arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2402.17389cs.CLcs.AI
收藏

FairBelief——评估语言模型中的有害信念

FairBelief -- Assessing Harmful Beliefs in Language Models

University of Pisa(比萨大学);University of Edinburgh(爱丁堡大学);Bocconi University(博科尼大学)

Mattia Setzu, Marta Marchiori Manerba, Pasquale Minervini, Debora Nozza

AI总结:

本文提出FairBelief方法,通过提示分析语言模型内嵌的信念,评估其在不同规模、似然性等维度上对特定性别的有害程度,揭示模型虽性能高但存在伤害性偏见。

摘要:

语言模型(LMs)已被证明会继承不良偏见,如果此类系统在未经仔细公平性审计的情况下被集成到现实世界应用中,这些偏见可能会伤害少数群体和代表性不足的群体。本文提出了FairBelief,一种分析方法,用于捕获和评估信念,即语言模型可能以不同程度的置信度嵌入并暗中影响其预测的命题。借助FairBelief,我们利用提示(prompting)来研究多个最先进语言模型在不同先前被忽视的轴上的行为,例如模型规模和似然性,评估在一个专门设计用于量化语言模型输出伤害性的公平性数据集上的预测。最后,我们以对模型所发出的信念进行深入的定性评估作为结论。我们将FairBelief应用于英语语言模型,揭示出尽管这些架构在多种自然语言处理任务上实现了高性能,但它们对特定性别表现出有害信念。有趣的是,训练过程和数据集、模型规模以及架构会引发不同程度的伤害性信念。

英文摘要:

Language Models (LMs) have been shown to inherit undesired biases that might hurt minorities and underrepresented groups if such systems were integrated into real-world applications without careful fairness auditing. This paper proposes FairBelief, an analytical approach to capture and assess beliefs, i.e., propositions that an LM may embed with different degrees of confidence and that covertly influence its predictions. With FairBelief, we leverage prompting to study the behavior of several state-of-the-art LMs across different previously neglected axes, such as model scale and likelihood, assessing predictions on a fairness dataset specifically designed to quantify LMs' outputs' hurtfulness. Finally, we conclude with an in-depth qualitative assessment of the beliefs emitted by the models. We apply FairBelief to English LMs, revealing that, although these architectures enable high performances on diverse natural language processing tasks, they show hurtful beliefs about specific genders. Interestingly, training procedure and dataset, model scale, and architecture induce beliefs of different degrees of hurtfulness.

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。

↑