Brown University's Shocking Experiment: Grades Dropped from 96% to 48% When Students Were Banned from Using AI on an Exam
Brown University professor Roberto Serrano banned AI on a final exam after suspecting widespread cheating — the average score plummeted from 96% to 48%, raising urgent questions about student dependence on artificial intelligence.
Introduction: The AI Education Wake-Up Call
In the summer of 2026, Brown University economics professor Roberto Serrano conducted an experiment that sent shockwaves through the academic world. He banned students from using AI on an exam — and the results were devastating. Average scores dropped from 96% to 48%. This is not just one university's story; it's a global warning about how deeply AI is embedded in student learning.
The Brown Experiment: What Happened?
Professor Serrano noticed that the 86 students in his online take-home exam had historically anomalous scores averaging 96%, compared to the historical norm of 65-80%. When he entered exam questions into ChatGPT and received nearly identical answers to what his students submitted, he knew AI was the cause. He moved the final exam to an in-person, proctored format. 18 students dropped the course, 9 didn't show up, and the average score fell to 48.6% — the lowest in course history.
Key Data Points
Take-home exam average: 96% (historical norm 65-80%); In-person final exam average: 48.6%; 19 students failed to meet the minimum threshold; Course dropout rate surged to 21% (historical 5-10%).
Global Evidence of AI Dependence
The Brown experiment echoes larger studies: a Chinese study of 26,000 students found banning ChatGPT dropped GPA from 4.1 to 3.3. UC Berkeley's study of 500,000+ grades found post-COVID online exam scores rose 0.2 GPA points, but in-person exams stayed at pre-COVID levels — confirming AI, not better learning, caused the bump.
Lessons for Georgian Universities
Georgian universities must develop balanced policies: clear AI usage guidelines, mandatory proctored exams, teaching critical AI literacy. As the Brown experiment reveals, blind AI use may undermine education's real purpose — knowledge transfer. The future belongs not to those who use AI best, but to those who can think without it.