Can Lessons From Human Teams Be Applied to Multi-Agent Systems? The Role of Structure, Diversity, and Interaction Dynamics

ACL ARR 2025 May Submission5093 Authors

Abstract:

Multi-Agent Systems (MAS) with Large Language Model (LLM)-powered agents are gaining attention, yet, fewer studies explore their team dynamics. Inspired by human team science, we propose a multi-agent framework to examine core aspects of team science: structure (flat vs. hierarchical teams), diversity (via demographic personas), and interaction dynamics (through pre-/post-task interviews and GPT-4o-based conversation analysis). We evaluate team performance across four tasks: CommonsenseQA, StrategyQA, Social Interaction QA, and Latent Implicit Hate, spanning commonsense and social reasoning. Our results show that flat teams tend to perform better than hierarchical ones, while diversity has a nuanced impact.
Interviews suggest agents are overconfident about their team performance, yet post-task reflections reveal both appreciation for collaboration and challenges in integration.
GPT-4o analysis highlights limited conversational coordination among agents.

Paper Type: Long

Research Area: Special Theme (conference specific)

Research Area Keywords: Language Modeling, Computational Social Science

Contribution Types: Model analysis & interpretability, Data analysis

Languages Studied: English

Submission Number: 5093