A new study from multiple esteemed universities has found that using AI for even brief periods of time can fry your brain.
Conducted by researchers at Carnegie Mellon, MIT, Oxford and UCLA, the study observed participants’ critical thinking abilities during and after the use of a chatbot powered by OpenAI’s GPT-5 model. In these roughly 10-minute sessions, people were tasked with solving problems related to fraction-based arithmetic and reading comprehension.
When they used the AI for the math questions, their solve rate was markedly higher, but once it was taken away, it dropped by around 20 per cent. What’s more, people were skipping questions entirely nearly twice as many times once AI was taken away.
And while the study didn’t find a significant increase in the assisted solve rate for the reading comprehension questions, there was still a drop in correct answers and an increase in skipping once AI was taken away.
Researchers took three key findings away from the study. First and foremost, they noted that “AI assistance reduces persistence and impairs independent performance,” with participants being “significantly more likely to give up on problems and performed significantly worse once the AI was removed, compared to participants who never had AI assistance.”
Additionally, they noted that these negative impacts were more severe in users who wanted AI to “directly” solve tasks for them, while there were no “significant impairments” when AI was used simply to provide “hints or clarifications.”
And finally, they noted that these effects were consistent across the fraction arithmetic and reading comprehension questions, indicating that the findings are “a general consequence of AI-assisted problem solving, not specific to any particular task.”
Ultimately, the researchers ruled that “AI assistance improves immediate performance, but it comes at a heavy cognitive cost” to users. “These findings raise urgent questions about the cumulative effects of daily AI use on human persistence and reasoning. We caution that if such effects accumulate with sustained AI use, current AI systems — optimized only for short-term helpfulness — risk eroding the very human capabilities they are meant to support,” they concluded.
All in all, it’s valuable data, illustrating that, like most things in life, moderation is key. The study’s findings corroborate similar research that has been conducted in recent years, like a 2025 MIT study into how ChatGPT can erode critical thinking skills or a 2025 Microsoft and Carnegie study showing AI-driven cognitive decline. (And that’s to say nothing of the many instances where AI simply gets things wrong, such as the numerous examples of “hallucinations” in Google Search in which AI Overviews provide flat-out false information.)
Image credit: OpenAI
