Scientists at Carnegie Mellon, MIT, Oxford, and the University of California at Los Angeles have concluded that ten minutes of communication with an AI helper is enough to weaken a person’s ability to think for themselves. The experiment confirmed that the "double-check the neurose" strategy does not save - the brain quickly gets used to delegating intellectual work to a machine.

The study involved two groups of volunteers, one performing GPT-5-based AI-powered jobs and the other without assistance. At the end of the experiment, the first group suddenly disconnected their assistant for the last three tasks. Total: Their accuracy of answers decreased by about 20% compared to those who worked alone all the time. And the rate of missed questions has doubled - people simply stopped trying to find a solution.

A similar pattern has been observed in both mathematics and text comprehension tests. Worst of all, when AI was disabled, those who were used to getting ready answers from it showed themselves. Those who used hints (rather than a complete solution) suffered significantly less. The authors of the study summarize: cognitive function decline is a universal consequence of AI work that does not depend on the nature of tasks.