Genetic Algorithms and Evolutionary Games

Genetic algorithms (GAs) have been used widely in evolving game-playing strategies since the mid-1980's. This paper looks at a particular game | the iterated prisoner's dilemma game, which is of interest to many economists, social scientists, evolutionary computation researchers and computer scientists. The paper describes a computational approach which uses a GA to evolve strategies for the 2 or more player iterated prisoner's dilemma game. Three important issues are addressed in this paper: (1) Can cooperation be evolved from a population of random strategies when the number of players is more than 2? (2) What is the impact of the group size, i.e., the number of players on the evolution of cooperation? (3) How stable are evolved strategies? Although the above three issues have been studied for the 2 player iterated prisoner's dilemma game, few results are available for the N player (N > 2) iterated prisoner's dilemma game. This paper concentrates on the N player game. The answers to the rst two questions are not xed. They depend on the group size, i.e., the number of players in the game. It is observed that as the group size increases it is more diicult to evolve cooperative strategies in a population, a phenomenon which is analogous to the real-life situation in our society, e.g., an agreement is normally more diicult to reach in a large group. This paper addresses the third issue by presenting a theoretical result on evolutionarily stable strategies.