| Resumen |
In this article, we propose a model based on Best Response Dynamics (BRD) to examine the behavior of a group of rational agents when an external regulatory entity enforces control policies that influence the agents’ dynamics. BRD is valuable for analyzing economic and social phenomena, as it captures the tendency of agents to seek to maximize their individual benefits—a common behavior in these contexts. However, these models frequently converge to a Nash equilibrium, which may not represent a socially optimal outcome. To address this limitation, we suggest introducing an external regulatory agent that employs reinforcement learning to enhance the convergence time to Nash equilibria or, ideally, to guide the system toward socially optimal solutions. We utilize an environment modeled after a Facility Location Game (FLG) to train a reinforcement learning agent and assess the impact of its policies on the FLG’s behavior. This methodology presents a novel application of game theory and reinforcement learning for regulating complex systems, with potential implications in economics, social systems, robotics, and engineering. We present preliminary results to support our findings. |
