Brown University economics professor Roberto Serrano writes that after allowing a take-home midterm in his welfare economics course, the median score came in at 98%, with 40 of 86 students scoring 100%. When he switched the final exam to an in-person, no-AI format, 27 students dropped or skipped it and the remaining students’ average fell to 48.6%, with three students scoring zero. Serrano argues the gap is strong evidence of widespread undisclosed AI use on the take-home exam and calls on universities to redesign assessments around in-person, process-based evaluation rather than relying on AI-detection software.
