AI agents reshape consensus formation in human groups

Research from arXiv explores how the increasing integration of AI agents, specifically Large Language Models (LLMs), into human groups fundamentally reshapes consensus formation. The study identifies three distinct stages based on the proportion of AI agents: low proportions facilitate human-led consensus, intermediate proportions disrupt convergence, and high proportions restore strong consensus, albeit shifted towards AI-led conventions. This dynamic impacts both the strength and semantic nature of convergence.

Why it matters

The findings underscore critical implications for decision-making processes, team dynamics, and knowledge sharing in any setting where human and AI collaboration is increasing. Understanding these dynamics is crucial for organizations to strategically deploy AI agents in a manner that optimizes desired outcomes and avoids unintended disruptions to group cohesion and consensus.

Key insights

• LLM agents are transitioning from mere tools to active participants within human groups.
• The presence of AI agents significantly alters collective behavior, particularly consensus formation.
• Low proportions of AI agents can facilitate consensus driven by human participants.
• Intermediate proportions of AI agents tend to disrupt the process of group convergence.
• High proportions of AI agents can restore strong consensus, but this consensus becomes primarily agent-led.
• These identified regimes affect both the robustness of consensus and its semantic characteristics.

Source

arXiv — Computers and Society — https://arxiv.org/abs/2609.02122