Ten minutes of AI help can make people give up sooner
Campus
Science
In a Berkeley-led study, people who solved problems with a chatbot did worse, and quit more, once it was taken away.
What happened
- On October 9, UC Berkeley described a study by researchers from Berkeley, Carnegie Mellon, MIT, Oxford and UCLA. It was presented at the Conference on Language Modeling.
- In three randomized experiments with 1,222 adults online, some people could ask a chatbot for help with fraction problems or SAT-style reading questions. After about ten minutes, the chatbot was taken away.
- With help, the AI group did better. Without it, their accuracy fell fast and more of them gave up. The group that never had AI kept trying and finished with better results.
Why it matters for you
Learning needs productive struggle: the effort of working a problem out yourself. A chatbot that gives answers removes that effort, and the study suggests the habit forms quickly. For students: try first, then ask AI to check your work, not to do it. For teachers: practice without AI still has a place. The study was short and online, so long-run effects are not yet known.
Sources
- Brief AI assistance may impair independent problem-solving and persistence, News-Medical (from UC Berkeley), October 9, 2026.
- Diez minutos de IA bastan para bajar la persistencia (in Spanish), El Economista, October 2026.
Notes for learning. Nothing here is investment advice. All news