Evaluating Nuanced Bias in Large Language Model Free Response Answers

TL;DR

The paper addresses the challenge of detecting nuanced bias in free-response outputs from large language models after business-oriented customization. It introduces a semi-automated, three-stage pipeline (automatic filtering, crowd evaluation, expert over-read) anchored to a name-reversal definition of bias and uses BBQ-context prompts to elicit diverse, free-form responses. The authors classify bias into five nuanced categories—Clear, Preferential, Implied, Inclusion, and Erasure—and provide automatic and human-in-the-loop methods (IDKNoNames, ReversedExactMatch, crowdsourcing, expert coding) to assess them at scale. This framework yields a practical, scalable approach for providing actionable feedback to refine LLM behavior, facilitating safer and more reliable deployment in real-world business settings.

Abstract

Pre-trained large language models (LLMs) can now be easily adapted for specific business purposes using custom prompts or fine tuning. These customizations are often iteratively re-engineered to improve some aspect of performance, but after each change businesses want to ensure that there has been no negative impact on the system's behavior around such critical issues as bias. Prior methods of benchmarking bias use techniques such as word masking and multiple choice questions to assess bias at scale, but these do not capture all of the nuanced types of bias that can occur in free response answers, the types of answers typically generated by LLM systems. In this paper, we identify several kinds of nuanced bias in free text that cannot be similarly identified by multiple choice tests. We describe these as: confidence bias, implied bias, inclusion bias and erasure bias. We present a semi-automated pipeline for detecting these types of bias by first eliminating answers that can be automatically classified as unbiased and then co-evaluating name reversed pairs using crowd workers. We believe that the nuanced classifications our method generates can be used to give better feedback to LLMs, especially as LLM reasoning capabilities become more advanced.