Walking out of Social Traps

Abstract

Theoretical and experimental studies of the iterated Prisoner's Dilemma have yielded a central insight into the logic of cooperation: the principle of reciprocity. However, previous models, based on either rational-actor or evolutionary processes of optimization, entail unduly restrictive assumptions. This essay proposes a behavioral alternative in which reciprocity is interpreted as operant conditioning - rewarding cooperation and punishing defection. The proposed stochastic learning model shows how players with no grasp of the structure of their relationship can escape social traps via random walk. Insights drawn from the two-person problem are then extended to multilateral encounters.