What's this about?
Over 1,000 first-year Bocconi undergraduates worked on a real business case — marketing recommendations for the university merchandise store — randomly assigned by class period to one of four groups: ChatGPT access, causal-reasoning training, both, or neither. Access to ChatGPT improved quality and coherence. The causal-reasoning exercise, which had nothing to do with AI, made students generate more unique ideas. Students who got both showed both effects, and the effects were largely independent.
What is "causal reasoning training"?
Training in linking cause and effect, and explaining *why* a proposed solution would or wouldn't work — delivered here through a game, examples and questions, with no AI involved. The finding that matters is what it did *not* do: it did not make the work more polished. It made the ideas more distinct. Polish and originality turn out to be separately produced, and only one of them is what the AI supplies.
Why is it worth noticing?
- The study's own conclusion is the actionable one: as AI makes polished answers easy, assignments have to start measuring something else — originality being the obvious candidate.
- It uses GPT-4o, which is now several generations old; the effect sizes may not hold for current models.
- It is OpenAI-affiliated research about ChatGPT's educational benefit, which is worth holding in mind while reading it.