We study environments in which agents are randomly matched to play the Prisoner’s Dilemma, and each player observes a few of the partner’s past actions against other opponents. We depart from the existing literature in two key aspects: (1) we allow few agents in the population to be commitment types, and (2) we do not assume a time zero in which the entire community start interacting. We show that the presence of few committed agents destabilizes the existing mechanisms to sustain cooperation, and we present a novel mechanism (which is essentially unique) that sustains stable cooperation in many environments.
