Particle.news
Download on the App Store

AI in Class Changed How Students Learn, Not How They Score

A controlled two-section trial at UMass Amherst found structured AI access boosted engagement, confidence, efficiency without raising proctored exam performance.

Overview

  • Researchers taught two back-to-back sections of the same upper-division antitrust economics course with identical lectures, assignments, and paper-and-pencil proctored exams that prohibited notes and technology.
  • One section was encouraged to use tools like ChatGPT with explicit guidance and disclosure requirements, while the other was barred from AI and received parallel non-AI support, with the AI permission intentionally assigned to a historically lower-performing afternoon section.
  • Proctored exam scores and final grades did not differ meaningfully between the sections despite the intervention.
  • Students allowed to use AI reported more real-time participation, longer and more substantive AI study sessions, habits like editing outputs and catching errors, a preference for their own answers, and less time spent on homework and exam preparation.
  • Standardized course evaluations were higher in the AI section and more students expressed interest in AI-intensive careers, as the authors recommend a permit-with-scaffolding approach while cautioning about the small sample, self-reported measures, and the study’s working-paper status on SSRN.