NYU professor creates AI exams to ‘fight fire with fire’ over ‘suspicious’ student work

by

A professor at New York University used artificial intelligence to proctor an exam, responding to a rise in potentially plagiarized or otherwise questionable student submissions by using the same technology.

As artificial intelligence becomes more common, it’s changing how we think about education. Many students are starting to depend on AI tools like ChatGPT, using them for help with assignments – and sometimes even letting the AI complete the work for them.

In late 2025, NYU business professor Panos Ipeirotis started to address a growing problem he observed in student work. He noticed that some assignments were exceptionally well-done – not just good for high-achieving students, but remarkably better than expected.

To get a better sense of how well his students really understood the material, the professor decided to use AI as a tool during an exam. He had students interact directly with AI chatbots, which allowed him to assess their knowledge and skills.

Professor uses AI voice bots for exam

Initially concerned about students using AI, the professor started unexpectedly calling on students in class. While many turned in high-quality assignments, some couldn’t explain their reasoning when questioned about fundamental decisions they’d made.

According to a recent blog post, students can now easily use AI tools to answer most exam questions and even create impressive presentations on topics they haven’t fully grasped. This poses a challenge to traditional assessment methods.

This prompted the professor to switch to oral exams, where students have to think on their feet and answer questions without relying on notes, textbooks, or tools like AI. As Professor Ipeirotis put it, these exams require students to reason in the moment, apply their knowledge to new situations, and explain their thinking.

A professor used ElevenLabs Conversational AI to create a custom quiz for his students. The AI bot personalized each assessment by using the students’ names and specific details about their projects, tailoring the experience for all 36 participants.

I was really fascinated to see how people interacted with the AI. On average, students spent about 25 minutes chatting with it, working through different case studies and answering its questions. It was cool to watch them figure things out!

After the conversations, three AI programs reviewed the transcripts and assessed the students’ performance. Remarkably, the quickest assessment—lasting only nine minutes—yielded the best score, a 19 out of 20.

Honestly, I was really surprised by the results of this experiment! Professor Ipeirotis said the feedback from the AI was actually better than what a human could give. But, and this is a big but, most students – 57% of them – actually preferred the old-fashioned written exams. They found the AI feedback way more stressful, which is something I hadn’t really considered.

Okay, so pretty much everyone thought that chatting with the AI was a really good way to see how well we actually understood the stuff we’d been learning. At the end of it all, the professor basically said the idea itself is solid, but they need to tweak how it actually works to make it even better.

Read More

2026-01-06 06:19