“Universities must measure learning, both with and without AI assistance”

Key points

• Trattner argues for measuring learning with AI-assisted and independent work.
• A study shows chatbot access improves practice scores but lowers later exam performance.
• Assessment that ends during assisted sessions may mislabel AI as a success.
• Reported in a correspondence in Nature by Christoph Trattner and Marc Schmitt.
• The example references a randomized trial in secondary-school mathematics and prior studies.