Fair and diverse perspectives discussion, aided by reliable AI intervention, can effectively burst filter bubbles and transform undemocratic behaviors into functional democratic processes in online forums. However, the impact of AI intervention on individuals may vary depending on the persona and social attributes of the AI. Here we conducted a two-timepoint online experiment to examine whether AI-mediated discussions on sensitive topics can form moderate perceptions. Our study focused on the impact of AI-intervention on participants' perceptions and attitudes towards engagement with AI technology, access to democracy and gender equality, and interethnic marriage rights in Afghanistan. We assigned 168 individuals with conservative, moderate, and liberal perspectives to three different settings in two experimental groups, with and without AI-mediation. Participants engaged in three-day online discussions, after which we measured their personality profiles through a self-reported survey. Our findings suggest that while AI intervention had a significant impact on participants' perceptions of AI chatbot as facilitator, but the participants' positions remained largely unaffected towards access to democracy and gender equality, and interethnic marriage rights, possibly due to the lack of contextual knowledge and human touch in the chatbot's responses. Our study provides insight into creating effective chatbot personas to address illiberal practices in online discussions.

Assessing the Impact of AI-mediation on Changing Extreme Perception in Online Discussion: Post-Intervention Responses in Post-2021 Afghanistan