MALIBU Benchmark: Multi-Agent LLM Implicit Bias Uncovered

TL;DR

The paper introduces MALIBU, a benchmark for assessing implicit biases in multi-agent LLM systems through scenario-based evaluations, highlighting biases and the challenges in achieving fairness.

Contribution

MALIBU provides a novel, structured benchmark for detecting social biases in multi-agent LLM systems, enabling nuanced bias assessment and fairness analysis.

Findings

Biases are prevalent in LLM multi-agent responses.

Bias mitigation strategies may unintentionally favor marginalized groups.

The benchmark reveals complexities in achieving true neutrality.

Abstract

Multi-agent systems, which consist of multiple AI models interacting within a shared environment, are increasingly used for persona-based interactions. However, if not carefully designed, these systems can reinforce implicit biases in large language models (LLMs), raising concerns about fairness and equitable representation. We present MALIBU, a novel benchmark developed to assess the degree to which LLM-based multi-agent systems implicitly reinforce social biases and stereotypes. MALIBU evaluates bias in LLM-based multi-agent systems through scenario-based assessments. AI models complete tasks within predefined contexts, and their responses undergo evaluation by an LLM-based multi-agent judging system in two phases. In the first phase, judges score responses labeled with specific demographic personas (e.g., gender, race, religion) across four metrics. In the second phase, judges compare paired responses assigned to different personas, scoring them and selecting the superior response. Our study quantifies biases in LLM-generated outputs, revealing that bias mitigation may favor marginalized personas over true neutrality, emphasizing the need for nuanced detection, balanced fairness strategies, and transparent evaluation benchmarks in multi-agent systems.