An Empirical Evaluation of Diversity in Multi-Agent Idea Generation

ACL ARR 2026 January Submission37 Authors

Keywords: Diversity Analysis, Idea Generation, Large Language Models

Abstract: Multi-agent systems (MAS) are increasingly used for open-ended idea generation, motivated by the belief that multi-agent interaction naturally increases diversity. However, when and why such collaboration truly expands the solution space remains unclear. We present a systematic empirical study of diversity in MAS-based ideation using scientific proposal generation as a controlled testbed. We analyze diversity across three levels: model intelligence, agent cognition, and system dynamics. At the model level, we identify a compute efficiency paradox, where stronger, highly aligned models yield diminishing information gain despite increased sampling or agent count. At the cognition level, we uncover a paradox of expertise: authority- or expert-driven collaborations consistently suppress semantic diversity, while junior-dominated interactions explore broader idea spaces. At the system level, larger groups and denser communication often accelerate premature convergence. Using complementary diversity metrics validated by expert judgments, we show that diversity collapse arises primarily from interaction structure rather than model insufficiency. Our findings highlight the importance of preserving independence and disagreement when designing MAS for creative tasks.

Paper Type: Long

Research Area: Dialogue and Interactive Systems

Research Area Keywords: Diversity Analysis, Agent Collaboration, Large Language Models

Contribution Types: Data analysis

Languages Studied: English

Submission Number: 37