Will Chatbots Really Stunt How Children Learn?

New York City has just told roughly 600,000 of its youngest students to put the chatbots away. The city's one-year moratorium bars student-facing generative AI from pre-kindergarten through eighth grade, prohibits companion chatbots at every grade, and follows a similar move in Los Angeles. Two of the largest school systems in the country have now decided, at least for a year, that AI does not belong in front of children who are still learning to read and reason.

The fear underneath the policy is easy to state. If a child can ask a machine for the answer, the worry goes, they never build the muscle that comes from struggling toward it themselves. Writing, arithmetic and argument are skills you earn by doing them badly first. Hand a ten-year-old a tool that does them well on demand, and you may be handing away the struggle that does the teaching.

What the evidence actually shows

It is a serious concern, and it deserves a serious look rather than a reflex. The honest position is that the science here is early and mixed, not settled. There is real research on cognitive offloading, the habit of leaning on an external tool instead of your own memory, and some of it is genuinely worth attention. A widely discussed MIT study on so-called cognitive debt, which we examined earlier, found weaker brain engagement in adults who wrote essays with a chatbot's help. But that was a small study, in adults, over a short window, and it measured effort during a task, not lasting harm to how people think.

What we do not have is strong long-term evidence that using AI stunts how children learn. We have reasonable theory, some suggestive short-term findings, and a great deal of understandable adult anxiety. Those are not the same as proof.

The distinction that matters

The useful question is not whether children touch AI, but how. A chatbot that spits out a finished essay replaces the thinking. A chatbot that asks a student guiding questions, or explains a step they are stuck on, can support it. The same tool can substitute for effort or scaffold it, and which one happens depends almost entirely on how a lesson is built around it. That nuance is hard to capture in a district-wide rule, which is part of why blanket bans are appealing. They are simple, and they buy time.

So the honest verdict sits between the headlines. The claim that AI will inevitably rot young minds runs well ahead of the evidence. So does the sunnier claim that dropping chatbots into classrooms will simply make kids smarter. New York's own policy hedges in exactly this way. It bans the youngest students from using AI while still letting teachers use it for planning, and it keeps a small, closely watched pilot running in some high schools.

Read charitably, the moratorium is not a verdict that AI harms children. It is an admission that we do not yet know, and a decision to wait rather than experiment on eight-year-olds while we find out. That is a defensible instinct. It is just worth being clear that caution is what this is, not settled science. For the districts already pulling AI from classrooms, the harder work starts when the year is up and the question of how, not whether, can no longer be deferred.