NYU Stern pilots AI voice exams, shifting assessment from papers to oral defense

In a pilot at New York University’s Stern School of Business, students submit written assignments that are often indistinguishable from AI‑generated work. Three days after submission, a voice‑based AI agent conducts a ten‑to‑fifteen‑minute oral exam, asking personalized questions based on the student's own text. Transcripts are graded by three separate language models, at a cost of about 42 cents (≈ 0.37 EUR) per student.

The experiment responds to a survey by the Center for Democracy and Technology indicating that 71 % of U.S. teachers find it difficult to verify whether submitted work is genuinely authored by students, while existing AI detectors produce high false‑positive rates. The new assessment model evaluates the student's ability to defend their work rather than the written document itself, highlighting the challenge of maintaining reliable measures of student understanding in the age of generative AI.