Forgiveness in strategies in noisy multi-agent environments

Abstract

Game theory has been widely used in modelling interactions among autonomous agents. One of the most oft-studies games is the iterated prisoner's dilemma. Prevalent assumptions in the majority of this work have been that no noise is present and that interactions and gestures by agents are interpreted correctly. In this paper, we discuss two classes of strategies that attempt to promote cooperation in noisy environments. The classes of strategies discussed include: forgiving strategies which attempt to re-establish mutual cooperation following a period of mutual defection; and memory-based strategies which respond to defections based on a longer memory of past behaviours. We study these classes of strategies by using techniques from evolutionary computation which provide a powerful means to search the large range of strategies' features.