A Co-evolution model of Scores and Strategies in IPD games: toward the understanding of the emergence of the social morals

It was shown that the evolution of a world consisting of agents that play Iterated Prisoner Dilemma (IPD) games each other is open-ended by Lindgren. The behavior of the world is very sensitive to the values of the payoff matrix used in IPD games, because the values have great influence on the population dynamics of the world. In general, the values are fixed throughout the simulations. In the real world, however, morals and the behaviors of individuals that follow the morals have been evolved influencing mutually.
In this paper, we propose a co-evolution model of agents and scores of IPD games toward the understanding of emergence of social morals. The coevolution model consists of two layers. In the first layer, scores for IPD games are evolved using a genetic algorithm. Scores vary within the range of dilemma games, and scores that attract more agents in the second layer will gradually increase. In the second layer, agents play IPD games with all other agents following the scores that they believe and are evolved using Lindgren's model.

Simulation results showed that the values of the scores evolve toward the score which gives more payoffs for cooperative strategies and less payoffs for defective strategies as the strategies of the IPD agents are evolved. The results also showed that small colonies of defective strategies repeatedly appear and disappear throughout the simulations.