B-score: Detecting biases in large language models using response history
This addresses bias detection in LLMs, which is a critical issue for ensuring fairness in AI applications, though it appears incremental as it builds on existing bias detection and verification techniques.
The paper tackles the problem of detecting biases in large language models (LLMs) by proposing B-score, a novel metric that effectively identifies biases across different question types, and shows that LLMs can reduce biases in multi-turn conversations for random questions, with B-score improving verification accuracy on benchmarks like MMLU, HLE, and CSQA compared to existing methods.
Large language models (LLMs) often exhibit strong biases, e.g, against women or in favor of the number 7. We investigate whether LLMs would be able to output less biased answers when allowed to observe their prior answers to the same question in a multi-turn conversation. To understand which types of questions invite more biased answers, we test LLMs on our proposed set of questions that span 9 topics and belong to three types: (1) Subjective; (2) Random; and (3) Objective. Interestingly, LLMs are able to "de-bias" themselves in a multi-turn conversation in response to questions that seek an Random, unbiased answer. Furthermore, we propose B-score, a novel metric that is effective in detecting biases to Subjective, Random, Easy, and Hard questions. On MMLU, HLE, and CSQA, leveraging B-score substantially improves the verification accuracy of LLM answers (i.e, accepting LLM correct answers and rejecting incorrect ones) compared to using verbalized confidence scores or the frequency of single-turn answers alone. Code and data are available at: https://b-score.github.io.