AI rise in classroom pushes professors back to old-style oral exams

Panos Ipeirotis, a data science professor at New York University's Stern School of Business, said he grew concerned that student assignments sounded polished but showed little real understanding.

When asked to explain their work in class, many students struggled, leading him to conclude that written submissions no longer accurately measured learning, he wrote in a blog published earlier this month and cited by Business Insider.

To address this, Ipeirotis revived oral exams and used an AI agent to conduct them at scale, in an attempt to "fight fire with fire."

"We need assessments that evolve toward formats that reward understanding, decision-making, and real-time reasoning," Ipeirotis wrote in his blog.

"Oral exams used to be standard until they could not scale," he added. "Now, AI is making them scalable again."

Using ElevenLabs’ conversational technology, he set up an AI examiner that questioned students about their capstone projects and tested their ability to analyze course cases in real time.

Over nine days, the system examined 36 students at a cost far lower than that of human-led oral exams. AI was also used for grading, with three large language models independently evaluating transcripts and then synthesizing a final grade. He said the AI grading was more consistent, stricter, and fairer than human grading, and provided high-quality feedback.

Student reactions were mixed, with many finding the exams stressful despite acknowledging their effectiveness. Ipeirotis said the experiment demonstrated how learning should work and warned against students relying too heavily on AI instead of using it to enhance their thinking.

Students at New York University, December 2025. Photo from the univeristy's Facebook page.

At the University of Wyoming, students in Professor Catherine Hartmann’s Religion course no longer write essays or take computer-based multiple-choice tests. Instead, they must complete a 30-minute face-to-face dialogue, according to The Washington Post.

Hartmann began overhauling her assessment methods in 2024. She requires students to avoid using electronic devices in class and to engage actively in discussions throughout the semester. Before the exam, students receive a list of core concepts and are tasked with explaining and debating these ideas convincingly in front of the professor to demonstrate deep understanding.

This trend is spreading across North America.

Mark Chin, a professor at Vanderbilt University in Tennessee, organized his first oral exam for a data science course earlier this month, believing oral exams are an effective way to test whether students are truly learning the material.

Speaking to The Post, he said he is well aware that AI excels at handling the programming assignments he gives. During the exam, he shows students examples of code written in the programming language they have studied and asks them on the spot to explain its function.

In a post discussing this approach on his LinkedIn account, Chin wrote in a comment that "I feel like there was definitely anxiety but also I felt that students appreciated the format for being able to talk through their thinking (and get credit for it) in a way that sometimes isn't possible for written work."

The approach is not limited to small classes. At Western University in Ontario, Canada, professors have conducted oral exams for a business class with 600 students. The University of California, San Diego has applied the format to six large-scale engineering courses.

According to a survey released last August by Inside Higher Ed, 85% of students admit using AI to brainstorm ideas or prepare for short quizzes, while 25% use it to complete assignments. To combat cheating, some professors have turned to AI-detection software but found the results were not always reliable.

Professor Hartmann said that in the past, every time she graded essays, she felt like a "detective" rather than a teacher, searching for AI-generated content. "I don’t like that adversarial relationship," she told The Post.

Since switching to oral exams, both professors and students have reported positive effects. After 11 rounds of oral exams, Professor Chin found that although students were initially nervous, they were able to answer questions that had stumped them months earlier. He believes learning a new and challenging skill is both the goal and the greatest reward of this testing method.

In an interview with CBC last December, Kyle Maclean, an assistant professor at Ivey Business School in London who researches the pros and cons of oral exams at the university level, said one drawback is that grading can be less uniform, as assessments often focus on a student's reasoning process and their ability to explain the rationale behind their answers.

"It also depends on the subject and discipline. That's what makes oral exams difficult to scale. But you do get a strong sense of how much someone understands the material," he said.

Maclean added that the time commitment is another challenge, as it can take up to 20 hours to assess all students. However, he believes oral exams are an "effective" solution for limiting inappropriate AI use and sees a future with more oral exams as universities move toward "smaller classes, more project-based learning, and deeper conversations between faculty and students."

Commenting on student reactions, he told CBC: "Students didn't like it. Oral exams are stressful. They require students to think on their feet, without rehearsing answers. And when it's live, they're speaking directly to a faculty member they respect, which adds pressure."

A study released in September in the academic journal "Assessment & Evaluation in Higher Education" described how AI has transformed student assessment into a "wicked problem." The researchers based their findings on interviews with 20 unit chairs at a major Australian university conducted in late 2024. In hourlong Zoom interviews, instructors reported feeling overwhelmed by increased workloads, uncertainty over how AI should be used, and a lack of consensus on what constitutes an AI-resistant form of assessment.

Faculty opinions were divided. Some participants argued that AI should be embraced as a skill students need to develop, while others viewed it as a form of academic misconduct that undermines learning. Many acknowledged uncertainty about the best way forward.

In a May episode of his podcast Possible, LinkedIn cofounder Reid Hoffman suggested that formats such as oral exams are harder to fake and require deeper, more comprehensive understanding, making them more suitable in an AI-rich environment.

He also noted the challenges of the method: "Part of the reason why oral exams are hard, generally reserved for Ph.D. students, sometimes master's students, etc., is because actually, in fact, to be prepared for oral exams, you've got to be across the whole zoom."

Hoffman does not support excluding AI from education. Instead, he believes it should be incorporated into curricula in ways that support learning rather than undermine it. He suggested using AI-generated work as examples of weak or superficial responses to help students understand what higher-quality work looks like.

Teachers have a responsibility to understand and use AI, as ignoring it harms both instructors and students, he said.

"The most central thing is preparing students to be capable, healthy, happy participants in the new world," he said. "And obviously your ability to engage with, deploy, leverage, utilize, AI -- AI agents, etc. -- is going to be absolutely essential."