Building a Foundation for Data-Driven, Interpretable, and Robust Policy Design using the AI Economist

Optimizing economic and public policy is critical to address socioeconomic\nissues and trade-offs, e.g., improving equality, productivity, or wellness, and\nposes a complex mechanism design problem. A policy designer needs to consider\nmultiple objectives, policy levers, and behavioral responses from strategic\nactors who optimize for their individual objectives. Moreover, real-world\npolicies should be explainable and robust to simulation-to-reality gaps, e.g.,\ndue to calibration issues. Existing approaches are often limited to a narrow\nset of policy levers or objectives that are hard to measure, do not yield\nexplicit optimal policies, or do not consider strategic behavior, for example.\nHence, it remains challenging to optimize policy in real-world scenarios. Here\nwe show that the AI Economist framework enables effective, flexible, and\ninterpretable policy design using two-level reinforcement learning (RL) and\ndata-driven simulations. We validate our framework on optimizing the stringency\nof US state policies and Federal subsidies during a pandemic, e.g., COVID-19,\nusing a simulation fitted to real data. We find that log-linear policies\ntrained using RL significantly improve social welfare, based on both public\nhealth and economic outcomes, compared to past outcomes. Their behavior can be\nexplained, e.g., well-performing policies respond strongly to changes in\nrecovery and vaccination rates. They are also robust to calibration errors,\ne.g., infection rates that are over or underestimated. As of yet, real-world\npolicymaking has not seen adoption of machine learning methods at large,\nincluding RL and AI-driven simulations. Our results show the potential of AI to\nguide policy design and improve social welfare amidst the complexity of the\nreal world.\n