From the fluency fallacy to the micro-to-macro validity gap: opportunities and pitfalls of LLMs in social simulation
The integration of Large Language Models (LLMs) into social simulation has generated considerable enthusiasm, but also raises substantial methodological and epistemological challenges. This critical review examines the use of LLMs as cognitive or decision-making components of simulated agents from a computational social science perspective. Rather than treating the psychological evaluation of LLMs as separate from simulation mechanics, we argue that their behavioral and epistemic limitations can become systemic vulnerabilities when scaled to multi-agent societies. We first map the rapidly evolving landscape of LLM-driven platforms, ranging from small narrative sandboxes to population-scale and spatially structured simulations. We then develop a unified critical framework for analyzing the Micro-to-Macro Validity Gap: the propagation, nonlinear amplification, or convergent contraction of micro-level limitations—including hallucinations, stochastic inconsistency, representational biases, and alignment effects—into agent-level distortions (such as average persona convergence) and macro-level risks (such as artificial consensus and behavioral homogenization), while exposing evaluators to epistemic vulnerabilities such as the Fluency Fallacy and automation bias. We identify contexts in which LLM-based agents offer genuine operational value, including serious games, participatory environments, and exploratory modeling, while distinguishing these uses from confirmatory research with unconstrained models and precise social forecasting. Finally, we examine theory-driven hybrid architectures that embed LLMs within explicit, mechanistic, and reproducible Agent-Based modeling (ABM) frameworks. We argue that such architectures offer a promising but not sufficient path toward improving epistemic control: their validity depends on multi-level evaluation of environmental dynamics, individual behavior, cross-level interactions, and aggregate outcomes, and they remain vulnerable to the risk of physics washing.
Authors
- Patrick Taillandier (ORCID: https://orcid.org/0000-0003-2939-4827)
- Benoît Gaudou (ORCID: https://orcid.org/0000-0002-9005-3004)
- Arnaud Grignard
- Jean‐Daniel Zucker (ORCID: https://orcid.org/0000-0002-5597-7922)
- Alexis Drogoul (ORCID: https://orcid.org/0000-0002-9685-9199)
- Nghi Quang Huynh
Institutions
- Université Toulouse III - Paul Sabatier (FR)
- Can Tho University (VN)
- Thuyloi University (VN)
- Université Toulouse-I-Capitole (FR)
- Sorbonne Université (FR)
- Institut National de Recherche pour l'Agriculture, l'Alimentation et l'Environnement (FR)
- Département mathématiques, informatique, sciences de la donnée et technologies du numérique (FR)
- Institut de Recherche en Informatique de Toulouse (FR)
- Université Toulouse - Jean Jaurès (FR)
- Unité de Modélisation Mathématique et Informatique des Systèmes Complexes (FR)
- Institut Polytechnique de Bordeaux (FR)
- Institut de Recherche pour le Développement (FR)
- Université de Toulouse (FR)
- Université Toulouse Capitole (FR)
- Mathématiques et Informatique Appliquées Toulouse (FR)
Publication Details
- Journal
- Frontiers in Artificial Intelligence
- Published
- 2026-11-17
- DOI
- https://doi.org/10.3389/frai.2026.1988865
- Citations
- 39
- Primary Topic
- Multi-Agent Systems and Negotiation
- Type
- article
- Field-Weighted Citation Impact
- 37.04
Funders
- Agence Nationale de la Recherche