AIGAC Debrief: AI Innovation in the Classroom, lessons from Fuqua

Across the country, educators (individually and institutions) are responding to AI in wildly different ways. Some schools have banned it outright, forcing a return to handwritten exams to sidestep concerns about cheating. Others have quietly handed students a vague policy and moved on. A few are building entire AI-focused curricula from scratch. The experiences are fractured, and the guidance is inconsistent at best.

So, when we encountered something genuinely different at Duke's Fuqua School of Business, it was worth pausing.

Scott Dyreng, Senior Associate Dean of Innovation at Fuqua, is tackling a question that most institutions haven't thought to ask: not how do we stop AI from undermining the classroom, but how do we use AI to make in-person learning better? His team began by asking whether AI could strengthen live, in-person discussion rather than replace it, with the goal of turning classrooms into dynamic environments where technology amplifies human connection.

We got to see the experiment in action last week as part of AIGAC’s 19^th Annual Conference. Fuqua's system uses an expanded network of microphones and cameras, combined with an internally developed check-in app, to produce time-stamped transcripts that accurately attribute who said what during class discussions. In early tests, accuracy has exceeded 90%. That transcript is then analyzed by an in-house AI system. Professors receive data on participation patterns — which students are driving conversations, which comments are generating meaningful follow-on discussion, and even which side of the room a professor tends to call on most.

Students, in turn, receive individualized reports on their contributions, with the system generating prompts to encourage stronger preparation for upcoming classes when engagement has been weak.

The ambition is significant. Dyreng describes it as giving every professor the equivalent of a full-time teaching assistant in every class, one that ignores appearances and accents, doesn't get tired or distracted, and can spot patterns that humans routinely miss. In his presentation, he shared many ways that the system can spot patterns from teachers as well, coaching them to engage with a class more effectively.

The system is certainly controversial, sparking feedback from students (though notably, as Dyreng pointed out, not from students who had been in the class that beta-tested the technology). A March 2026 op-ed in the Duke Chronicle argued that the system transforms the classroom from a space of intellectual risk-taking into a monitored performance arena, with research on surveillance in education consistently showing chilling effects on student speech. Some admitted students, upon hearing about the system's planned expansion to all Fuqua core classes, found it unsettling.

What struck us wasn't that the debate doesn't exist — it does, and it should. What struck us was the thoughtfulness with which Dyreng and his team have engaged with dissenting opinions. Privacy has been a guiding principle: students in the beta phase opted in, and the output is available only to the professor, used solely for teaching and learning. Dyreng has been transparent about the system through public writing and presentations, and he's been willing to sit with the hard questions rather than dismiss them.

His core argument is that AI technology doesn't have to replace face-to-face discussion. Used thoughtfully, it can make human connection more meaningful and human judgment sharper.

Whether or not every institution should follow Fuqua's path, this is the kind of engaged, values-driven experimentation that separates genuine innovation from either reflexive fear or uncritical adoption. It's a model worth watching and that we’re going to guide discussion about through AIGAC’s Summer With the Schools program that Andrea has coordinated for seven years.