In late 2025, a computer science professor at New York University’s Stern School of Business faced a reckoning. Even though the students in his Artificial Intelligence / Machine Learning Product Management class were submitting high-quality assignments, when in class most of them could not talk deeply about the material or their assignments.
It was clear to the professor, Panos Ipeirotis, and his teaching partner Konstantinos Rizakos, that their students were outsourcing their homework to AI and, in doing so, had not put in the effort to learn the material.
“We have zero problems with students using AI for their work,” Ipeirotis wrote on his blog. “But here’s the distinction that matters: using AI to enhance your thinking versus outsourcing your thinking entirely and learning nothing at the end. One of these is education. The other is expensive credential theater.”
His solution? Bring back oral exams, where students would need to discuss their work, justify decisions, and reason and think critically in the moment.
The problem was that oral exams are time consuming and, therefore, expensive. With more than 30 students in the class, it would take weeks to plan, organize, schedule and conduct the exams. That logistical headache is why “oral exams don’t scale, and why we abandoned them in the first place,” he wrote.
To solve the problem, Prof. Ipeirotis constructed an AI voice agent to conduct live oral exams with each student. The agent recorded each exam, transcribed it, and then shared it for grading with two other AI agents that were programmed with academic standards that Ipeirotis and Rizakos used when grading assignments.
It wasn’t perfect, but the result was a basic, cheap oral exam administration tool that produced remarkably accurate grades – and kept students honest about their academic work.
Why Oral Exams Matter More in the Age of AI
Once, oral exams were the standard. From Plato’s defense of Socratic dialogue to Lev Vygotsky’s social development theory, educators and researchers have long known that language, reasoning and listening are intertwined. For centuries, education relied on dialogue, debate and rhetoric. This type of learning is mentally taxing but can have a positive long-term influence on student learning.
However, oral exams are expensive and cumbersome, so education eventually dropped them in favor of paper-and-pencil exams, essays and homework as the primary forms of assessment. AI has thrown a wrench into that approach. When students can turn to free, easily-accessible AI platforms to produce polished work, the traditional “paper trail” of using homework and projects to help document what a student has learned no longer holds.
Oral exams can fill the gap. They are a great way to assess whether real learning has occurred because they force students to demonstrate in a face-to-face conversation their understanding of the concepts and principles they have been studying.
According to Joanna Woronkowicz, a professor at Indiana University who once taught in Denmark, oral exams are still common in Europe. She wrote on her Linkedin page that “oral exams not only help address challenges created by AI, but they close the loop on teaching in a way most written products simply can’t. They also offer something uniquely rewarding as an instructor—the chance to sit with a student and hear, in their own words, what they’ve taken away from the course.”
Capturing Thinking, Not Just Answers
A traditional paper-and-pencil test only captures a final output, such as an essay or a multiple-choice answer sheet. While these materials have come to stand as a proxy for student learning, they reveal little about a student’s internal thought process. Oral exams provide a way for teachers to see how their students think and react in real time.
For educators who want to help students sharpen their critical thinking skills, oral exams can be an effective strategy. Instead of checking to see if students arrived at the “right” answer, a professor could use an oral exam to assess how well their students are able to connect disparate concepts, how they handle a counter-argument, or how they recover when a flaw in their logic is exposed.
While a master’s student in computer science at Georgia Tech, Ray Hung developed The Socratic Mind, an AI-powered oral assessment platform that uses Socratic questioning to challenge students to showcase their understanding by explaining, justifying and defending their work.
The effort won a $50,000 prize from the Learning Engineering Tools Competition, one of the largest ed tech competitions in the world. In 2025, the platform was piloted with 2,000 students across several classes at Georgia Tech. The resulting research study, published in July 2026, found that not only did students who used the platform learn more, they also improved their critical thinking skills. According to the research report: “Thematic analysis of qualitative feedback revealed substantial perceived improvements in higher-order thinking skills, including problem-solving, critical thinking, and self-reflection.”
In a 2024 article, Hung said that’s exactly what he wanted his platform to do: “I wanted to promote an education paradigm in which critical thinking, material understanding, and communication skills play integral roles and are at the forefront of our education.”
The Scaling Problem Educators Have Always Faced
The biggest problem with using oral exams to prevent AI cheating and to promote deeper learning is their expense. In addition to time needed for the exams themselves (typically 20 to 30 minutes each) educators must spend hours developing the grading criteria, scheduling (and often re-scheduling) exams, assessing the results and providing feedback.
Jason Linn, a lecturer in the History Department at Cal Poly in San Luis Obispo, Calif., gave 20-minute oral exams to each of his 70 students – 23 hours worth of live assessments, a process he said “tested my own mental stamina and ability to think on my feet.”
That’s why, despite their pedagogical superiority, oral exams are used primarily today at small, elite schools, where the numbers are manageable, or they are used for graduate defenses or small seminars. Educators at bigger schools with larger courseloads are forced to use more standardized forms of assessment – like multiple choice tests, or in class essays – simply because they are cheaper and manageable.
AI systems can help eliminate those bottlenecks.
The system Prof. Ipeirotis set up at NYU conducted 36 oral exams over nine days for about $15 in AI costs.
The Student Experience: Anxiety, But Results
Traditional final exams can feel like high-stakes, make-it-or-break-it events. And they can be. Students have reported that oral exams can feel even more pressurized, given their unfamiliar format and the face-to-face questioning.
In a survey after Prof. Iperirotis’ course ended, 93 percent of his students said they found the oral exam “more stressful” than a traditional exam. Nevertheless, 70 percent of them also agreed that the oral exam tested their actual understanding of the material better than a written test.
Jason Linn, the lecturer at Cal Poly, reported that several of his students “appreciated how much shorter an oral exam was compared to the traditional multiple-choice test.”
Further, a few of his students said participating in the oral exam over Zoom helped ease their anxiety. One student wrote to him, “Having it on zoom put me a bit at ease as it allowed me to study right up until the time to meet and didn’t leave the limbo time of getting to school to think and stress about the exam.”
For students who excel at writing or test-taking, an oral exam can seem like a disadvantage. However teachers who have conducted oral exams say those issues mostly resolve themselves once students become familiar with the new format.
“Here is the delicious part,” Prof. Ipeirotis wrote. “You can give the whole setup to the students and let them prepare for the exam by practicing it multiple times. Unlike traditional exams, where leaked questions are a disaster, here the questions are generated fresh each time. The more you practice, the better you get. That is actually how learning is supposed to work.”
What Educators Should Take Away
provides the inexpensive, easy-to-use infrastructure needed to restore explanation, debate, and verbal reasoning to learning. Furthermore, because an AI oral exam agent can generate fresh questions for every interaction, students can use the platform for low-stakes practice.
In traditional schooling, leaked exam questions undermine evaluation. In an adaptive oral framework, endless practice with dynamic questions is simply called studying. The more a student practices explaining, defending, and revising their ideas out loud, the more deeply they learn.
By Helen Lee Bouygues, President of the Reboot Foundation
