LLM 智能体行为是否自洽？面向社会模拟的 Latent Profiling 方法

Are LLM Agents Behaviorally Coherent? A Latent Profiling Approach for Social Simulation

AI 导读

研究提出 Latent Profiling 框架，用于检验 LLM 智能体的行为自洽性——其对话行为是否如人类行为模型所预测的那样，由短问卷elicited出的自身潜在画像所决定。在九个开源模型上测试了关于偏好、开放性和一致性的六项假设，结果显示智能体虽能复现表层趋势，却在深度测试中系统性失败：即便个体回答看似像人，它们也并不行为自洽。

正文 · 原文

View PDF HTML (experimental)

Abstract:LLM agents are increasingly proposed as substitutes for human participants in social research, usually on the strength of survey responses that match those of the humans they represent. We argue that such output-level alignment is not enough. Drawing on the behavioral sciences, we propose Latent Profiling, a framework for testing behavioral coherence: whether an agent's conversational behavior follows, as human behavioral models predict, from its own latent profile elicited by short surveys. As a case study, we test six hypotheses about preference, openness and agreement across nine open models. Agents reproduce surface-level trends but systematically fail in-depth tests: they are not behaviorally coherent, even when their individual responses look human. Latent Profiling thus offers a theory-grounded instrument for a behavioral science of LLMs.

Comments:	19 pages. Project page: this https URL
Subjects:	Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG)
Cite as:	arXiv:2509.03736 [cs.AI]
	(or arXiv:2509.03736v3 [cs.AI] for this version)
	https://doi.org/10.48550/arXiv.2509.03736 arXiv-issued DOI via DataCite

Submission history

From: James Mooney [view email]
[v1] Wed, 3 Sep 2025 21:55:29 UTC (1,179 KB)
[v2] Fri, 8 May 2026 15:32:06 UTC (490 KB)
[v3] Sun, 11 Oct 2026 06:56:04 UTC (479 KB)

来源：arXiv cs.AI · arxiv.org