Neutral Voice in Echo Chambers: Effects of AI Participation on Group Polarization and Social Perception

Group discussions among like-minded individuals often intensify polarization, producing echo-chamber effects and prompting questions about whether conversational AI can facilitate more rational deliberation. Prior HCI research has predominantly conceptualized AI as a facilitator or support tool, leaving underexplored the social role of AI as an agent for reducing polarization. We report an experimental study examining how a neutral AI agent influences group discussions of a contentious public policy issue in Taiwan—the Digital National ID. Groups of three participants sharing the same initial stance engaged in structured text-based discussions either with or without an AI participant contributing neutral, information-oriented responses. Results indicate that AI participation did not significantly reduce attitude polarization. However, social perception outcomes reveal a notable asymmetry: while the AI was generally perceived as competent, participants holding opposing stances reported significantly higher levels of discomfort toward the AI. These findings suggest that introducing AI as a discussion participant reshapes affective and social dynamics even when attitudinal outcomes remain unchanged. We discuss implications for designing human–AI interaction systems for deliberative contexts