Short AI use may impair problem-solving skills, study finds

Critics of artificial intelligence have long warned about its long-term effects on the human brain, given its relatively recent emergence as a mainstream tool. Now, new research suggests that even short-term AI use may pose cognitive risks.

A study conducted by researchers at Carnegie Mellon, Oxford, MIT, and UCLA examined how AI assistance impacts problem-solving abilities. Participants were given a set of fraction-based math problems to solve. Half completed the test independently, while the other half used an AI assistant powered by OpenAI’s GPT-5 model—only to have the AI removed without warning for the final three problems.

While the AI-assisted group initially performed better, their solve rate dropped sharply once the tool was taken away. Once both groups worked independently, the AI-assisted group’s solve rate was approximately 20% lower than the control group. Additionally, they skipped questions at twice the rate of the control group, suggesting a reliance on AI even after brief exposure.

The participants had access to their AI assistants for only about 10 minutes, indicating that even short-term AI reliance can weaken independent problem-solving skills.

The researchers replicated the experiment using reading comprehension tasks instead of math. The results mirrored the math study, with one key difference: the AI-assisted group showed no initial advantage in the first portion of the exam.

How AI use affects cognitive performance

Not all AI usage harmed problem-solving abilities equally. Participants who asked the AI for direct solutions experienced the steepest decline in solve rate and the highest increase in skip rate. A majority of participants—61%—reported seeking direct answers from the AI.

However, those who used AI for hints or clarifications maintained performance levels comparable to the control group. This suggests that total reliance on AI, rather than occasional assistance, is the primary driver of cognitive impairment.

Previous research supports these findings

The study’s conclusions align with prior research linking AI use to cognitive decline. A MIT study measured brain activity during essay writing and found that independent writers exhibited significantly higher brain connectivity than those using large language models (LLMs). Over a four-month experiment, LLM-assisted writers underperformed in neural, linguistic, and behavioral metrics.

Additional studies in fields like knowledge work and medicine have shown that workers who rely on AI to complete tasks struggle to perform those same tasks without AI assistance.