The convergence of agent-based modeling (ABM) and large language model (LLM)-driven autonomous agents presents a transformative approach to social simulation, particularly for addressing the complexities of the United Nations sustainable development goals (SDGs). ABM enables the exploration of emergent behaviors in complex systems but is limited by static, rule-based agents that fail to capture the nuances of human decision-making. LLM-based agents, equipped with adaptive reasoning, contextual awareness, and generative capabilities, offer a solution by enhancing realism, diversity, and inclusivity in simulations. This paper explores how integrating these technologies can advance policy experimentation, enabling simulations that reflect diverse cultural contexts and emergent social norms. We discuss the technical and ethical challenges of LLM-based systems, including reasoning limitations, hallucinations, and alignment constraints, and propose strategies for governance that balance innovation with accountability. This paper, based on insights from the UNU Macau AI Conference 2024 session titled “AI Agents in Practice: Harnessing AI for All,” advocates for interdisciplinary research and human-in-the-loop frameworks to ensure responsible AI use in sustainable policymaking.
Liu et al. (Thu,) studied this question.