Dynamics of internal models in game players

A new approach for the study of social games and communications is proposed . Games are simulated between cognitive players who build the opponent's in ternal model and decide their next strategy from predictions based on the m odel. In this paper, internal models are constructed by the recurrent neura l network (RNN), and the iterated prisoner's dilemma game is performed. The RNN allows us to express the internal model in a geometrical shape. The co mplicated transients of actions are observed before the stable mutually def ecting equilibrium is reached. During the transients, the model shape also becomes complicated and often experiences chaotic changes. These new chaoti c dynamics of internal models reflect the dynamical and high-dimensional ru gged landscape of the internal model space. (C) 1999 Elsevier Science B.V. All rights reserved.