LLM Economist: Large Population Models and Mechanism Design in Multi-Agent Generative Simulacra
Seth Karten, Wenzhe Li, Zihan Ding, Samuel Kleiner, Yu Bai, Chi Jin
2025-07-22
Summary
This paper talks about the LLM Economist, a new system that uses AI models to simulate and study economic policies by creating virtual populations of agents with realistic jobs and behaviors.
What's the problem?
The problem is that designing effective economic policies is very hard because real economies have many people with different needs and actions, and traditional models can’t easily capture this complexity or predict the outcome of new policies well.
What's the solution?
The authors built a framework that uses large language models to create detailed virtual agents based on real demographic data. These agents interact in simulations where a tax policy planner tries different strategies using reinforcement learning to improve social welfare. The model can also simulate voting and political effects.
Why it matters?
This matters because it allows researchers and policymakers to test and improve economic policies in a virtual, safe environment before applying them in the real world, potentially helping create fairer and more effective fiscal systems.
Abstract
The LLM Economist framework uses agent-based modeling and reinforcement learning to design and assess economic policies, demonstrating improved social welfare through simulations of large, demographically realistic agent populations.