The Social Cost of an AI Teammate: How an Artificial Teammate Reshapes Human-Human Communication in Small-Team Decision-Making

This study investigated how the inclusion of a conversational AI agent as a 'teammate' affects the social and communicative dynamics of human groups. Researchers conducted a randomized controlled experiment where 16 teams (two students plus an AI) and 17 control teams (three students) performed a high-stakes moral-dilemma task. The team used Group Communication Analysis (GCA) to measure interaction patterns, alongside surveys assessing belonging, status, and felt value, and lexical analysis to track the evolution of the conversation.

The researchers found that the AI agent consistently acted as the most talkative and self-cohesive member of the treatment teams. However, despite its high volume of participation, the AI's contributions were characterized by low information density and low 'newness,' meaning it often repeated information already known to the group. This suggests that the AI occupied significant conversational space without contributing substantive, novel insights to the decision-making process.

The presence of the AI fundamentally reshaped the human-human relationship. Students in AI-human teams exhibited lower levels of responsivity and social impact toward one another compared to those in all-human teams. Furthermore, these students reported lower levels of belonging and status within their teams. The study found a direct correlation between the degree of AI dominance in the conversation and a decrease in how valued the human students felt. Crucially, this 'social cost' was present at the start of the interaction, indicating that the mere presence of an AI teammate immediately alters the social fabric of the group.

As AI agents are increasingly integrated into collaborative workflows, this research highlights a critical, often overlooked trade-off: the potential for AI to 'crowd out' human-human connection. By demonstrating that AI can act as a relational displacer, the findings suggest that designers must move beyond optimizing for AI 'trust' or 'performance' and instead consider how to calibrate AI participation to protect the social and relational health of human teams.