Interaction of intelligent agents in prisoner's dilemma

Abstract

A model of an intelligent agent based on recurrent neural network is presented and assessed in prisoner``s dilemma game. This method is superior than classical reinforcement learning especially for the ability of searching continuous action and continuous state space. Through experiments, we show that the agent has good adaptation ability to fixed strategies. In the both learning case, it is not guaranteed that they find optimal solution. To overcome this problem, we introduce personality to each agent. This is an attempt to build an agent retaining high adaptation capability to various environment by imitating the inherent property in human. Through evolutionary simulation, we show that the community of agents with personality evolves to bring social prosperity. In addition, the extension to a model of a stockmarket is presented in which independent adaptive agents can buy and sell stock on a central market.