Research Seminar — One Hand on the Wheel: Who Steers When AI Joins the Conversation?

AI agents are no longer just tools we query; instead, they are increasingly participants in our conversations, shaping how we relate to one another. What happens when an agent decides who speaks next in a meeting, mediates a family argument, or coaches someone through the vulnerability of a first date? Drawing on a systematic review of 53 group conversational agent designs, this talk maps the emerging design space: the group challenges these agents target, the roles they play, and the tensions—around visibility, proactivity, and authority—that every designer in this space must navigate.

Design vocabulary, however, cannot tell us which values to encode. Through two scenario-based design studies, on AI-mediated self-disclosure in online dating and chatbot support for intergenerational family communication, we show that going deep on a specific social context surfaces tensions invisible from above. Among the findings: a correct interpretation delivered at the wrong moment can damage a relationship more than saying nothing at all; users care about felt authorship---whether a message is genuinely theirs---in ways that "user control" and transparency frameworks miss; and people spontaneously expect chatbots mediating family conflict to exercise moral judgment, not just manage process. These are not edge cases. They point to a question that applies wherever AI mediates close human relationships: when an agent shapes how we communicate, whose values is it enacting? And, did anyone choose that deliberately?

This seminar is moderated by Prof. Huaxin Wei, Associate Professor, PhD Advisor and Specialism Leader of BA Interaction Design.

 

• All PolyU PhD students and SD staff are welcome. No registration is required. Seating is available on a first‑come, first‑served basis.