LLM-powered Social Mediator (in progress)

As part of a group research project, I explored how AI could support student teams when subtle but harmful comments occur during collaboration. The broader project focused on microaggressions in group work, and how an AI mediator might help teams acknowledge these moments instead of letting them pass unnoticed.

My individual research focused on the input side of this mediator. I compared an audio only AI mediator with a version that also included a private chat window, allowing students to personally share context, feelings or concerns with the system.

Using a Wizard of Oz prototype, I tested both versions through scripted roleplays with student groups. The research showed that personal text input can give students more agency, especially when speaking up in the moment feels difficult. At the same time, it introduced risks, such as distraction, suspicion and the feeling of “snitching”.

The final result was a proof of concept prototype and design insights for future AI mediation systems. The main takeaway was that personal input can be valuable, but only when it is designed in a way that keeps the group interaction socially safe.