotic behavior for learning in a basic two-person game. As we argue here, chaos provides an important self-consistency condition for determining when adaptive players will learn to behave as though they were fully rational.