A representational sensitivity study of game theoretic simulations

The iterated prisoner's dilemma is a widely used computational model of cooperation and conflict. Many studies report emergent cooperation in a population of agents trained to play the prisoner's dilemma with an evolutionary algorithm. We demonstrate levels of emergent cooperation ranging from 0% to over 90% by varying the representation of the evolving agents. The representations used in this study are finite state machines, feedforward neural networks, if-skip-action lists, parse trees storing two types of logical functions, simple look-up tables, and Markov chains. This study adds weight to the proposition that choice of representation is critical and suggests any soft computing system intended to simulate behavior must be deeply concerned with the representation issue.