From Prediction to Conditional Exploration in LLM Agent Simulations for AI Policy Governance

Dilrukshi Gamage

Keywords: LLM Agents, Policy, Goverance

Abstract:

Large Language Model (LLM)–based agent simulations are increasingly proposed as tools for informing AI governance. However,
policymaking is inherently uncertain, high-stakes, and socially embedded. This position paper argues that LLM-agent simulations
should not function as predictive policy engines but as structured uncertainty exploration systems. Drawing on generative agent ar-
chitectures and recent multi-agent simulation work, I examine two challenges: interpreting conditional simulation outcomes under
deep uncertainty, and establishing transparency in architecturally opaque systems. I propose a co-evolutionary framework integrat-
ing robustness stress-testing and layered auditability, reframing simulation from prediction toward conditional exploration and
institutional accountability in AI policy design.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 11