Investigating LLM-Powered Dissenting Minority Support in Power-Imbalanced Group Decision-Making: Counterargument and Mediation as Intervention Strategies

This study investigates how LLM-powered interventions can support minority voices in power-imbalanced group decision-making. The researchers tested two strategies: AI-generated counterarguments (AIGC), where the AI acts as a neutral devil's advocate, and AI-mediated messages (AIMM), where the AI paraphrases and posts anonymous contributions from minority members. The experiment involved 96 participants in 24 groups, each with a 3:1 senior-to-junior power structure, to evaluate how these tools affect psychological safety, engagement, and decision-making satisfaction.

The researchers identified a critical support paradox. While the AIGC condition fostered a more flexible and satisfying atmosphere for minority members, the AIMM condition—despite increasing the volume of minority contributions—unexpectedly eroded their psychological safety and satisfaction. Participants reported that when their ideas were revoiced by the AI, they felt a loss of agency and authorship, leading to a sense of marginalization. Conversely, the AIGC condition successfully normalized dissent without attaching it to specific individuals, which helped maintain a more inclusive group dynamic.

This research challenges the assumption that anonymity-by-mediation is a straightforward solution for protecting minority voices. It highlights that while technology can increase the frequency of participation, it can also diminish the perceived legitimacy and ownership of those contributions. The study provides actionable design implications, suggesting that AI should act as a supportive scaffold rather than a substitute for human voice, and that designers must carefully balance anonymity with the need for relational acknowledgment to avoid inadvertently silencing the very individuals they aim to empower.