In almost all tech debates, there is a point where the early adopter’s optimism meets something harder, like data, consequences, or real-life experience. It looks like that time has come for AI in schools.
People in the EdTech industry have been secretly dreading a new study from the Brookings Institution’s Center for Universal Education. It took a serious, methodical look at AI in K–12 classrooms, without the happy filter of a product launch. It’s hard to argue with the results, which come from focus groups, interviews, and reading hundreds of research papers with students, parents, teachers, and tech experts from 50 different countries. The main point is that using generative AI in schools right now is more dangerous than helpful.
The tech industry’s responses have ranged from a careful response to something that sounds like defensiveness dressed up as nuance. To which I can relate. A story that this report makes a lot harder to understand is linked to billions of dollars being invested in EdTech.
The Brookings study is harder to ignore than a normal opinion piece from a critic because it covers a lot of ground and has specific points to make. The researchers don’t say that AI shouldn’t be used in schools. They agree that it can help students learn a second language, make teachers’ jobs easier by giving them a few extra hours a week, and even reach students in places like Afghanistan where girls haven’t been allowed to go to school. Indeed, these are good things. The report doesn’t try to hide the truth.
The main worry, though, and probably the one that keeps most people in EdTech up at night, is what’s happening to kids’ minds. The report talks about a feedback loop that is so logical it almost sounds like doom: students give AI their thinking, and in the process, they lose the cognitive muscles they were supposed to be building. One student said to researchers, “It’s easy,” without much irony. “Don’t think about it.” That sentence is like a small, quiet alarm in the report.
Someone who helped write the study and is a senior fellow at Brookings, Rebecca Winthrop, puts it in a way that’s hard to disagree with. When a kid asks AI what the answer is and accepts it, they’re not learning how to think critically about an argument, sort through different points of view, or tell the difference between useful and useless information. These aren’t just school skills. They are the building blocks of adult judgment that works. For that reason, the present moment seems dangerous: the long-term effects of that choice might not become clear for another ten years.
There’s another, less obvious worry that the report brings up that policymakers probably don’t pay as much attention to but should: AI is designed to be agreeable. That’s right. It makes you feel better. And when kids talk to a system a lot about things like their homework, their feelings, or how frustrated they are with their parents, and the system basically tells them they’re right, something in their social development may quietly bend. Winthrop makes a strong point when he says that a friend who disagrees with you and even annoys you can teach you something that a chatbot can’t. It turns out that echo chambers don’t help people feel empathy.
The most sobering part of the report might be the part about equity. EdTech has long seen itself as a force that makes things more fair and equal—technology as the great equalizer. What Brookings found, though, is that free AI tools are often the least accurate. This is because they are easy for schools with few resources to use. It costs more for models that are more advanced and reliable. Winthrop says that this might be the first time in the history of EdTech that being accurate about facts costs money. That’s a pretty big irony. That’s a problem with the structure.
The report doesn’t make the case for keeping AI out of schools. It’s a call for more than that—for educators and the companies that make these tools to work together more closely, think more slowly, and make better designs. In the Netherlands, there is already a model that does this by putting teachers and tech companies in the same room to test and evaluate new tools before they are used by everyone. It seems pretty clear. When something like this doesn’t happen very often, it says something important.
The study from Brookings is called a “premortem” because it tries to figure out what could go wrong before it’s too late to fix things. This moment may tell us more about the EdTech industry than any product roadmap. How they handle it will show how serious they are about their work.
