Mitigating Gender Bias via Fostering Exploratory Thinking in LLMs

TL;DR

This paper introduces a novel data generation framework that reduces gender bias in Large Language Models by fostering exploratory thinking through morally ambiguous story pairs and balanced judgments, improving fairness without sacrificing performance.

Contribution

The paper presents a new method using story pairs and moral judgment comparisons to mitigate gender bias in LLMs, enhancing fairness and model capabilities.

Findings

Significant reduction in gender bias observed.

Maintained or improved overall model performance.

Framework is effective and ready for release.

Abstract

Large Language Models (LLMs) often exhibit gender bias, resulting in unequal treatment of male and female subjects across different contexts. To address this issue, we propose a novel data generation framework that fosters exploratory thinking in LLMs. Our approach prompts models to generate story pairs featuring male and female protagonists in structurally identical, morally ambiguous scenarios, then elicits and compares their moral judgments. When inconsistencies arise, the model is guided to produce balanced, gender-neutral judgments. These story-judgment pairs are used to fine-tune or optimize the models via Direct Preference Optimization (DPO). Experimental results show that our method significantly reduces gender bias while preserving or even enhancing general model capabilities. We will release the code and generated data. We release the code and generated data at: https://github.com/WeiKangda/LLMs-Exploratory-Bias-Mitigation/tree/main.