EduPolicySim: LLM-Agent Simulation for Navigating the Pedagogical Policy Space

Weirui Peng, Hyoungwook Jin

Keywords: LLM Agent, Social Simulation, AI for Education

Abstract:

Pedagogical policy design requires navigating a combinatorially vast space of instructional choices---such as technique, dosage, and
timing---whose effects interact in ways that are difficult to anticipate. Existing LLM-based student simulations operate at the micro level (individual learner responses) or meso level (classroom dynamics), but do not address macro-level policy evaluation. In this position paper, we propose \textbf{EduPolicySim}, a simulation framework that uses LLM agents as proxies for diverse student populations to support pedagogical policy exploration at scale. The framework operates in four stages: student profile configuration grounded in authentic artifacts, policy specification along multiple instructional dimensions, iterative simulation with decision support, and co-evolutionary refinement using real-world outcome signals. EduPolicySim aims to surface hidden interdependencies among instructional decisions and scaffold evidence-informed policy making before costly classroom deployment.

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: 9