跳转至

理论指导下的欺骗检测：基于 RAG 的人工智能探索

文章背景与核心概要

本文探讨了将检索增强生成（RAG）模型与主流欺骗理论相结合，用于评估和检测书面陈述中欺骗行为的应用。研究分析了来自 700 份陈述的 39,200 个欺骗判断，测试了四种大语言模型（gpt-4o、claude-sonnet-4-6、ollama/llama3 和 deepseek-v4-flash）及多种配置。结果发现，基于 RAG 的模型的平均准确率为 54.5%，与基线模型（54.6%）及人类的典型表现相当。

尽管理论框架对整体准确率的提升作用有限，但它们对响应偏置产生了深远影响，使模型的偏好从高度倾向于判定谎言（例如，可验证性方法为 32.2%）转变为高度倾向于判定实话（例如，真相默认理论为 88.1%）。研究表明，虽然在当前的参数下，理论指导下的 AI 判断依然不可靠，但通过优化数据集、模型测试以及理论与数据的匹配，它在未来具有巨大的发展潜力。

Document Metadata

Field Detail
arXiv ID arXiv:2608.08881 [cs.AI]
Subjects Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
Authors David M. Markowitz, Timothy R. Levine
Submission Date August 9, 2026
DOI 10.48550/arXiv.2608.08881
License Creative Commons Attribution 4.0 license icon

Abstract

当前研究开发了七个基于主流欺骗理论的检索增强生成（RAG）模型，并将其做出的欺骗判断与基线模型进行了对比。研究涵盖了来自五个已发表欺骗数据集的 700 份陈述、四种大语言模型（gpt-4o、claude-sonnet-4-6、ollama/llama3 和 deepseek-v4-flash）以及两种运行类型（RAG 与基线），总共生成了 39,200 个欺骗判断。

The current work developed seven Retrieval-Augmented Generation (RAG) models based on leading deception theories and compared how deception judgments were made relative to baseline models. Across 700 statements drawn from five published deception datasets, four large language models (gpt-4o, claude-sonnet-4-6, ollama/llama3, deepseek-v4-flash), and two run-types (RAG vs. baseline), a total of 39,200 deception judgments were rendered.

检测准确率与人类典型的准确率一致，且 RAG 模型（54.5%）与基线模型（54.6%）之间在统计学上没有显著差异。基于 RAG 的模型（57.0%）比基线模型（59.7%）更少表现出真相偏置（truth-biased），但其效应量非常小。理论视角对准确率的影响微乎其微，但对响应偏置产生了重大影响：响应偏置从高度倾向于谎言（可验证性方法，32.2%）跨越到高度倾向于实话（真相默认理论，88.1%）。内容效应和模型效应进一步调节了结果。在当前的参数下，理论指导下的 AI 判断尚不可靠，但在配合更多数据集、模型测试以及理论-数据匹配时，它们展现出了潜在的前景。

Detection accuracies were consistent with typical human accuracies and not statistically different across RAG (54.5%) and baseline models (54.6%). RAG-based models (57.0%) were less truth-biased than baseline models (59.7%), but the effect size was quite small. Theoretical perspective mattered little for accuracy yet mattered substantially for response bias, which ranged from highly lie-biased (the verifiability approach, 32.2%) to highly truth-biased (truth-default theory, 88.1%). Content effects and model effects further moderated the results. Theory-guided AI judgments are unreliable with current parameters, yet they might show promise with additional datasets, model testing, and theory-to-data matching.