Reinforcement learning rules in a repeated game

This paper examines the performance of simple reinforcement learning algorithms in a stationary environment and in a repeated game where the environment evolves endogenously based on the actions of other agents. Some types of reinforcement learning rules can be extremely sensitive to small changes in the initial conditions, consequently, events early in a simulation can affect the performance of the rule over a relatively long time horizon. However, when multiple adaptive agents interact, algorithms that performed poorly in a stationary environment often converge rapidly to a stable aggregate behaviors despite the slow and errative behavior of individual learners. Algorithms that are robust in stationary environments can exhibit slow convergence in an evolving environment.