大型语言模型越来越多地用作社交模拟器,，包括用作综合调查受访者。大多数评估都会询问模拟结果是否类似于人类结果。我们认为这是必要的，但太弱了: 模拟器可以在使用错误的基本原理派生原因模式时匹配最终答案。我们通过 94 人防晒霜概念测试来研究这个问题，每个受访者评估三个产品概念并写出开放式的理由。我们将这些基本原理映射到签名原因状态 $Z$, 中，其中正号支持采用，负号则阻止采用。这给出了实际审计:，持有受访者描述符$D$,类别上下文$K$,和概念处理$X$固定,人类理由衍生的原因是否有助于预测行为$Y$,并且LLM可以在不看到人类理由或结果的情况下模拟相同的原因状态?人类理由衍生的原因大大改善了对购买意图的预测。 LLM 模拟的原因更加脆弱: 它们通常听起来似乎合理, 但经常重复概念板，而不是恢复受访者的接受或拒绝路径。该论文为社交模拟器提供了一个评估框架。原因状态本身并不识别自然因果效应,，但它们提供了一个可解释的测试，用于测试模拟器'所陈述的原因是否与人类证据相符。
Large language models are increasingly used as social simulators, including as synthetic survey respondents. Most evaluations ask whether simulated outcomes resemble human outcomes. We argue that this is necessary but too weak: a simulator can match the final answer while using the wrong rationale-derived reason pattern. We study this problem through a 94-person sunscreen concept test in which each respondent evaluated three product concepts and wrote open-ended rationales. We map those rationales into signed reason states $Z$, where positive signs support adoption and negative signs block it. This gives a practical audit: holding respondent descriptors $D$, category context $K$, and concept treatment $X$ fixed, do human rationale-derived reasons help predict behavior $Y$, and can an LLM simulate the same reason state without seeing the human rationale or outcome? Human rationale-derived reasons substantially improve held-out prediction of purchase intent. LLM-simulated reasons are more brittle: they often sound plausible, but frequently echo the concept board rather than recover the respondent的 acceptance or rejection path. The paper contributes an evaluation framework for social simulators. Reason states do not identify natural causal effects by themselves, but they provide an interpretable test of whether a simulator的 stated reasons align with human evidence.
科目:人工智能(cs.AI)
Subjects: Artificial Intelligence (cs.AI)