Evolution of Generosity and Its Infrastructure in Self-organizing Cooperative Societies

Abstract

We are studying a society of evolving cooperative agents that show a continuous tit-for-tat like behavior in an adequately modified Iterated Prisoner’s Dilemma game. Each agent evaluates the action of its opponent with the help of a threshold as cooperative or defective and responds differently to the two cases. The evolutionary mechanism consists in copying the behavior of more successful agents met at random in the society. First, we study various evolutionary schemes and we show that copying of the evaluation threshold does not offer any evolutionary advantage, while copying of the visible actions leads consistently to higher fitness. In all cases, higher fitness has to be attributed, somewhat counter-intuitively, to the tendency of society to be dominated by extremely generous agents rather than by rationally reciprocal agents, as one might expect from the experimental setup. Moreover, for generosity to become prominent and visible, the society needs to start from initially rational settings rather than random or more natural ones. These findings are consistent across many interaction configurations. A final test on whether such initial proto-rationality may be selected by evolution and/or co-evolve with evolution of generosity shows that rationality in reciprocity has to pre-exist as infrastructure for this particular social context and therefore must be selected beforehand or in parallel for some other reason.