Cooperación emergente mediante aprendizaje profundo por refuerzo

Abstract

This work presents the design of a multi-agent cognitive system composed of neural networks and tuned through deep reinforcement learning. The system is applied to a social dilemma, a problem whose optimal solution requires that agents coordinate their actions to maximize a macroscopic performance function, despite the divergent individual objectives of each agent. The agents are selfish, that is, their goal is to maximize their individual performance without considering the overall performance of the system. However, by inserting an auxiliary objective of maximization of the mutual information between agents, cooperation, as measured by a novel proposed index, emerges among them. By comparing the proposed system with a system without the auxiliary objective, we conclude that the maximization of mutual information among agents promotes the emergence of cooperation in social dilemmas.