Imitation and cooperation in coupled dynamical recognizers

Abstract

A coupled dynamical recognizer is proposed as a model for simulating intelligent game players, who can imitate the other player’s behavior. A kind of recurrent neural network called a dynamical recognizer is used as an internal model of the other player to imitate the behavior. The Rashevskyan game is examined, where each player moves along a separate spatial axis to take an advantageous position over the other player. Though the players are egocentric in principle, it is shown that some altruistic behavior will be performed as a dynamical attractor phase. The altruistic behavior is no longer attainable by continually modeling the opponent player merely as a Tit for Tat player. Rather, players have to dynamically change their model of imitation to achieve mutual co-operation, otherwise they go to a static non-cooperative Nash solution. Enhancement of a minute difference in players’ action patterns, called the pragmatic paradox, is the key issue throughout this paper.