AI facilitation and affective dynamics of online deliberation

Abstract Artificial intelligence systems are increasingly used to facilitate online group discussion, yet their influence on the affective conditions of collective interaction remains poorly understood. Existing research has largely focused on participation, efficiency, or content quality, leaving the emotional dynamics of AI-mediated deliberation underexamined. This paper investigates how AI facilitation relates to the expressed sentiment in online group discussion, understood as one measurable dimension of its affective dynamics. Drawing on two empirical studies conducted in different discussion settings in Afghanistan, one involving a rule-based facilitator and the other a large language model-based facilitator, we analyze differences in expressed sentiment between AI-facilitated and non-AI discussions. Across both datasets, participant sentiment was modestly higher under AI facilitation, although the mixed-effects analyses did not establish statistically significant condition differences. In addition, AI-generated facilitation messages exhibited higher sentiment than participant-authored messages, with a statistically reliable overall gap in both studies and a consistent direction across all five city-specific boards in the first study and all four AI-facilitated groups in the second, suggesting that AI facilitators introduced more positive affective signals into the discussion environment. Rather than treating increased positivity as an unqualified improvement, we interpret these findings as suggesting that AI systems may act as affective mediators, potentially influencing the emotional norms and tone of collective interaction. This raises normative questions about emotional regulation, power, and democratic deliberation, particularly regarding whether affective smoothing may constrain disagreement or reconfigure deliberative agency. By foregrounding affect as a dimension of AI-mediated discussion, this study contributes empirical evidence to debates on the societal implications of AI facilitation and highlights the need for greater attention to the affective norms and expressive patterns introduced by AI facilitation.