FairMonitor: A Dual-framework for Detecting Stereotypes and Biases in Large Language Models

TL;DR

FairMonitor introduces a dual-framework combining static and dynamic detection methods to comprehensively identify explicit and implicit stereotypes and biases in Large Language Models across diverse scenarios.

Contribution

This work presents a novel dual-framework that effectively detects nuanced stereotypes and biases in LLMs using static tests and multi-agent dynamic scenarios.

Findings

Static tests evaluate explicit and implicit biases with over 10,000 questions.

Dynamic scenarios reveal subtle biases through multi-agent interactions.

Combined methods improve bias detection accuracy in LLMs.

Abstract

Detecting stereotypes and biases in Large Language Models (LLMs) is crucial for enhancing fairness and reducing adverse impacts on individuals or groups when these models are applied. Traditional methods, which rely on embedding spaces or are based on probability metrics, fall short in revealing the nuanced and implicit biases present in various contexts. To address this challenge, we propose the FairMonitor framework and adopt a static-dynamic detection method for a comprehensive evaluation of stereotypes and biases in LLMs. The static component consists of a direct inquiry test, an implicit association test, and an unknown situation test, including 10,262 open-ended questions with 9 sensitive factors and 26 educational scenarios. And it is effective for evaluating both explicit and implicit biases. Moreover, we utilize the multi-agent system to construst the dynamic scenarios for detecting subtle biases in more complex and realistic setting. This component detects the biases based on the interaction behaviors of LLMs across 600 varied educational scenarios. The experimental results show that the cooperation of static and dynamic methods can detect more stereotypes and biased in LLMs.