Cognitive debt: what AI could be doing to kids' thinking

You can usually tell when a child has handed the thinking to a chatbot. The homework comes back tidy, correct, and oddly weightless, as if no one had actually wrestled with it. Most parents read that as a cheating problem. It is the smaller problem.

The bigger one has an unlovely name: cognitive debt. The idea is simple. When you routinely let a tool do the mental heavy lifting, you save effort now and settle up later, in skills that never got built. For an adult with a fully formed set of habits, that might be a rounding error. For a child who is still assembling the equipment, it is a different kind of bill.

The MIT study everyone half-remembers

In 2025, a team at the MIT Media Lab ran a small, careful experiment with a memorable title: “Your Brain on ChatGPT”. Fifty-four adults wrote essays across several sessions while the researchers recorded their brain activity with EEG. One group wrote with ChatGPT, one with a search engine, and one with nothing but their own heads.

The pattern was consistent. Brain connectivity scaled down with the amount of outside help: the no-tools group showed the strongest, most distributed neural networks, the search group came in the middle, and the ChatGPT group showed the weakest coupling. The AI writers also reported the least sense of owning what they had produced, and many struggled to quote sentences from an essay they had finished minutes earlier. When some of them were later asked to write without the tool, they underperformed the group that had never leaned on it.

Now the honest part, because it matters. This was a small study of adults, not children, on a single kind of task, and it was released as a preprint rather than a finished, peer-reviewed paper. The authors themselves caution that the results are context-dependent and may not carry over to other tasks. So it is not proof that AI rots young minds. What it is, is a real signal, measured in actual brains, that outsourcing the effort changes what the brain does while you work.

A global review reached a blunter conclusion

If the MIT paper is a single data point, the Brookings Institution’s report from January 2026 is the wide-angle view. Its Center for Universal Education spent a year consulting more than 500 students, teachers, parents, education leaders and technologists across 50 countries, and reviewed over 400 studies, before landing on a sentence that surprised a lot of people: “At this point in its trajectory, the risks of utilizing generative AI in children’s education overshadow its benefits.”

At the top of their list of risks was cognitive growth itself, how children learn new skills and learn to work through problems. This is not a fringe panic. In the same work, teacher use of AI in classrooms had jumped from 34 percent to 61 percent, and it was the students who sounded the alarm: 65 percent of those surveyed worried that relying on AI would lead to cognitive decline. One of them put it more plainly than any researcher could: “It’s easy. You don’t need to use your brain.”

Why the struggle is the point

Here is the thing that gets lost when we frame all of this as cheating. Learning is not the moment you have the right answer. It is the friction on the way there: the wrong turn, the reread, the slow assembly of a thought that was not there before. That friction is not a bug in homework. It is the homework.

A chatbot is very good at removing exactly that friction. And children are reaching for it at scale. A Pew Research Center survey of American teenagers, conducted in late 2025, found that 54 percent had used chatbots to help with schoolwork. This is not a niche behaviour to get ahead of. It is already the water they swim in.

So the useful question is not whether your child uses AI. It is whether the AI did the thinking or helped them do it. Those are opposite tools wearing the same interface. One hands over a finished paragraph. The other asks, “What do you already know about this?” and makes the child do the next step.

(A disclosure, since it is our own patch: Mentus AI builds a mentoring product designed to coach rather than answer, so we have a stake in that distinction. The evidence above stands on its own regardless of what we sell.)

What to actually do about it

You do not need monitoring software or a household ban. You need one habit. When a piece of work comes back suspiciously smooth, ask the child to explain it to you without the screen. Not as a gotcha, as a genuine question: talk me through how you got here. If they can, the thinking happened somewhere, and the tool was a help. If they cannot, you have found the debt before it compounded.

The kitchen-table question worth keeping is not “did you use AI for this.” It is “can you do it without it.” A child who can answer yes has lost nothing to the machine. A child who cannot has told you exactly where the next lesson is.