Simultaneous Evolution of Learning Rules and Strategies

We study a model of local evolution. Players are located on a network and play games agains their neighbors. Players are characterized by three properties: (1) The stage game strategies they use agains their neighbors. (2) The repeated game strategy that determines the former. (3) A learning rule that selects the repeated game strategy, on the basis of the player's own and the neighbors' payoff and repeated game strategy. The dynamics that specifies learning rules is given exogenously. Players sample their neighbors' learning rules and their respective payoff. Then they construct a model that related parameters of the learning rules to payoffs. Given this model they choose an optimal learning rule. We find that under this dynamics learning rules emerge in the long run which behave deterministically but which are asymmetric in the sense that while learning they put more weight on the learning players experience then on the observed players one. Nevertheless stage game behavior under these learning rules is similar to behavior using symmetric learning rules.