Evaluating Gender, Racial, and Age Biases in Large Language Models: A Comparative Analysis of Occupational and Crime Scenarios

2024-09-22

arXiv.org

AI Summary

This study systematically evaluates gender, racial, and age biases in four leading large language models (LLMs) released in 2024 within two high-stakes domains: occupational representation and crime-related inference. Method: Leveraging real-world statistics from the U.S. Bureau of Labor Statistics (BLS) and the Federal Bureau of Investigation (FBI), we construct a cross-dimensional quantitative bias analysis framework—first enabling comparative assessment of coupled biases across occupational and criminal contexts. We employ refined Winogender/Winobias prompting and multi-model horizontal evaluation. Results: In occupational tasks, female representation deviates by 37% from BLS benchmarks; in crime-related tasks, gender, racial, and age biases reach 54%, 28%, and 17%, respectively. Crucially, we identify that mainstream debiasing techniques induce “over-indexing” among subgroup populations, revealing unintended overcorrection risks. These findings establish a new empirical benchmark for LLM fairness evaluation and governance.

Technology Category

• Machine Learning: Ethics, Bias, and Fairness
• Philosophy and Ethics of AI: Bias, Fairness & Equity
• Computer Vision: Bias, Fairness & Privacy

Application Category

• Social Networks and Social Media: Fairness and bias in social network and social media analysis
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work

Abstract

Recent advancements in Large Language Models(LLMs) have been notable, yet widespread enterprise adoption remains limited due to various constraints. This paper examines bias in LLMs-a crucial issue affecting their usability, reliability, and fairness. Researchers are developing strategies to mitigate bias, including debiasing layers, specialized reference datasets like Winogender and Winobias, and reinforcement learning with human feedback (RLHF). These techniques have been integrated into the latest LLMs. Our study evaluates gender bias in occupational scenarios and gender, age, and racial bias in crime scenarios across four leading LLMs released in 2024: Gemini 1.5 Pro, Llama 3 70B, Claude 3 Opus, and GPT-4o. Findings reveal that LLMs often depict female characters more frequently than male ones in various occupations, showing a 37% deviation from US BLS data. In crime scenarios, deviations from US FBI data are 54% for gender, 28% for race, and 17% for age. We observe that efforts to reduce gender and racial bias often lead to outcomes that may over-index one sub-class, potentially exacerbating the issue. These results highlight the limitations of current bias mitigation techniques and underscore the need for more effective approaches.

Problem

• Evaluating gender bias in occupational scenarios in LLMs
• Assessing gender, racial, age bias in crime scenarios in LLMs
• Analyzing limitations of current bias mitigation techniques in LLMs

Innovation

• Debiasing layers to reduce model biases
• Specialized datasets for bias evaluation
• Reinforcement learning with human feedback