Evolving memory: Examining the impact memory has on evolutionary outcomes in iterated game

The genetic algorithm approach to evolutionary game theory offers a promising way to explain the formation of institutions and explore the complex dynamics that govern the establishing of and transitions between various cooperative and non-cooperative equilibria. I introduce a model that simulates these events by using evolving populations of deterministic agents. I demonstrate how different memory lengths affect these outcomes for agents with a general memory of all opponents and specific memory of individual opponents. Finally, I introduce a model whereby memory length is a fully endogenous part of the model, allowing agents with different memory lengths to interact and reproduce. I find that there exists a memory length, and therefore level of strategic complexity, that maximizes the well-being of a population. This level depends on the specification of memory, with general-memory populations cooperating better with longer memories and specific-memory populations cooperating better with shorter memories, ignoring direct memory cost. Populations with specific memory maintain cooperation better but do not achieve as high average payoffs during times of cooperation as populations with general memory. I also find that, in populations with endogenous memory, the individual's preference for shorter memory overrides the population-wide benefits of longer memory. These results imply that policies designed to reduce uncertainty might have the unintended consequence of increasing the risk of cooperation collapse. Another implication is that policies that extend individual actor time horizon should improve outcomes for society as a whole.