It began with a certificate. There was a note in a third-grader’s backpack that said he was good at “demonstrating an understanding of the basic concepts of Artificial Intelligence.” The certificate was given after a drag-and-drop computer game with a cartoon dancer and songs that were easy to find on Amazon Music. The game was made in collaboration with Amazon Future Engineer. The parents weren’t asked. No one really talked about what the kids had learned. Almost no one even knew it happened.
To put it another way, A.I. got into a lot of American classrooms without a vote, a policy debate, or any real discussion about whether it should be there at all. Things often come into schools in a quiet way, wrapped up as progress and speaking the language of chance.
A growing number of cognitive scientists and some parent groups are starting to think that something went wrong but no one said it out loud. When Chromebooks were released during the COVID-19 pandemic, they gave Google a huge institutional foothold in K–12 education. By late 2020, year-over-year sales of Chromebooks had risen nearly 290 percent in a single quarter. Google’s AI suite, Gemini, was already on those devices. You can now tell kids who are sitting down to write an essay, “Help me write.” Are they about to start a presentation? “Help me visualize.” There are still prompts. They wait.
In the meantime, studies are piling up in ways that make them hard to ignore. A study from 2025 at M.I.T. warned that putting big language models into learning environments “may unintentionally contribute to cognitive atrophy.”” Scientists from M.I.T., Carnegie Mellon, U.C.L.A., and Oxford all worked together on a study that found that students who used AI to help them with math problems and then lost access to those tools did much worse and were more likely to give up. That’s not a small discovery. The researchers said that persistence is one of the best ways to tell if someone will learn something for a long time. Taking away the need to be persistent is not a good way to teach it.
It’s still not clear how many school leaders have read any of this research before signing the next ed-tech contract for their district. The Brookings Institution looked at about 400 studies on AI and education for kids and came to the conclusion that these tools “undermine children’s foundational development.” A study by Education Week that looked at thirteen hundred school districts found that about one in five interactions between students and generative AI led to cheating, self-harm, bullying, or other bad behavior. These are not strange findings from academics who are not sure about them. Those are the kinds of numbers that would make you reevaluate right away in any other situation involving children.
Of course, the businesses say something different. An adaptive learning platform says that it can personalize learning in real time, and Google talks about putting teachers “at the center of the experience.” Some private school chains say that A.I. can finish schoolwork in just two hours a day. It sounds great, until you talk to a sixth-grader who knows how to ignore an AI chatbot that keeps offering to read her poems and has her Google password. There’s something stuck in my mind about that picture.
That the technology isn’t going away is what makes this situation really tough. Chromebooks won’t be taken away from Gemini because of a school board resolution. There is no federal guidance that takes away the permission that districts have already given. The New York City Department of Education asked for feedback on its AI guidelines, saying, “The question is not whether AI belongs in schools.” It’s a strange move to start a debate by saying that the main point is no longer valid. But that’s where we are.
But the real question is whether the tech companies, school administrators, and policy teams that are making these choices are really aware of what’s at stake. It is not a side effect of schooling to learn how to struggle through a tough problem, deal with confusion, and fail and try again. Their job is to teach. It’s not innovative to give that job to a chatbot. It’s about switching. Also, the price might not be clear until after it has been paid.
