In a new study, researchers claim to have provided the first causal evidence that relying on AI to assist with “reasoning-intensive” cognitive labor (mental tasks ranging from writing to studying to coding to simply brainstorming new ideas) can rapidly undermine a user’s intellectual capacity and willingness to persevere despite difficulties.
“We find that AI assistance improves immediate performance, but at a significant cognitive cost,” the study declares its findings. “After this [about] After 10 minutes of AI-assisted problem solving, those who lost access to AI performed worse and gave up more often than those who never used AI. ”
The study was conducted by a multidisciplinary group of scientists from across the US and UK and has not yet been peer-reviewed. But the study builds on a growing body of research suggesting that widespread use of AI can distort and weaken users’ thinking and independence, and warns that outsourcing cognitive tasks to AI tools could put humans in a “boiled frog” quandary, as experts work to understand the impact on people of widely used chatbots that are deployed in real time. In other words, our cognitive “muscles” are gradually eroded without us realizing it, leading to terrifying challenges in the long run.
“If continued use of AI undermines the motivation and persistence that drive long-term learning, the effects will accumulate over years and become visible, and will be difficult to reverse,” the study claims. “It’s similar to the ‘boiled frog’ effect: each incremental action feels costless until the cumulative effects become too much to handle.”
To conduct the study, the researchers recruited a cohort of approximately 350 Americans who were asked to complete a series of short fractional equations. Just over half of the participants were randomly given access to a chatbot (a specialized bot built on OpenAI’s GPT-5 that provides specific answers to each question in a simple exam) for help. All others were recruited into a control group without AI.
As the results revealed, firstly, the chatbot proved useful in helping AI-assisted participants pass the test smoothly. However, midway through the short trial, access to the AI was suddenly cut off. At that point, the participants’ ability to tackle reasoning problems without AI assistance rapidly declined, as did their motivation to continue working on the problem when the going got tough.
For a follow-up experiment, the researchers recruited another larger group of about 670 participants. They were once again divided into roughly equal halves and asked to take a simple mathematical reasoning test, with one group given access to a chatbot assistant, but once again suddenly abandoned by their AI companions and left to fend for themselves cognitively. The results were about the same, worse performance and less patience.
These same results again persisted in a final experiment in which about 200 participants completed a series of short reading comprehension questions, showing that such results were not limited to just math questions.
“People become less persistent,” Rashit Dubey, a computational cognitive scientist and assistant professor at the University of California, Los Angeles, who co-authored the study with colleagues from the Massachusetts Institute of Technology, Carnegie Mellon University, and the University of Oxford, said in an interview. futurism. “When AI is taken away from people, people don’t just give wrong answers; they don’t even try without it.”
One bright spot from the study is that how participants used AI appears to make a difference in individual outcomes. People who self-reported that they essentially encouraged chatbots to spit out answers unsurprisingly fared even worse when the AI rug was pulled from them. Participants who said they asked the chatbot for hints or clarification, rather than outright cheating, seemed to be better off without AI assistance.
Darby worries that relying too much on chatbots to replace cognitive labor could make people more impatient and even create a situation where they become overly dependent on AI and act like an addict. But above all, he says he is concerned about how reliance on AI will change the confidence and values of individuals who struggle to think independently about problems.
“The most important thing I learned in college was the value of hard work…If you work hard, you can achieve a lot,” Dubey recalled, noting that schools and communities need to think very carefully about “blindly” incorporating chatbots into educational programs. “These are very important core human elements that we learn throughout our childhood, high school, and college years.”
“If we are offloading everything to AI at scale, what will happen to our own beliefs about ourselves?” Dubey continued, adding, “Practice makes you better in so many areas, and AI takes that away…That’s what worries me the most. We’re going to have a generation of learners and people who don’t know what they’re capable of, and that really diminishes human innovation and creativity.”
And as the researchers look to expand their work into long-term experiments, they are asking people across industries to “think about optimizing not just what you can do with AI, but what you can do without it.”
Learn more about AI and cognitive functions: University students will no longer be able to participate in class discussions as they leave their thinking to AI
