Effects of Group-based Agents on Iterated Prisoner's Dilemma Strategies in Evolving Spatial Environments

Abstract

The feeling of belongingness, to be a member of a group, is rooted in human evolutionary history. Cooperative behaviour within such groups has since been an important research topic. The evolution of cooperation in the iterated prisoner's dilemma (IPD) has been shown to be an effective tool of simulating and analysing this behaviour. However, it is unclear what the effects of group-based agents on IPD strategies in evolving spatial environments are. This paper investigates how this cooperation emerges by proposing an evolving spatial model that applies a genetic algorithm to its agents, extended to work with three distinct group types. This genetic algorithm makes use of four genetic operators: cloning, crossover, mutation and inversion. The groups considered are kin, clans and their combination. Cooperation is measured by examining (1) populations levels divided into nice, balanced and nasty groups, and (2) average cooperation levels of both strategies and games played per iteration. Experiments for all group types with two reproduction preferences were conducted. Three distinct conclusions can be drawn from the results. First, strategies evolved through domestic reproduction exhibit more cooperation for in-group opponents and more defection for out-group opponents. Second, strategies evolved through wealthy reproduction exhibit the same increase and decrease of cooperation as domestic reproduction, but to a smaller degree. Third, an evolving spatial environment with group-based agents develops subgroups, defined by similar strategies and restricting group-wide cooperation. Thus, agents with cooperative domestic strategies and defective foreign strategies win, and there is a positive correlation between group size and restriction of group-wide cooperation.