10/05/2026
For centuries, one of the simplest ways to find out whether someone actually understood something was to sit across from them and ask them to explain it.
Now, as AI makes it increasingly easy to produce polished essays and completed assignments, that old method is making a comeback.
A growing number of college professors are bringing back oral exams and one-on-one “oral defenses” as a way to see whether students actually understand the work they submit. At Cornell, for example, students in one biomedical engineering course must defend their written problem sets through a 20-minute conversation with an instructor. At the University of Pennsylvania, professors are also pairing written work with oral examinations. (The Washington Post)
The reason is straightforward. An AI system can produce a convincing answer, but a student still has to understand that answer when a professor starts asking follow-up questions.
Why did you choose that approach? What does this line of code do? Can you explain this argument another way? What would happen if one of the assumptions changed?
Those questions require students to think in real time rather than simply submit a finished document.
Some educators are even using AI to conduct the oral exams themselves. At NYU Stern, one professor developed a voice-based AI examiner that asks students questions about their projects and follows up based on their answers. The goal is to make oral assessment practical even when instructors have large numbers of students. (The Washington Post)
The shift isn’t necessarily about trying to eliminate AI from education altogether. Many educators recognize that students will use AI in college and eventually in their careers. The bigger question is whether students are using it as a tool while still developing the ability to think, solve problems, communicate and understand their own work.
And oral exams aren’t a perfect solution. They can take considerable time, create anxiety for students, and don’t work equally well for every course or class size. Research and teaching guides describe them as one possible part of a broader assessment strategy rather than a complete answer to AI-related problems. (Worcester State University)
Still, there is something interesting about the return of such an old-fashioned method.