Biased or Flawed? Mitigating Stereotypes in Generative Language Models by Addressing Task-Specific Flaws

TL;DR

This paper presents a method to reduce stereotypes in generative language models by addressing comprehension failures through instruction-tuning, achieving over 60% reduction in stereotypical outputs without explicit debiasing.

Contribution

It introduces a targeted stereotype mitigation framework that disentangles bias from comprehension errors and demonstrates its effectiveness across multiple models and bias dimensions.

Findings

Over 60% reduction in stereotypical outputs

Effective across multiple bias categories

Maintains model utility while reducing bias

Abstract

Recent studies have shown that generative language models often reflect and amplify societal biases in their outputs. However, these studies frequently conflate observed biases with other task-specific shortcomings, such as comprehension failure. For example, when a model misinterprets a text and produces a response that reinforces a stereotype, it becomes difficult to determine whether the issue arises from inherent bias or from a misunderstanding of the given content. In this paper, we conduct a multi-faceted evaluation that distinctly disentangles bias from flaws within the reading comprehension task. We propose a targeted stereotype mitigation framework that implicitly mitigates observed stereotypes in generative models through instruction-tuning on general-purpose datasets. We reduce stereotypical outputs by over 60% across multiple dimensions -- including nationality, age, gender, disability, and physical appearance -- by addressing comprehension-based failures, and without relying on explicit debiasing techniques. We evaluate several state-of-the-art generative models to demonstrate the effectiveness of our approach while maintaining the overall utility. Our findings highlight the need to critically disentangle the concept of `bias' from other types of errors to build more targeted and effective mitigation strategies. CONTENT WARNING: Some examples contain offensive stereotypes.