WORLD - Researchers found that people who relied on artificial intelligence tools performed less accurately and were less persistent when the tools were removed, raising concerns about the effects of AI dependence on human learning and problem-solving.
The study, involving 1,222 participants, found that access to AI assistance was associated with a decline in performance once the technology became unavailable. The researchers conducted three randomized controlled experiments examining participants’ ability to solve mathematics problems and answer reading comprehension questions.
The findings were presented at the Conference on Language Modeling in October 2026.
AI Assistance and Problem-Solving
In the first experiment, 354 participants were divided into two groups. One group solved 15 basic fraction problems without assistance, while the other could use ChatGPT to help answer the questions.
Participants with AI access initially achieved greater accuracy. However, when researchers removed the tool after 12 questions, their accuracy dropped sharply.
A second experiment involving 667 participants produced similar results. Participants who had used AI were more likely to answer incorrectly or abandon the task after losing access to the tool, while those who had worked without AI were more successful at completing the problems.
A third experiment, involving 201 participants, examined reading comprehension using an SAT-style prompt. Researchers observed a similar decline in accuracy and persistence after AI assistance was withdrawn.
Brian Christian, a research fellow at the University of California, Berkeley’s Center for Human-Compatible AI, was part of the research team.
The study builds on growing concerns that excessive reliance on AI tools could weaken people’s ability to concentrate, persist through difficult tasks and develop independent problem-solving skills.
The Importance of Independent Thinking
Researchers highlighted the role of what they described as “productive struggle,” the process of working through challenges independently, which can help people develop expertise and discover their strengths.
AI tools can provide rapid answers, potentially encouraging users to avoid tasks that require sustained effort. The researchers questioned whether systems designed to help people might sometimes reduce learning opportunities.
Christian said AI should be developed to strengthen human abilities rather than replace them, including by providing more instructional responses that guide users through problems instead of simply supplying answers.
Implications for Research and Education
The findings also raise questions about the use of AI in academic research, where developing expertise often requires sustained engagement with data, close examination of evidence and collaboration among researchers.
Christian warned that relying heavily on AI could affect the skills and working practices that contribute to scientific discoveries.
The research team included scholars from Carnegie Mellon University, the Massachusetts Institute of Technology, the University of Oxford and the University of California, Los Angeles.
The researchers’ findings highlight the need to consider not only how AI improves efficiency, but also how its use may affect learning, independent judgment and the development of human expertise.
