UC Berkeley Study Finds 10 Minutes of AI Use Can Undercut Persistence

A brief window with AI, then a harder task

A UC Berkeley research fellow, Brian Christian, co-authored a study that found spending just 10 minutes relying on an AI tool significantly impairs the ability to focus and persist with difficult tasks. The updated, peer-reviewed paper was presented at the Conference on Language Modeling, after a preliminary draft drew attention earlier in the year.

Christian has spent 20 years thinking about how artificial intelligence will change minds and society, and his 2020 book The Alignment Problem was praised as one of the best on AI. He also keeps a notebook on his desk and writes in it every day, a habit that grounds him amid concern that too much work with AI could erode his own scholarly edge.

Fraction and reading tests showed the same pattern

Across randomized controlled trials, researchers recruited 1,222 participants online. In the first experiment, 354 people solved 15 basic fraction problems either without help or with ChatGPT open on the side for assistance and even to ask for answers.

The group with AI access started off more accurate than the group without it. After 12 problems, researchers took away the AI tool, and almost immediately that group stopped solving questions accurately.

A second, larger group of 667 participants showed a similar result. When AI assistance was withdrawn, the AI users got answers wrong or gave up entirely, while the group without AI stuck with the task and was more successful at the end.

Researchers then ran a third experiment with 201 participants using an SAT reading comprehension prompt. The results echoed the fraction findings: after the tool was removed, persistence and accuracy dropped, and participants skipped more questions.

The cost of losing productive struggle

The study team included scholars from Carnegie Mellon University, the Massachusetts Institute of Technology, the University of Oxford and UCLA. The findings align with anecdotes and surveys that overreliance on AI for answers can erode cognitive fitness.

Christian said the systems were built to be helpful, but the results show they can help in ways that end up unhelpful. When AI produces answers in seconds, people can get used to that speed, and anything that takes longer feels inefficient. The team also pointed to a loss of productive struggle, the kind of effort that can help people find research strengths and passions.

Augmenting people instead of replacing them

Frontier AI companies have acknowledged that offloading tasks to chatbots can reduce baseline understanding and mastery of a subject. Christian suggested that instead of defaulting to quick answers, AI responses could default to being more instructive, like a tutor.

He argued for a future in which human abilities are augmented rather than supplanted, especially in research and academia. He worries that overreliance on AI in science could weaken what makes a good scientist: staying close to data, being immersed in the material and working as part of a team.

For Christian, the concern goes beyond fractions and SAT problems. A research team of 10 developing ideas that lead to a breakthrough could become two or three researchers and a chatbot. He said the same shift is happening to human knowledge and expertise from grade school students to leading experts.

Sources: vcresearch.berkeley.edu