Case-Based Reasoning, Social Dilemmas, and a New Equilibrium Concept

In this paper social dilemmas are modelled as n-player games. Orthodox game theorists have been able to provide several concepts that narrow the set of expected outcomes in these models. However, in their search for a reduced set of solutions, they had to pay a very high price: they had to make disturbing assumptions such as instrumental rationality or common knowledge of rationality, which are rarely observed in any real-world situation. We propose a complementary approach, assuming that people adapt their behaviour according to their experience and look for outcomes that have proved to be satisfactory in the past. These ideas are investigated by conducting several experiments with an agent-based simulation model in which agents use a simple form of case-based reasoning. It is shown that cooperation can emerge from the interaction of selfish case-based reasoners. In determining how often cooperation occurs, aspiration thresholds, the agents' representation of the world, and their memory all play an important and interdependent role. It is also argued that case-based reasoners with high enough aspiration thresholds are not systemically exploitable, and that if agents were sophisticated enough to infer that other players are not exploitable either, they would eventually cooperate.