Towards Implicit Bias Detection and Mitigation in Multi-Agent LLM Interactions

TL;DR

This paper investigates implicit gender biases in multi-agent LLM interactions, develops a dataset and metric to assess biases, and proposes two mitigation strategies—self-reflection with in-context examples and supervised fine-tuning—that effectively reduce biases.

Contribution

It introduces a novel dataset and metric for detecting implicit biases in multi-agent LLM interactions and evaluates two mitigation strategies, demonstrating their effectiveness.

Findings

LLMs exhibit implicit gender biases in over 50% of outputs.

Biases tend to increase after multi-agent interactions.

Combining fine-tuning with self-reflection best reduces biases.

Abstract

As Large Language Models (LLMs) continue to evolve, they are increasingly being employed in numerous studies to simulate societies and execute diverse social tasks. However, LLMs are susceptible to societal biases due to their exposure to human-generated data. Given that LLMs are being used to gain insights into various societal aspects, it is essential to mitigate these biases. To that end, our study investigates the presence of implicit gender biases in multi-agent LLM interactions and proposes two strategies to mitigate these biases. We begin by creating a dataset of scenarios where implicit gender biases might arise, and subsequently develop a metric to assess the presence of biases. Our empirical analysis reveals that LLMs generate outputs characterized by strong implicit bias associations (>= 50\% of the time). Furthermore, these biases tend to escalate following multi-agent interactions. To mitigate them, we propose two strategies: self-reflection with in-context examples (ICE); and supervised fine-tuning. Our research demonstrates that both methods effectively mitigate implicit biases, with the ensemble of fine-tuning and self-reflection proving to be the most successful.