10 minutes of AI use lowers problem-solving ability by 20%, study finds

Just 10 minutes of AI use cut independent math scores by 20%, per a study from Carnegie Mellon, Oxford, MIT, and UCLA. Participants who asked for direct answers saw the worst drops.

Categorized in: AI News Science and Research
Published on: May 12, 2026
10 minutes of AI use lowers problem-solving ability by 20%, study finds

Study Shows 10 Minutes of AI Use Impairs Problem-Solving Skills

Researchers at Carnegie Mellon University, Oxford, MIT, and UCLA found that brief reliance on AI assistants degrades the ability to solve problems independently. Participants who used an AI tool for just 10 minutes performed roughly 20% worse on math problems once the AI was removed, compared to those who worked without assistance.

The study split participants into two groups. One solved fraction-based math problems alone. The other used an AI assistant powered by OpenAI's GPT-5 model for most of the test, then had it removed without warning for the final three questions.

While the AI-assisted group solved more problems initially, their performance collapsed once they lost access. The control group, working independently throughout, maintained consistent performance.

Dependency Builds Quickly

The AI-assisted group also skipped questions at twice the rate of the control group after losing their tool. Rather than attempt problems, participants abandoned them.

A follow-up experiment testing reading comprehension showed similar results, though the AI provided no initial advantage for that task.

The finding suggests that cognitive reliance on AI develops faster than previously understood. Ten minutes was enough to measurably degrade independent problem-solving capacity.

How You Use AI Matters

Not all AI use affected participants equally. Those who asked the AI for direct answers saw the largest performance drops. Sixty-one percent of participants reported requesting direct solutions rather than using the tool to guide their thinking.

Participants who used AI differently-presumably asking for hints or explanations rather than answers-showed smaller declines in performance after losing access.

The research suggests the mechanism matters. Outsourcing thinking entirely produces different cognitive outcomes than using AI as a reference point.

For professionals working with research data or complex problems, the implications are direct: how you interact with AI tools shapes whether those tools support or replace your own analytical capabilities.

Learn more about Generative AI and LLM Courses and AI Research Courses to understand how to work effectively with these tools.


Get Daily AI News

Your membership also unlocks:

700+ AI Courses
700+ Certifications
Personalized AI Learning Plan
6500+ AI Tools (no Ads)
Daily AI News by job industry (no Ads)