Detecting Prefix Bias in LLM-based Reward Models
This work addresses fairness issues in AI by exposing biases in reward models used for RLHF, which is crucial for developers and researchers aiming to build more equitable systems, though it is incremental in proposing mitigation strategies.
The paper tackled the problem of prefix bias in LLM-based reward models trained on human preference datasets, revealing significant biases across racial and gender dimensions through novel detection methods and demonstrating that these biases persist across diverse datasets and architectures.
Reinforcement Learning with Human Feedback (RLHF) has emerged as a key paradigm for task-specific fine-tuning of language models using human preference data. While numerous publicly available preference datasets provide pairwise comparisons of responses, the potential for biases in the resulting reward models remains underexplored. In this work, we introduce novel methods to detect and evaluate prefix bias -- a systematic shift in model preferences triggered by minor variations in query prefixes -- in LLM-based reward models trained on such datasets. We leverage these metrics to reveal significant biases in preference models across racial and gender dimensions. Our comprehensive evaluation spans diverse open-source preference datasets and reward model architectures, demonstrating susceptibility to this kind of bias regardless of the underlying model architecture. Furthermore, we propose a data augmentation strategy to mitigate these biases, showing its effectiveness in reducing the impact of prefix bias. Our findings highlight the critical need for bias-aware dataset design and evaluation in developing fair and reliable reward models, contributing to the broader discourse on fairness in AI.