People who rely on artificial intelligence (AI) chatbots to solve everyday problems may become less capable of completing similar tasks independently when the technology is taken away, according to new research examining the effects of AI-assisted learning.
The study, presented at the 2026 Conference on Language Modeling in San Francisco, found that participants who used AI to obtain quick answers performed worse and were more likely to abandon tasks when subsequently required to work without assistance.
Researchers led by Liu conducted three randomised controlled experiments involving 1,222 participants to examine how AI assistance affected mathematical problem-solving, reading comprehension and the willingness to persist when faced with difficult tasks.
The findings raise concerns about the growing practice of cognitive offloading, in which individuals transfer mental work, reasoning and problem-solving to external tools rather than developing their own abilities.
In the first experiment, 354 participants were divided into two groups and given 12 mathematical fraction problems. One group had access to an AI chatbot, while the other had to solve the problems without assistance.
The AI-assisted participants were initially more successful and less likely to abandon questions. However, when researchers unexpectedly removed access to the chatbot and presented three additional problems, their performance declined. They were also more likely to skip questions or fail to complete them.
A second experiment involving 667 participants produced similar findings, suggesting that reliance on AI for direct answers could undermine perseverance and motivation when individuals are required to solve problems independently.
The third experiment examined whether the effects extended beyond mathematics. Researchers recruited 201 participants and tested their reading comprehension using five initial questions, followed by three additional questions after chatbot access was withdrawn.
The results were similar: participants who had relied on AI were less successful at completing the subsequent questions independently and were more likely to skip them. This suggested that the effects were not limited to mathematical calculations but could also affect reading and comprehension.
“The most significant finding is that people who offload more have higher declines in persistence (increased skip rate), which is an important skill that cuts across domains,” Liu said.
Researchers suggested that repeated exposure to immediate answers may condition users to view independent problem-solving as unnecessarily difficult or inefficient. However, they cautioned that further long-term research is needed to establish whether brief periods of AI reliance produce lasting declines in intellectual ability.
Researchers call for AI to support, rather than replace, human thinking
The study’s authors argue that AI tools should be designed to encourage learning and independent reasoning rather than simply provide completed answers.
They recommend that chatbots function more like tutors by offering explanations, guidance and clues that help users work towards solutions themselves.
Brian Christian, a computer scientist and researcher at the University of California, Berkeley’s Center for Human-Compatible AI and a co-author of the study, said:
“We need to work toward a future in which human abilities are, as much as possible, augmented rather than supplanted.”
The findings add to growing scientific discussion about the consequences of cognitive offloading and the potential impact of excessive dependence on AI on education, critical thinking and independent decision-making.
While AI can improve efficiency and provide valuable assistance, the researchers warn that convenience should not come at the expense of the mental effort required to develop knowledge, skills and intellectual independence.
