Melioration learning in iterated public goods games: The impact of exploratory noise

ABSTRACT Experimental observations in iterated public goods games are explained by a simple but empirically well-grounded model of long-term reinforcement learning. In many experiments, medium levels of cooperation at the beginning decrease with further repetitions. However, in some settings, the actors only slowly learn the individual benefits of defection. In the present model, the decay in cooperation is mitigated by high individual returns, a large group size or stability in the group’s composition. Results from agent-based simulations are presented, and the underlying mechanisms are disclosed. The proposed explanation stresses the role of exploratory noise: if multiple actors explore their alternatives simultaneously, the marginal benefit of defection diminishes and cooperation can be sustained.