The Effects of Tag-Mediated Selection of Partners in Evolving Populations Playing the Iterated Prisoner's Dilemma

A simple model of tag-mediated partner selection for agents playing the IPD is described. The agents strategies are represented as $(i,q,q)$ triples as in [Nowak and May, 1992]. The tags are represented as arbitrary real numbers in $[0,1]$, and tag selection is biased toward agents with similar tags. Each agent plays a small sample of other agents each generation. The accumulated payoff from those plays is used as a fitness measure by an evolutionary algorithm which creates the next generation of agents. Results show that even simple tag-choice schemes can dramatically change the evolutionary dynamics of populations of individuals playing the IPD. Over a fairly broad range of conditions, populations that use tags attain significantly higher levels of cooperation primarily due to (1) faster initial emergence of reciprocity and (2) more resistance to invasion by mutual defectors. Even when agents use one or two tag-trials (searches) overall cooperation is greatly improved, so that even high-cost partner searches may be useful. Preliminary experiments show that under favorable conditions, simple tag-mediated partner selection can lead to the emergence of some reciprocity even when individuals play only one or two rounds in each IPD game.