The Oldest Test in Education Is Making a Comeback

Socrates Didn't Grade Essays

Formal oral examination is older than the university itself. Socratic dialogue in ancient Athens was built on a simple idea: understanding shows itself in conversation, in the back-and-forth of a question you didn't expect. Medieval universities inherited that idea directly. The viva voce, Latin for "living voice," became the standard way a scholar defended a thesis, because a written argument alone couldn't prove the person in front of you actually built it. Oxford and Cambridge kept the tutorial and the oral defense at the center of undergraduate life long after most of the world moved on. American higher education, by contrast, built itself around the written exam and the machine-graded test as class sizes grew and grading needed to scale.

That was the trade the modern system made, mostly without noticing it was a trade. Written assessment is efficient, standardized, and easy to grade at volume. What it quietly gave up was any guarantee that the person who turned in the work is the person who understood it.

The Fallback Teachers Have Always Used

Long before AI, and even before the internet made copying easy, oral questioning was already the tool teachers reached for whenever they suspected a student hadn't done their own work. I saw this firsthand, early.

I was teaching in a one-to-one laptop program starting in 2001, years before plagiarism was a widespread classroom concern. In 2002, a student turned in a paper that clearly wasn't written in his own voice. The tool I had at the time was Ask Jeeves. I took the first paragraph of the paper and pasted it in, and found it word for word in an Encyclopedia Britannica entry. The student hadn't even bothered to change the wording.

Instead of just failing him, I asked him to stay after class and talk through the paper with me. He couldn't answer any of my questions coherently, because he hadn't written it. He'd copied and pasted it.

That instinct, asking a student to explain their work when something feels off, has always been the teacher's fallback. It just never had a name, a system, or a way to happen for every student on every assignment. It was something you did after the fact, one student at a time, when you were suspicious. What's changed now is the scale of the suspicion. AI hasn't created a new problem. It's made an old, occasional problem universal.

The One Place That Never Gave It Up

There's one stage of education where oral defense never went anywhere: the doctoral dissertation defense. A PhD candidate can write five hundred pages, but the degree isn't conferred until they sit in a room and defend that work, live, answering questions they haven't seen in advance. Every field that treats original, independent understanding as the whole point has landed on the same mechanism. Committees don't require a defense as tradition for its own sake. They require it because it's the only format that reliably tells apart a student who did the thinking from one who assembled a document that looks like thinking.

For decades, that was treated as a PhD-level luxury: too slow and too expensive for anywhere else. Thirty students at five minutes each is two and a half hours of one-on-one time, repeated every unit, forever. No K-12 teacher or large-lecture professor could make that math work.

Then Writing Stopped Being Proof of Anything

Generative AI broke the trade completely. A student can now produce a flawless essay, a clean lab report, or a well-argued term paper without having done, or even understood, the underlying thinking. Written work, the proxy the whole system was built on, stopped reliably showing whether learning happened.

Higher education has noticed fast, and the response is a genuine return to the oldest method in the field. Cornell biomedical engineering professor Chris Schaffer added an "oral defense" requirement to his class after noticing take-home work coming back flawless while students couldn't explain it in person. His point is simple: you can't AI your way through a live conversation. At the University of Pennsylvania, Emily Hammer now pairs written seminar papers with an oral defense, and she's clear that the goal isn't catching cheaters. She frames it as students losing cognitive capacity and creativity, not as an integrity crackdown. Her department describes it as part of a much broader shift toward in-person assessment at Penn. At NYU's Stern School of Business, Panos Ipeirotis built an AI-powered oral exam agent for his product management course and now wants to attach an oral defense to every written assignment he teaches. He's said plainly that he no longer trusts written work as proof of real thinking.

This isn't limited to elite research universities. The same reporting found the shift accelerating across smaller programs and community colleges too, and engineering researchers at UC San Diego have spent several years studying how to make oral exams work at real class scale, not just in small seminars. The pattern is the same everywhere: schools that once treated oral defense as too expensive to run outside a dissertation committee are deciding they can't afford to skip it anymore.

Why It Actually Works

Oral assessment doing more than stopping cheating isn't a new idea. Plenty of research on it predates AI entirely, and the findings are consistent:

• Explaining your reasoning out loud uses a different cognitive process than retrieving a memorized or copied answer. The idea goes back to Vygotsky: putting a thought into words is part of how it becomes genuinely your own thought, not just a report on someone else's.
• Structured oral assessment reaches inter-rater reliability comparable to standardized written tests, nowhere near the coin-flip reliability people assume "just talking to a student" would produce.
• Multiple studies, including a widely cited meta-analysis spanning biology, mathematics, nursing, and law, find students score as well or better on oral assessment than written, with no evidence it disadvantages any group. Some findings suggest first-generation students respond to it with more motivation, not less.

None of this is new information. Until recently, nobody needed to relearn it, because writing was still a decent proxy for thinking.

The Real Obstacle Isn't Pedagogy, It's Logistics

Every professor mentioned above is describing the same bottleneck: oral exams work, and they don't scale. Hammer, Schaffer, and Ipeirotis are each hand-building their own workaround, whether that's office hours, an AI chat agent, or one-on-one slots squeezed into a semester, because the tooling to run oral assessment at normal class size hasn't existed. That's also the real reason "show and tell" disappeared from elementary classrooms decades before ChatGPT existed. It was never a weak idea pedagogically. It was an idea that ran out of minutes in the school day.

This is precisely the gap SayVeritas is built to close. It follows the same tradition Cornell, Penn, and NYU are independently rediscovering, that a student's understanding has to be spoken, not just submitted, but it's built for the exact constraint those professors are currently solving one office hour at a time. Students record a short spoken defense of their work. AI transcribes it, asks a follow-up question based on what they actually said, and produces a first-pass score against a rubric. A teacher reviews the flagged and exceptional responses and makes the final call. What used to require one professor in one room with one student at a time now runs for an entire class, asynchronously, with the teacher spending review minutes instead of exam hours.

The name is a deliberate nod to that lineage. Veritas is Latin for truth, and viva voce, the living voice, is the exact mechanism NYU, Cornell, and Penn are quietly reaching back for. The tools are new. The idea that understanding has to survive being spoken out loud, in the moment, to someone who might ask a question you didn't expect, is as old as Socrates. It turns out it was never actually broken. It just couldn't be sustained at scale. Now it can.

Sources: Associated Press reporting on oral exams at Cornell, Penn, and NYU (March to April 2026); The Conversation, "Oral exams are making a comeback to stop AI cheating" (August 2026).