Research from multiple universities on 1,222 people showed that just 10 minutes of using AI to get direct answers significantly reduced the brain's ability to solve problems independently.
A new study from Carnegie Mellon University, MIT, Oxford, and UCLA has just been published, revealing worrying results. The analysis indicates that 10 minutes of using AI to directly obtain answers significantly reduced independent thinking ability and the willpower to persevere without AI assistance. The study conducted three randomized controlled experiments with 1,222 participants.
In the first experiment, 354 people were divided into two groups to solve fraction problems. The first group used the GPT-5 model for the first 12 questions, then had their access revoked for the last 3 questions. The control group solved all 15 questions independently. The results for the last 3 questions showed that the AI group only achieved an accuracy rate of 0.57, compared to 0.73 for the control group. The AI group's question skipping rate was 0.20, almost twice as high as the control group's.
The second experiment involved removing confounding variables to test the initial capabilities of all participants. Among AI users, 61% asked for direct answers, 27% only asked for hints or guidance, and 12% hardly used any AI at all. The group that asked for direct answers had the lowest independent accuracy rate and the highest skip rate. The group that only used AI for hints had results that were not significantly different from the group that did not use AI.
"If this dependence persists for months or years, we could create an entire generation of learners who lose the ability to grapple with problems on their own. Without technological support, they won't know how to think effectively," the research team wrote in the report.
The third experiment shifted to reading comprehension with 201 SAT test-takers and yielded similar results. The AI-powered group scored only 0.76 when doing the test themselves, compared to 0.89 for the control group.
The research team explained the phenomenon through two mechanisms. First, there is a shift in expectations. Once the brain becomes accustomed to receiving answers within seconds, tasks requiring more than three minutes of concentration feel overwhelming. Second, there is a loss of self-assessment, as the process of grappling with a problem and finding an answer helps learners understand their own limitations and capabilities.
Finally, MIT also published its own research showing that people who used AI to write long essays often couldn't remember the content and didn't even recognize their own work in subsequent tests. The research team named this phenomenon "cognitive debt," where users consume content without critical thinking, causing the brain to lose its ability to remember