AI use for just 10 minutes damages problem-solving skills, new study reveals

Brief AI Use Triggers Sharp Cognitive Decline

A new study from researchers at Carnegie Mellon, MIT, Oxford, and UCLA has uncovered troubling evidence that AI use can damage human problem-solving skills in surprisingly little time. The research, which tested over 1,000 participants across three experiments, found that relying on AI assistants for as little as 10 minutes led to measurable cognitive impairment when the technology was suddenly removed1

1

2

2

.

The experiments tasked participants with solving various challenges, including fraction-based math problems and SAT-style reading comprehension questions, through an online platform that paid them for their work. Some participants received access to AI chatbots powered by OpenAI's GPT-5 model that could solve problems autonomously, while others worked independently throughout4

4

.

Performance Collapses When AI Support Disappears

The results proved stark. During the AI-assisted portion, participants using the technology showed higher solve rates and completed tasks faster with fewer mistakes. But when researchers removed the AI helper without warning for the final three questions, performance collapsed dramatically. The AI-assisted group's solve rate dropped approximately 20% lower than those who had worked independently the entire time2

2

4

4

.

In the math section specifically, people who had previously used AI scored only 57%, while those who never used the tool scored 73%. Reading comprehension showed similar patterns, with AI users scoring 76% compared to 89% for independent thinkers5

5

. Perhaps most concerning, AI users demonstrated nearly double the skip rate, simply choosing to abandon difficult questions rather than persist2

2

.

Critical Thinking and Persistence Under Threat

Michiel Bakker, an assistant professor at MIT involved with the study, emphasizes that persistence represents a crucial element of human cognitive capabilities. "It is fundamentally a cognitive question—about persistence, learning, and how people respond to difficulty," Bakker explains. A person's willingness to persist with problem-solving predicts their capacity to learn over time and acquire new skills1

1

.

The cognitive scientists found that participants who used AI to get direct answers showed the largest declines in performance. However, those who used AI for hints rather than complete solutions didn't see significant impairment compared to the control group2

2

. This distinction suggests the manner of AI assistance matters enormously.

Redesigning AI as Educational Aid, Not Answer Machine

Bakker stresses that over-relying on AI poses particular risks but argues the solution isn't abandoning the technology entirely. "The takeaway is not that we should ban AI in education or workplaces," he says. "AI can clearly help people perform better in the moment, and that can be valuable. But we should be more careful about what kind of help AI provides, and when"1

1

3

3

.

The research suggests AI tools need fundamental redesign to work like effective human teachers—scaffolding learning rather than replacing it. "Systems that give direct answers may have very different long-term effects from systems that scaffold, coach, or challenge the user," Bakker notes1

1

3

3

. When AI offered explanations instead of providing answers directly, participants showed no signs of impaired skills, demonstrating that AI as an educational aid can avoid cognitive damage5

5

.

These findings align with previous research, including a Microsoft study on knowledge workers that found increased AI dependence correlated with worse performance without support, and Polish research showing doctors performed below baseline after AI assistance was removed2

2

. As AI chatbots become embedded in daily workflows for both children and adults, understanding how brief AI use shapes thinking patterns grows increasingly urgent for maintaining foundational human problem-solving skills.