Reproduce “AI Agents as Policymakers in Simulated Epidemics” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Goshi Aoki, Navid Ghaffarzadegan

Researchers at Virginia Tech developed and evaluated generative AI agents as policymakers within simulated epidemic environments, demonstrating that these agents can make repetitive, high-stakes decisions and that their performance in reducing cumulative cases significantly improves with the integration of systems-level knowledge and ensemble decision-making.

Key insights

  • Generative AI agents can exhibit reactive, human-like policy behavior in dynamic environments, and the framework provides a computational model for studying bounded rationality in decision-making.
  • Integrating brief, theory-informed systems-level knowledge into an agent's prompt significantly enhances its ability to make timely and effective policy decisions, leading to a substantial reduction in epidemic impact.
  • Ensemble decision-making is most effective when combined with guiding causal knowledge; aggregating raw, stochastic LLM outputs without such knowledge can lead to suboptimal or worse performance.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.