This paper introduces a social learning algorithm for recursive decision problems faced by players in large anonymous games. The algorithm keeps track of only the distributions of agents over possible state-action pairs. State update, value update and behavior update constitute the three stages...