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 2405.03098cs.CL
收藏

FairMonitor:一种用于检测大语言模型中刻板印象与偏见的双框架

FairMonitor: A Dual-framework for Detecting Stereotypes and Biases in Large Language Models

School of Computer Science and Technology, East China Normal University(华东师范大学计算机科学与技术学院);School of Computer Science, Fudan University(复旦大学计算机学院)

Yanhong Bai, Jiabao Zhao, Jinxin Shi, Zhentao Xie, Xingjiao Wu, Liang He

AI总结:

FairMonitor通过静态测试与多智能体动态场景协同,系统检测大语言模型中的显隐性刻板印象与偏见,实验证明其检测能力优于传统方法。

摘要:

检测大语言模型(LLMs)中的刻板印象和偏见对于提升公平性、减少这些模型在应用中给个人或群体带来的不利影响至关重要。传统方法依赖嵌入空间或基于概率度量,难以揭示不同语境中细微且隐性的偏见。为应对这一挑战,我们提出FairMonitor框架,并采用静态-动态检测方法对LLMs中的刻板印象和偏见进行全面评估。静态部分由直接询问测试、内隐联想测试和未知情境测试组成,包含10,262个开放式问题,涉及9个敏感因素和26个教育场景,能够有效评估显性和隐性偏见。此外,我们利用多智能体系统构建动态场景,以在更复杂和真实的环境中检测细微偏见。该部分基于LLMs在600个不同教育场景中的交互行为来检测偏见。实验结果表明,静态与动态方法的协同能够检测出LLMs中更多的刻板印象和偏见。

英文摘要:

Detecting stereotypes and biases in Large Language Models (LLMs) is crucial for enhancing fairness and reducing adverse impacts on individuals or groups when these models are applied. Traditional methods, which rely on embedding spaces or are based on probability metrics, fall short in revealing the nuanced and implicit biases present in various contexts. To address this challenge, we propose the FairMonitor framework and adopt a static-dynamic detection method for a comprehensive evaluation of stereotypes and biases in LLMs. The static component consists of a direct inquiry test, an implicit association test, and an unknown situation test, including 10,262 open-ended questions with 9 sensitive factors and 26 educational scenarios. And it is effective for evaluating both explicit and implicit biases. Moreover, we utilize the multi-agent system to construst the dynamic scenarios for detecting subtle biases in more complex and realistic setting. This component detects the biases based on the interaction behaviors of LLMs across 600 varied educational scenarios. The experimental results show that the cooperation of static and dynamic methods can detect more stereotypes and biased in LLMs.

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。

↑