arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2605.20767cs.CLcs.LGstat.ME
收藏

干预的幻觉:你的LLM模拟实验实际上是一个观察性研究

The Illusion of Intervention: Your LLM-Simulated Experiment is an Observational Study

Google DeepMind(谷歌DeepMind);Carnegie Mellon University(卡内基梅隆大学)

Victoria Lin, Taedong Yun, Maja Matarić, John Canny, Arthur Gretton, Alexander D'Amour

AI总结:

本文探讨了大型语言模型在模拟人类行为中的潜在作用,指出在LLM模拟的合成用户中进行干预可能引起潜在用户属性的意外变化,从而导致用户漂移,影响效果估计。本文提出了使用负对照结果来检测分布变化的方法,并通过调整角色描述以减少偏倚来缓解漂移问题。

摘要:

大型语言模型(LLMs)显示出作为人类行为模拟器的潜力,提供了一种可扩展的方式研究对干预的反应。然而,由于LLMs主要基于观察性数据进行训练,在与LLM模拟的合成用户进行实验时,干预可能会引起潜在用户属性的意外变化,导致用户漂移,其中隐含的模拟总体在不同处理条件下有所不同,这可能会扭曲效应估计。我们正式化了由于用户漂移可能产生的混淆或选择偏差,并展示了干预依赖性变化如何放大或减弱干预下用户响应的观测差异。为了诊断混淆,我们提出使用负对照结果——在干预下应保持不变的属性——来识别干预条件间的分布变化,提供用户漂移的证据。为了缓解漂移,我们研究了通过获取额外的混杂因素来调整角色描述,发现针对特定场景的相关混杂因素可以显著减少调查式和多轮代理评估中的偏倚。

英文摘要:

Large language models (LLMs) show potential as simulators of human behavior, offering a scalable way to study responses to interventions. However, because LLMs are trained largely on observational data, interventions in experiments with LLM-simulated synthetic users can induce unintended shifts in latent user attributes, causing user drift where the implicit simulated population differs across treatment conditions, potentially distorting effect estimates. We formalize the confounding or selection bias that can arise due to user drift and show how intervention-dependent shifts can inflate or attenuate observed differences in user responses under intervention. To diagnose confounding, we propose using negative control outcomes--attributes that should remain invariant under intervention--to identify distribution shifts across intervention conditions, providing evidence of user drift. To mitigate drift, we study adjusting the persona specification by eliciting additional confounders, finding that targeted, setting-relevant confounders can substantially reduce bias across survey-style and multi-turn agent evaluations.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑