Investigating Gender Stereotypes in Large Language Models via Social Determinants of Health

TL;DR

This paper examines how large language models propagate gender stereotypes within social determinants of health in French medical records, highlighting the importance of context-specific bias assessment in sensitive domains.

Contribution

It introduces a method to probe gender stereotypes in LLMs through social determinants of health, emphasizing the need to evaluate interactions among these factors for bias detection.

Findings

LLMs rely on embedded stereotypes for gendered decisions

Probing SDoH inputs reveals gender bias in LLMs

Interaction evaluation among SDoH factors enhances bias assessment

Abstract

Large Language Models (LLMs) excel in Natural Language Processing (NLP) tasks, but they often propagate biases embedded in their training data, which is potentially impactful in sensitive domains like healthcare. While existing benchmarks evaluate biases related to individual social determinants of health (SDoH) such as gender or ethnicity, they often overlook interactions between these factors and lack context-specific assessments. This study investigates bias in LLMs by probing the relationships between gender and other SDoH in French patient records. Through a series of experiments, we found that embedded stereotypes can be probed using SDoH input and that LLMs rely on embedded stereotypes to make gendered decisions, suggesting that evaluating interactions among SDoH factors could usefully complement existing approaches to assessing LLM performance and bias.