A data science professor at NYU Stern School of Business, Panos Ipeirotis, faced a challenge: students' written assignments looked perfect—like McKinsey-style consulting memos—but during oral presentations it became clear that many students lacked a real understanding of the material.They couldn’t explain their own solutions or logic.
A data science professor at NYU Stern School of Business, Panos Ipeirotis, faced a challenge: students' written assignments looked perfect—like McKinsey-style consulting memos—but during oral presentations it became clear that many students lacked a real understanding of the material.They couldn’t explain their own solutions or logic.
To combat the consequences of AI involvement in education, the professor decided to fight fire with fire — reintroducing oral exams in a scalable format powered by AI. Working with a colleague, he created an AI examiner that used ElevenLabs' conversational technology. The setup took only a few minutes and required a single prompt describing what questions the agent should ask.
The exam consisted of two parts. First, the AI questioned students in detail about their final projects and decision-making processes; it then selected a course case study and required students to reason through it in real time. Over nine days the system conducted 36 exams, each lasting about 25 minutes. The computational cost was around $15 per exam — significantly cheaper than in-person oral exams run by teaching assistants.
Grading was also handled by AI. Three models (Claude, Gemini, and ChatGPT) independently evaluated the transcripts, compared scores, and produced a final grade, with Claude acting as the chair. According to the professor, this AI panel proved stricter but fairer than human graders and provided higher-quality feedback. Moreover, the analysis revealed gaps not only in student understanding but also in how course material was presented.
Student reactions were mixed. Most students found oral exams more stressful than written ones but acknowledged that the oral exams were better at assessing genuine understanding. The professor himself sees the core issue as students outsourcing their thinking to AI rather than using it to enhance their own capabilities.
Business Insider notes that this story is part of a broader crisis in knowledge assessment in the AI era. Universities struggle to evaluate students when essays and homework can be easily automated. Educators are perplexed, and industry leaders — including LinkedIn co-founder Reid Hoffman — say that traditional methods no longer suffice; AI examiners and oral comprehension tests may become the new standard.
Релоцировались? Теперь вы можете комментировать без верификации аккаунта.
Неужели теперь со студентов будут требовать знание и понимание, а не домашки и курсовые?
Ну если выпускники будут болваны, которые снижают рейтинг заведения, то очевидно что да, нужно менять что то в датском королевстве
Ну, в бытность мою студентом, когда я помогал товарищам с работами – я всегда добивался того, чтобы человек нормально понимал, что там написано, и мог "защитить" работу. Не вижу проблемы при генерации ии сделать то же самое – просто пишешь "объясни решение" и уточняешь у него непонятные моменты, можно и попросить проверить уже твоё понимание, задавая вопросы.
это надо потратить время то есть всё равно учиться 🧐