Benefits of memory for the evolution of tag-based cooperation in structured populations

Abstract

We study the effects of working memory capacity and network rewiring probability on the evolution of cooperation in the standard and modified versions of an agent-based model of tag-mediated altruism. In our evolutionary model, computational agents populate a large complex network, engage into multiplayer Prisoner’s Dilemma-like interactions, and reproduce sexually. Agents carry discernible phenotypic traits subject to mutation, memorize their own experiences, and employ different strategies when interacting with different types of co-players. Choices made are selected from a pool of two conditional and two unconditional strategies, depending on the available memory contents and phenotypic similarity among interactors. For the dominating strategy in our standard model version, we found a strong dependence of cooperation on network structure and a weak one on memory, whereas in the modified version, the structural effect was weaker than that of memory. Most importantly, we found that the previously reported decline of cooperation in memory-based models, typically observed at a high memory capacity, is now prevented with the help of tags. This suggests that the evolutionary advantages of memory capacity limits may be far more complex than previously assumed. For much smaller systems, we observed a quasi-symmetric alternation of the two winning groups of strategists. This result provides an example of ingroup biased interactions that are characterized by bursts of intra-tag cooperation interspersed with periods of unconditional transient altruism. Such switches of strategies may represent a boosting mechanism necessary for the emergence and stability of global altruism in its early evolutionary stages.