The $0.42 Solution to Academic Cheating

Image source note: The image is AI-generated, and the image licensing service provider is Midjourney

When NYU professors Panos Ipeirotis and Konstantinos Rizakos noticed suspiciously polished assignments in their "AI/ML Product Management" course, they didn't expect a simple oral exam would reveal such glaring knowledge gaps. Students who submitted flawless papers stumbled through basic explanations of their own work.

How AI Became the Ultimate Teaching Assistant

The professors turned this discovery into an innovative assessment method using ElevenLabs' voice AI technology. Their two-part oral exam first asked students to defend project decisions, then randomly quizzed them on course material. Over nine days, 36 students completed the 25-minute exams at a total cost of just $15 - cheaper than a pizza delivery.

"At first, students complained the AI sounded like a stern professor," Ipeirotis admits. Early versions sometimes fired multiple questions simultaneously, creating confusion. After tweaking the system, the virtual examiner became more conversational while maintaining rigorous standards.

The Grading Revolution

Scoring presented another challenge. Using Claude, Gemini and ChatGPT to evaluate responses initially produced inconsistent results. "It was like having three teaching assistants who never agreed," Rizakos jokes. By having the AIs cross-check each other's assessments, they achieved remarkably consistent final grades.

While 70% of students acknowledged the tests effectively measured true understanding, many found them more stressful than written exams. "You can't bluff an AI," one participant noted. "It immediately spots vague answers and asks follow-ups."

Beyond Cheat Detection

The experiment revealed unexpected benefits beyond catching academic dishonesty. "Some students clearly understood concepts but struggled to articulate them," Ipeirotis observes. "Now we know where to focus teaching efforts."

The professors believe AI oral exams could become standard practice, especially for technical courses where practical understanding matters more than polished writing. At 42 cents per test, they're also solving the eternal problem of academic budget constraints.

Key Points:

• Written vs Verbal Discrepancy: High-scoring assignments often didn't reflect actual comprehension when tested verbally
• Budget-Friendly Innovation: AI proctoring slashed oral exam costs from hundreds to mere dollars per class
• Stress with Purpose: While more intense than written tests, most students recognized the method's effectiveness
• Teaching Insights: The exams identified not just cheating but genuine learning gaps needing attention