The AI Economist: Optimal Economic Policy Design via Two-level Deep Reinforcement Learning

AI and reinforcement learning (RL) have improved many areas, but are not yet widely adopted in economic policy design, mechanism or economics at large. At the same time, current methodology is limited by a lack of counterfactual data, simplistic behavioral models, opportunities to experiment with policies evaluate responses. Here we show that machine-learning-based simulation powerful design framework overcome these limitations. The Economist two-level, deep RL trains both agents social planner who co-adapt, providing tractable solution highly unstable novel two-level challenge. From simple specification an economy, learn rational agent behaviors adapt learned vice versa. We demonstrate efficacy on problem optimal taxation. In one-step economies, recovers tax theory. complex, dynamic substantially improves utilitarian welfare trade-off between equality productivity over baselines. It does so despite emergent tax-gaming strategies, while accounting for interactions change more accurately than theory. These results first time can be used understanding as complement theory unlocking new computational learning-based approach policy.