Berkeley study: 10 minutes of AI use can weaken persistence

Summary

A UC Berkeley-led team ran randomized controlled trials with 1,222 online participants. Groups allowed to use ChatGPT while solving fraction problems became less accurate and skipped more questions once the tool was taken away, and the same pattern appeared in a larger 667-person experiment and a 201-person SAT reading test. The peer-reviewed paper was presented this week at the Conference on Language Modeling, after an early draft drew wide attention in the spring.

Why it matters for our work

The findings suggest that the instant speed of AI can crowd out the productive struggle that builds the ability to stick with hard problems. For teams mixing human and AI work, deciding when to lean on AI and when to persist unaided becomes a real design question for research and learning.

Translated from the Korean original. Summaries may be translated and edited. Commentary reflects our perspective; forecasts remain the source’s views.