Social Bias in Popular Question-Answering Benchmarks

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Kraft, Angelie, Simon, Judith, Schimmler, Sonja
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866908750395211776
author Kraft, Angelie
Simon, Judith
Schimmler, Sonja
author_facet Kraft, Angelie
Simon, Judith
Schimmler, Sonja
contents Question-answering (QA) and reading comprehension (RC) benchmarks are commonly used for assessing the capabilities of large language models (LLMs) to retrieve and reproduce knowledge. However, we demonstrate that popular QA and RC benchmarks do not cover questions about different demographics or regions in a representative way. We perform a content analysis of 30 benchmark papers and a quantitative analysis of 20 respective benchmark datasets to learn (1) who is involved in the benchmark creation, (2) whether the benchmarks exhibit social bias, or whether this is addressed or prevented, and (3) whether the demographics of the creators and annotators correspond to particular biases in the content. Most benchmark papers analyzed provide insufficient information about those involved in benchmark creation, particularly the annotators. Notably, just one (WinoGrande) explicitly reports measures taken to address social representation issues. Moreover, the data analysis revealed gender, religion, and geographic biases across a wide range of encyclopedic, commonsense, and scholarly benchmarks. Our work adds to the mounting criticism of AI evaluation practices and shines a light on biased benchmarks being a potential source of LLM bias by incentivizing biased inference heuristics.
format Preprint
id arxiv_https___arxiv_org_abs_2505_15553
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Social Bias in Popular Question-Answering Benchmarks
Kraft, Angelie
Simon, Judith
Schimmler, Sonja
Computation and Language
Artificial Intelligence
Computers and Society
Question-answering (QA) and reading comprehension (RC) benchmarks are commonly used for assessing the capabilities of large language models (LLMs) to retrieve and reproduce knowledge. However, we demonstrate that popular QA and RC benchmarks do not cover questions about different demographics or regions in a representative way. We perform a content analysis of 30 benchmark papers and a quantitative analysis of 20 respective benchmark datasets to learn (1) who is involved in the benchmark creation, (2) whether the benchmarks exhibit social bias, or whether this is addressed or prevented, and (3) whether the demographics of the creators and annotators correspond to particular biases in the content. Most benchmark papers analyzed provide insufficient information about those involved in benchmark creation, particularly the annotators. Notably, just one (WinoGrande) explicitly reports measures taken to address social representation issues. Moreover, the data analysis revealed gender, religion, and geographic biases across a wide range of encyclopedic, commonsense, and scholarly benchmarks. Our work adds to the mounting criticism of AI evaluation practices and shines a light on biased benchmarks being a potential source of LLM bias by incentivizing biased inference heuristics.
title Social Bias in Popular Question-Answering Benchmarks
topic Computation and Language
Artificial Intelligence
Computers and Society
url https://arxiv.org/abs/2505.15553