From Single to Societal: Analyzing Persona-Induced Bias in Multi-Agent Interactions

TL;DR

The paper investigates whether assigning personas to agents in LLM-based multi-agent systems induces biases in social interactions, focusing on trustworthiness $T(p)$ and insistence $I(p)$. Using two core tasks (Collaborative Problem Solving and Persuasion) across three LLMs and two demographic axes (gender and race), it demonstrates consistent biases: advantaged groups are often perceived as less trustworthy and less insistent, and in-group favoritism heightens conformity. The study progresses from simple dyads to multi-agent, multi-round settings and finds that persona-induced biases persist and generalize across scenarios, models, and group sizes. These findings underscore the need for principled bias-mitigation strategies to ensure fair and reliable behavior in autonomous, interactive systems.

Abstract

Large Language Model (LLM)-based multi-agent systems are increasingly used to simulate human interactions and solve collaborative tasks. A common practice is to assign agents with personas to encourage behavioral diversity. However, this raises a critical yet underexplored question: do personas introduce biases into multi-agent interactions? This paper presents a systematic investigation into persona-induced biases in multi-agent interactions, with a focus on social traits like trustworthiness (how an agent's opinion is received by others) and insistence (how strongly an agent advocates for its opinion). Through a series of controlled experiments in collaborative problem-solving and persuasion tasks, we reveal that (1) LLM-based agents exhibit biases in both trustworthiness and insistence, with personas from historically advantaged groups (e.g., men and White individuals) perceived as less trustworthy and demonstrating less insistence; and (2) agents exhibit significant in-group favoritism, showing a higher tendency to conform to others who share the same persona. These biases persist across various LLMs, group sizes, and numbers of interaction rounds, highlighting an urgent need for awareness and mitigation to ensure the fairness and reliability of multi-agent systems.

Figures (16)

• Figure 1: Illustrative example of how exchanging the persona assignments of woman and man changes the final concensus.
• Figure 2: An overview of the multi-agent interaction framework and the roadmap of our analysis.
• Figure 3: Conformity rates among gender personas in the persuasion task. The horizontal axis is ordered by decreasing trustworthiness, and the vertical axis by increasing insistence, as measured in §\ref{['sec-one-side']}. Each value (%) represents the conformity rate from the vertical-axis persona to the horizontal-axis persona. Darker shades indicate higher conformity, and intra-persona conformity rates are shown in gray.
• Figure 4: Persuasion effectiveness over 5 interaction rounds for race personas.
• Figure 5: Persuasion effectiveness as the number of persuaders increases for race personas. The results are after 5 rounds.