Does Evolution Make Reasoning Improve Learning?

Bernard Borges and Peter M. Todd - Max Planck Institute for Psychological Research

This paper investigates the usefulness of various simple reasoning strategies (introspection) in combination with reinforcement learning (RL) in the ultimatum game. In particular, evolutionary simulations are used to pit different combinations of reasoning and RL against each other. Our simulations cover a wide range of reasoning strategies, and we present tests with respect to the viability and evolutionary stability of these strategies. In contrast to the findings of previous researchers who compared RL against exogenously specified reasoning rules, we find that evolved reasoning rules can actually aid an agent's RL process. In particular, our results suggest that evolution will find and use only those reasoning abilities that aid RL.

Scheduled for Session 4.6 Agent-Based Computational Economics - II