arXiv cs.CY· Lucca Pfr\"under, Riccardo Loconte, Bruno Verschuere, Bennett Kleinberg·· 6 天前AI 评分32
对抗攻击揭示人类与生成式语言模型在欺骗能力上的差异
Adversarial Attacks Reveal Divergent Deception Competence in Humans and Generative Language Models
AI 导读
一项预注册实验（n=501）对比了人类与生成式语言模型对自动欺骗分类器的改写攻击效果，发现人类更擅长将真话改写成谎言，而生成式语言模型更高效地让谎言听起来像真话。计算文本分析显示，人类有针对性地利用欺骗线索，语言模型则施加通用的真实性信号改动。这一失败是双向的：既暴露了机器检测谎言的盲点，也意味着语言模型可能助长成功的欺骗。
正文 · 原文
Abstract:Machine learning models typically outperform humans in deception detection. A threat to machine learning models is their vulnerability to adversarial attacks: subtle yet targeted perturbations of input data that trick a model into misclassification. Yet, this also provides an opportunity: understanding how humans versus machines construct attacks offers new insights into second-order beliefs about deception. In a preregistered experiment (n = 501), we compare the effectiveness, efficiency, and nature of a paraphrasing attack by humans and generative language models on an automated deception classifier. We find that humans were more effective attackers when rewriting truthful statements to be deceptive, while the generative language model was more effective and efficient at making lies sound like truths. Computational text analysis suggests that this is because humans make targeted use of deception cues, while generative language models, instead, apply generic changes that signal truth. The failure to mimic deception is two-sided: while it exposes a blind spot in the machines increasingly used to detect lies, it also means that language models might facilitate successful lying.
| Subjects: | Computers and Society (cs.CY) |
| Cite as: | arXiv:2610.13363 [cs.CY] |
| (or arXiv:2610.13363v1 [cs.CY] for this version) | |
| https://doi.org/10.48550/arXiv.2610.13363 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Lucca Pfründer [view email]
[v1]
Fri, 9 Oct 2026 10:38:39 UTC (2,260 KB)
来源：arXiv cs.CY · arxiv.org