A new study found that AI assistance improved performance in the moment but reduced problem-solving ability once the tool was removed.
Researchers observed the effect after just 10 to 15 minutes of AI-assisted work on math and reading tasks.
The findings suggest that how people use AI may matter, with hints and clarification appearing less problematic than asking for direct answers.
Artificial intelligence has quickly become part of everyday life. People use AI tools to write emails, answer questions, brainstorm ideas, and even help with school or work assignments. While these tools can save time and make difficult tasks feel easier, new research suggests there may be a downside to relying on them too heavily.
Researchers from several universities, including Carnegie Mellon University, the University of Oxford, MIT, and UCLA, wanted to understand what happens when people use AI for reasoning-intensive tasks and then have to continue without it.
Their findings suggest that AI can boost performance in the short term, but that benefit may come at a cost when users are later asked to work independently.
The study
The study focused on two abilities that are important for learning and problem-solving: persistence, or the willingness to keep working through a challenge, and independent performance, meaning how well someone can complete a task without assistance.
The research involved more than 1,200 participants across a series of randomized experiments.
In one experiment, participants solved fraction-based math problems. Some completed the work on their own, while others were given access to an AI assistant for about 10 minutes before the assistance was unexpectedly removed.
The results
Not surprisingly, the AI-assisted group performed better while the tool was available. However, once the AI was taken away, their performance dropped noticeably.
Researchers found that participants who had relied on AI solved fewer problems correctly and were more likely to skip questions altogether than those who never had access to AI in the first place. Similar patterns appeared in reading-comprehension tasks.
The researchers also discovered an important distinction. Participants who used AI mainly to obtain direct answers experienced the largest declines. Those who used the tool more as a tutor — asking for hints, explanations, or clarification — did not show the same level of impairment.
What this means for consumers
The findings do not suggest that AI is inherently harmful or that people should avoid using it. Instead, the study highlights the importance of how these tools are used. According to the researchers, AI may be most beneficial when it supports learning rather than replaces the thinking process altogether.
For consumers, the takeaway may be simple: use AI as a guide rather than a substitute. Asking for explanations, examples, or feedback could help reinforce understanding, while relying on AI to provide complete solutions may make it harder to develop and maintain problem-solving skills over time.
The researchers note that their experiments measured short-term effects, not years of AI use. Still, because the changes appeared after only a brief interaction with AI, they say more research is needed to understand what frequent, long-term reliance on these tools could mean for learning, persistence, and independent thinking.