Using generative AI for a few minutes to solve a complex math problem or draft an email may be a quick and easy way to get the job done. But your critical thinking, creativity and reasoning skills sit unused if AI is doing all the hard work for you. That might have more of an effect than you think.
A recent preprint study by researchers from Carnegie Mellon University, the University of Oxford, the Massachusetts Institute of Technology and the University of California, Los Angeles, found that participants who relied on AI to practice doing a task for just 10 minutes struggled to do that task without it, compared to those who never used AI for it in the first place.
The researchers said AI chatbots can help with “reasoning-intensive tasks” like studying and brainstorming, but that relying on bots for these tasks might hinder our ability to do those tasks on our own. There’s a big difference between having an AI tool solve problems for you and using it for hints and clarifications — or not at all.
The study raises ongoing concerns about how using AI affects our ability to do the same tasks without the machine’s help. It’s one thing to have a tool like ChatGPT or Claude provide some guidance or answer some questions that point us toward a solution, but it’s another to have it do our tasks wholesale.
Using AI to complete a school assignment before a deadline is not the same as letting it help you manage your time. AI’s overuse, especially in workplaces and schools, could be making us, as humans, worse at doing things we really should be able to do on our own.
Solving problems, with and without AI
The study consisted of 1,200 US participants on the research platform Prolific and three experiments. The tests tasked participants with solving fraction problems and answering SAT-style reading comprehension questions. Some people were given the AI assistant for help, and were able to use it however they liked, but it was taken away after 10 minutes. The study found that participants who had used the AI for help were more likely to give up on problems and perform worse than participants who didn’t have it.
Negative effects of AI use were observed only among those who relied on it to solve problems, not among those who didn’t use it during the study. There were similar findings in an MIT study last year that focused on using AI to write essays.
Grace Liu, a Ph.D. student at Carnegie Mellon University and one of the study’s authors, told CNET in an email that only short-term effects were measured, and we don’t yet have a full picture of long-term effects. “How significant this effect is at scale and in the long term needs more research,” Liu said. Even though AI can be used differently across use cases, the study doesn’t evaluate how AI tools are used.
The researchers also found that when participants who had used AI couldn’t use it, they didn’t feel as confident in their ability to solve the problems.
“People do not merely become worse at tasks, but they also stop trying,” the researchers wrote. “If such effects accumulate over months and years of AI use, we may end up creating a generation of learners who have lost the disposition to struggle productively without technological support.”
Is AI all that different from a calculator?
The study makes me question whether using AI is similar to other problem-solving methods or shortcuts, like using a calculator to solve a math problem. One distinction is that generative AI can be used for almost anything, such as personal decisions, editing and research with follow-up questions.
“Both phenomena certainly share similarities because they allow people to outsource cognitive tasks,” Liu said. “We believe that it is especially important to study cognitive outsourcing to AI because AI can be used ubiquitously across many reasoning tasks, whereas previous tools are task-specific.”
Should you let AI do your work for you?
The researchers said the findings raise questions about the effects of our persistence and reasoning when using AI daily.
“We caution that if such effects accumulate with sustained AI use, current AI systems — optimized only for short-term helpfulness — risk eroding the very human capabilities they are meant to support,” the researchers added.
Liu recommends exercising caution when using AI tools.
“Our results suggest that we should be more intentional about how and when AI assistance is used and deployed — particularly in learning contexts,” she said. “It’s not a reason to avoid AI, but it is a reason to design and use these tools carefully.”
