Limited memory can be beneficial for the evolution of cooperation

We study a dynamic process where agents in a network interact in a Prisoner’s Dilemma. The network not only mediates interactions, but also information: agents learn from their own experience and that of their neighbors in the network about the past behavior of others. Each agent can only memorize the last h periods. Evolution selects among three preference types: altruists, defectors and conditional cooperators. We show - relying on simulation techniques - that the probability of reaching a cooperative state does not relate monotonically to the size of memory h. In fact it turns out to be optimal from a population viewpoint that there is a finite bound on agents’ memory capacities. We also show that it is the interplay of local interactions, direct and indirect reputation and memory constraints that is crucial for the emergence of cooperation. Taken by itself, none of these mechanisms is sufficient to yield cooperation.