arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2309.08573cs.CLcs.CY
收藏

Indian-BhED：一个用于衡量大型语言模型中以印度为中心的偏见的数据集

Indian-BhED: A Dataset for Measuring India-Centric Biases in Large Language Models

Oxford Internet Institute, University of Oxford（牛津大学牛津互联网研究所）

Khyati Khandelwal, Manuel Tonneau, Andrew M. Bean, Hannah Rose Kirk, Scott A. Hale

备注：
To be published in GoodIT '24, doi:10.1145/3677525.3678666. 14 pages
阅读 PDF
arXiv 页面 ↗HTML 原文 ↗

AI总结：

本文构建Indian-BhED数据集，从种姓与宗教维度评估流行LLM的印度中心刻板印象偏见，发现其偏见显著，并分析成因与干预方向。

摘要：

如今被数百万人日常使用的大型语言模型（LLM）可能会编码社会偏见，使其用户面临表征性伤害。关于LLM偏见的大量研究已经存在，但它们主要采用以西方为中心的框架，相对较少关注全球南方的偏见水平和潜在伤害。在本文中，我们通过Indian-BhED这一首个此类数据集，按照以印度为中心的框架量化流行LLM中的刻板印象偏见；该数据集包含印度种姓和宗教刻板印象语境下的刻板印象示例与反刻板印象示例。我们发现，大多数接受测试的LLM在印度语境中都有很强的输出刻板印象的倾向，尤其是与西方语境中传统研究的偏见维度（如性别和种族）相比时更为明显。值得注意的是，我们发现GPT-2、GPT-2 Large和GPT 3.5在种姓（63–79%）和宗教（69–72%）维度上，偏好刻板印象输出的句子占全部句子的比例特别高。最后，我们调查了LLM中此类有害行为的潜在原因，并提出干预技术，以同时减少刻板印象偏见和反刻板印象偏见。本研究的结果凸显了在研究AI公平性和评估LLM时纳入更多元声音的必要性。

英文摘要：

Large Language Models (LLMs), now used daily by millions, can encode societal biases, exposing their users to representational harms. A large body of scholarship on LLM bias exists but it predominantly adopts a Western-centric frame and attends comparatively less to bias levels and potential harms in the Global South. In this paper, we quantify stereotypical bias in popular LLMs according to an Indian-centric frame through Indian-BhED, a first of its kind dataset, containing stereotypical and anti-stereotypical examples in the context of caste and religious stereotypes in India. We find that the majority of LLMs tested have a strong propensity to output stereotypes in the Indian context, especially when compared to axes of bias traditionally studied in the Western context, such as gender and race. Notably, we find that GPT-2, GPT-2 Large, and GPT 3.5 have a particularly high propensity for preferring stereotypical outputs as a percent of all sentences for the axes of caste (63-79%) and religion (69-72%). We finally investigate potential causes for such harmful behaviour in LLMs, and posit intervention techniques to reduce both stereotypical and anti-stereotypical biases. The findings of this work highlight the need for including more diverse voices when researching fairness in AI and evaluating 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。

↑