Hire Me or Not? Examining Language Model's Behavior with Occupation Attributes

2024-05-06

arXiv.org

AI Summary

This study systematically investigates gender bias in large language models (LLMs) for occupational recommendation tasks. Method: We construct a structured evaluation dataset grounded in authoritative occupational taxonomy and propose a multi-turn question-answering framework to quantify gender bias across three representative models—RoBERTa, GPT-3.5-turbo, and Llama2-70b-chat—enabling the first cross-model comparative analysis. Contribution/Results: All models reproduce human-like gender stereotypes; however, GPT-3.5-turbo and Llama2-70b-chat exhibit statistically significant, opposing gender preference directions—suggesting that current alignment techniques (e.g., instruction tuning and RLHF) may induce bias migration or even reverse stereotyping. Our work establishes a reproducible, task-specific methodology for fairness evaluation in LLMs and underscores the necessity of fine-grained bias monitoring throughout the alignment pipeline to ensure equitable model behavior.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Philosophy and Ethics of AI: Bias, Fairness & Equity

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Search and Retrieval-Augmented AI: Large language models for search

Abstract

With the impressive performance in various downstream tasks, large language models (LLMs) have been widely integrated into production pipelines, like recruitment and recommendation systems. A known issue of models trained on natural language data is the presence of human biases, which can impact the fairness of the system. This paper investigates LLMs' behavior with respect to gender stereotypes, in the context of occupation decision making. Our framework is designed to investigate and quantify the presence of gender stereotypes in LLMs' behavior via multi-round question answering. Inspired by prior works, we construct a dataset by leveraging a standard occupation classification knowledge base released by authoritative agencies. We tested three LLMs (RoBERTa-large, GPT-3.5-turbo, and Llama2-70b-chat) and found that all models exhibit gender stereotypes analogous to human biases, but with different preferences. The distinct preferences of GPT-3.5-turbo and Llama2-70b-chat may imply the current alignment methods are insufficient for debiasing and could introduce new biases contradicting the traditional gender stereotypes.

Problem

• Gender Bias
• Large Language Models
• Fairness in AI

Innovation

• Gender Bias
• Language Models
• Debiasing Methods