Could AI make Duke business school students more human?
AI-generated summary reviewed by our newsroom.
- Duke Fuqua pilots AI that records a classroom and group meetings to assess participation.
- Students and a professor said AI feedback changed behavior and increased preparedness.
- A Fuqua spokesperson said recordings are anonymized and stored behind Duke firewalls.
If an artificial intelligence agent informed you that you were talking over your peers too often in group discussion, would you heed its advice?
Some Duke business school students do.
It’s part of an experiment in which Duke professors and students let AI record them work and learn, and then inform them about how they could improve their interpersonal dynamics and participation.
In one classroom at Duke’s Fuqua School of Business, an artificial intelligence agent is constantly recording. The classroom is home to courses on topics like taxes and accounting. The agent knows who sat where and said what during class, and uses the recording to evaluate the quality and quantity of individual students’ participation, determining which comments pushed discussion the furthest, and helping the professor evaluate the students.
The students also record their group project meetings outside of class, and the AI agent assesses the discussion for balance, productivity, and creativity. The professor can then review the AI’s analysis of group dynamics.
Four Duke MBA students and one professor spoke to The News & Observer about how this use of AI has affected their learning experience. In the so-called “AI classroom,” students said the tool feels less like surveillance and more like accountability.
Still, student Jenny Laurence, from London, said there is a “teething pain of getting used to the experience of being watched.” Students might be less inclined to engage in small talk with each other or chat about topics outside of the class material, said second-year student Samantha Low.
They pause the AI recording if they want to complain about the assignment itself.
Phillip Coale, a third-year MBA student and Marine veteran, described the kind of feedback that AI gives after it listens to group discussions.
“’You guys are going from ‘interesting point’ to ‘good enough’ to ‘done,’ and you’re not really getting to the best solution,’ [the AI said]. That was actionable feedback that helped us change our behavior in the next meetings,” Coale said. “I noticed folks showing up differently. If it said they only spoke for 5% of the time, they came prepared because it had been highlighted that they were not contributing in the same way.”
Coale said the AI’s analysis of a moment of conflict on their team forced certain members to evaluate how they were being perceived, and made the group confront that not all of them were on the same page, even after the meeting had ended.
“It’s just so much easier to get that feedback from an unemotional tool,” he said. “It’s not a person that you feel is accusing you of something.”
Duke business professor Bill Mayew echoed Coale’s sentiment. He says he can “let AI rock the boat,” by providing feedback that students may not be comfortable giving each other about how controlling they are or how much they talk over one another.
Mayew said that reviewing the AI summaries of groups’ discussions allows him to call out specific students to share their thoughts in class, because he knows which teams had polarizing ideas on an issue — thanks to this window into his students’ thought processes he never had before.
“It’s a professor scaling their impact,” Coale said. “If he had two students, he could spend that kind of personal time with them. When you have 70 students, he can have way more of an impact with this tool.”
The recordings of class and of group discussion are preserved but anonymized, and all the data stays safely behind Duke’s firewalls, according to Duke Fuqua spokesperson Ashley Cimino. All personal information is anonymized before it’s analyzed by AI, she said.
Mayew built a custom AI tool that students can use for the course, which he sees as a kind of course companion or evolving textbook. He’s trained this custom ChatGPT model on his own notes and case studies, and has instructed it never to provide answers to students, but rather to help them get to the answers themselves, he said.
“When you leave Fuqua, it used to be that you’d have some slide decks and notes you took,” Mayew said. “Now, we have a full history of the course, every lesson. It’s like a custom website that has all the topics, all the questions students asked — not with names, but just ‘Hey, there were questions about this. Here was a controversial thing, here was the resolution.’ They can walk away from Fuqua with a document that is searchable five years from now, and almost relive the class.”
Plus, Mayew says that the AI could give students quantifiable data about their interpersonal skills that they could present to potential employers down the line.
Research from elsewhere in Duke, however, appears to provide a foil to the experiences of these business students. A recent study by Duke computer science researcher Emily Wenger concluded that groups of people are more creative than AI, as large language models tends to provide homogenized creative outputs that are similar among different models.
For these Duke business students, however, they feel like they’re participating in something truly new.
“A lot of times, people talk about: ‘What will AI be like, what will society look like?’” said Coale.
The answer, he thinks? Like this.
“We’re actually doing it, actually experimenting,” Coale said. “We’re not just using the ChatGPT website to ask questions, but using [AI] in a novel way. I had no idea that teams would be strengthened, that our interpersonal connections would be strengthened by AI, but now I know, because we got to experiment with it.”
“I obsess over how we can learn to grow as leaders and as teammates,” Coale said. “And this was the coolest tool I’ve seen at Fuqua to help us do that.”
