Live·Open questions in longevity research
All news
Scientific Computing

Annual oral exams proposed to assess mathematics PhD students’ knowledge when they use AI

15 November 2026· 261115007

Annual oral exams proposed to assess mathematics PhD students’ knowledge when they use AI

In an interview released on YouTube on November 14, Alex Kontorovich, chair of the mathematics department at Rutgers University, proposed assessing PhD students’ knowledge separately from the original results in their dissertations. He considers a written dissertation essential: its author must understand the ideas and be able to explain the proofs.

In Kontorovich’s view, a dissertation traditionally demonstrated two things: the student had produced an original result and mastered the mathematics needed to obtain it. AI can help produce a result, so universities need to assess students’ own understanding of their work separately. In his conversation with physicist Brian Greene, Kontorovich points out that an advisor could always have suggested an idea. Students must work through that idea themselves; he expects the same when they use AI.

The logical validity of a proof can be checked in Lean, a program that takes a proof written in a formal language and checks each step against specified rules. Kontorovich proposes asking students to give a lecture on their topic without preparation and explain any step of the argument in detail. This would allow a committee to assess how well students connect the ideas and understand the details of the proof.

Rutgers University already has an oral qualifying exam and a dissertation defense. Kontorovich proposes annual discussions with a committee that could monitor students’ understanding as they work on their dissertations, independently of their advisors’ assessments. Mathematician Frank Calegari has proposed in-person interviews for PhD admissions and hiring; Kontorovich’s proposal would make assessment a regular part of the PhD program itself.

“We develop understanding,” Kontorovich says of mathematicians’ purpose. He asks us to imagine a future in which AI can prove theorems without human help. People would still need to choose interesting questions, understand the results, and explain them to one another. When hiring, he would therefore want colleagues who know a great deal of mathematics and can teach it to others.

Kontorovich recalls PhD students who had a deep knowledge of entire areas of mathematics, saw connections between them, and could answer difficult questions, yet struggled to produce an original result. In the past, he says, such people left academic research because it required them to produce original results. He predicts that AI could help them write a dissertation, become experts in its subject, and begin an academic career.

Greene asks whom a university would favor under this approach to hiring: a researcher capable of making discoveries but poor at explaining their ideas, or someone with a deep knowledge of the subject who can explain other people’s results. Kontorovich replies, “My hope is that we would hire both.” He expects demand for mathematics education to grow, creating a need for more university teachers and room for both kinds of colleagues.

Originally published on Telegram by Ukhvat NewsView on Telegram
Sources
#ai-assisted-mathematics#doctoral-education#oral-exams#mathematical-proofs#lean#academic-hiring