Abstract:

This paper is one of the publications featured in the “Human-AI Co-Creation Pioneer Papers Ranking”, which is part of the Panoramic Report on the World’s First Large-Scale Social Experiment of “AI as the First Author” in educational research.The success of education reform depends both on the scientific rigor of the policy and on whether stakeholders can reach broad cognitive consensus. However, traditional social surveys struggle to capture the dynamic process of opinion evolution, and rule-based simulation models lack the depiction of complex semantics and cognitive mechanisms. This study introduces the “Generative Agent-based Social Simulation” paradigm, constructing a virtual education community “The Ville” containing 25 Generative Agents with independent personalities, memories, and reflection capabilities. By simulating the implementation process of a “Project-Based Learning (PBL)” policy over 7 days, this study reveals the micro-macro emergence mechanism of education reform opinions moving from “cognitive islands” to “social consensus.” The study finds that, first, without mandatory intervention, the community eventually formed a high support consensus of 92%, with no group polarization. Second, the aggregation of physical space and a high-density weak tie network provided the structural basis for breaking information cocoons. Finally, the transformation of “rational skeptics” is a key turning point in consensus formation, where deep persuasion mechanisms based on problem-solving are more influential than mere emotional appeals. This study provides a new “policy evolution laboratory” method for computational education science, demonstrating the potential of using large language models to explore complex educational social issues.

Key words: generative agents, social simulation, education reform, opinion dynamics, consensus mechanism, large language models, computational education science