LLM-powered system boosts minority voices in group decisions — with a catch
AI counterarguments increase satisfaction; AI mediation boosts participation but hurts safety.
Minority viewpoints often get suppressed in hierarchical group settings due to social pressure. To address this, researchers developed an LLM-powered system that intervenes either by generating counterarguments from the minority perspective or by mediating messages between members. In a mixed-method experiment with 96 participants across 24 groups, they compared baseline discussions against these two AI interventions.
The results reveal a nuanced trade-off: AI-generated counterarguments created a more flexible atmosphere and enhanced minority satisfaction, while AI-mediated messaging increased how often minority members participated but unexpectedly lowered their psychological safety. The study, accepted at CSCW 2026, provides empirical evidence on how different AI implementations affect group dynamics, identifies a critical support paradox between participation and safety, and offers design implications for more equitable AI support in hierarchical decision-making.
- 96 participants in 24 groups tested AI interventions for supporting dissenting minorities in power-imbalanced decisions.
- AI-generated counterarguments improved satisfaction and flexibility; AI-mediated messages increased participation but reduced psychological safety.
- Research accepted at CSCW 2026 highlights a 'support paradox' between participation and psychological safety.
Why It Matters
LLMs can amplify marginalized voices in meetings, but design choices critically impact user safety and satisfaction.