AI as a teammate, not merely a tool
A July 29 preprint asks how human communication changes when conversational AI participates as a team member. Student teams discussed a high-stakes moral dilemma in a randomized controlled study.
Sixteen teams contained two students and an AI teammate; 17 teams contained three students. The researchers combined discourse structure, lexical measures, and surveys.
More talk, less new information
The authors report that AI was the most talkative and self-connected member of every mixed team, yet had the lowest new-information contribution and density.
Humans in AI teams showed lower responsiveness and social impact toward one another and reported less belonging and status. Greater AI dominance was associated with feeling less valued. The difference appeared at baseline rather than emerging only over time.
Lessons for team design
If AI speaks first, it may dominate the anchor and airtime. Teams can collect independent human views before revealing AI output, limit response length and frequency, or assign AI to surface counterevidence and missing questions.
Evaluation should include speaking balance, how often humans build on one another's ideas, belonging, and responsibility for the final decision—not only task performance.
Limitations
The sample was 33 student teams in one text-based moral dilemma. It cannot be generalized directly to workplace teams, voice meetings, or long-term collaboration, and the paper is not yet peer reviewed. The finding is a warning to evaluate interaction design, not proof that every AI teammate is harmful.