Discovering effective strategies for the iterated prisoner's dilemma using genetic algorithms

Abstract

The iterated prisoner's dilemma is used to illustrate and model the phenomena in economics, sociology, psychology, as well as in the biological sciences such as evolutionary biology. The discovery and optimization of IPD strategies in real-world applications requires flexible strategy representation. The comparison of deterministic and non-deterministic finite state machines as the representations of strategies for the iterated prisoner's dilemma is presented. A novel chromosome representation scheme for non-deterministic Mealy finite state machines is proposed. The research on efficiency of the strategies evolved using genetic algorithms was made. Best results in competition with unknown strategies were obtained by non-deterministic strategies.