We’ve always leaned on tools to lighten our mental load, from calculators to GPS to search engines, and it’s generally made us more efficient. But AI chatbots like ChatGPT, Claude, and Gemini take cognitive offloading to a new level, handling everything from essay writing to medical image analysis at the touch of a prompt.
This has sparked a pressing question: is AI actually dulling our minds? According to a new paper from an international team of psychologists, the answer is both yes and no—AI can boost learning through immediate feedback, but people who lean on it too heavily often perform worse once the tool is taken away.
The concern centers on what researchers are calling “cognitive debt.” Unlike simple memory aids like shopping lists, AI can offload critical thinking itself. An MIT study found that people who wrote essays using ChatGPT showed weaker brain connectivity, struggled to recall their own writing, and produced weaker essays when later asked to write without AI assistance.
Similar patterns emerged elsewhere: high school students who used AI to solve math problems performed worse on unaided tests, and doctors who relied on AI to spot abnormalities during colonoscopies saw their own detection rates drop once the tool was removed. Even how people build general knowledge is affected: participants who used ChatGPT to learn about a topic gave notably shallower advice than those who researched and synthesized information themselves.
Despite these worrying signs, the picture isn’t entirely bleak. The same MIT study found that people who first wrote independently and only later brought in ChatGPT produced more creative, better-argued work while still preserving their own voice and perspective. Similarly, students who used AI as a tutor, asking for hints rather than direct answers, retained strong performance even after losing access to the tool. The key difference seems to lie in how AI is used: as a total substitute for thinking, it erodes skills, but as a collaborator that challenges ideas and fills gaps, it can genuinely enhance learning and performance.
Encouragingly, researchers found that core cognitive abilities, like attention, reasoning, and working memory, appear to be far more resistant to erosion than specific skills. Since becoming skilled at a task usually reflects learning to use these mental resources more efficiently rather than expanding raw brainpower, AI may chip away at particular abilities without damaging this deeper cognitive architecture. Still, whether this resilience holds over decades of habitual AI use, or during childhood when the brain is still developing, remains an open question.
Ultimately, the article frames AI’s effect on cognition much like financial debt: a tempting shortcut that offers immediate convenience but can carry long-term costs if overused. Many questions remain unanswered, including whether people will eventually adapt to AI the way they did to calculators and smartphones, whether refresher training can prevent skill decay, and how to encourage people to offload strategically rather than completely. As the piece suggests, the more philosophical question may be the most important one: as we increasingly share our thinking with machines, will our very definition of thinking begin to change?