Strategy evolution in a spatial IPD game where each agent is not allowed to play against itself

Evolution of cooperative behavior has been examined in many studies on the IPD (Iterated Prisoner's Dilemma) game under various conditions. In some studies, each agent is allowed to play against itself. However, this setting is somewhat strange because we do not play any real-world games against ourselves. In this paper, we examine the effect of this somewhat strange setting on the evolution of cooperative behavior in a spatial IPD game. Two cases are compared with each other: Each agent is allowed to play against itself in one case and not allowed to do so in the other case. It is shown through computational experiments that similar results are obtained from these two cases when opponents of each agent are selected from a large number of its neighbors. However, the difference between the two cases is large when the number of neighbors is small. Actually the evolution of cooperative behavior is strongly facilitated by allowing each agent to play against itself when the number of neighbors is small. Our computational experiments are performed on a spatial IPD game with various specifications of the neighborhood size where binary and real number strings are used as game strategies.