NYC bans AI in schools to protect real learning
The most dangerous thing about artificial intelligence in education is not that it fails. It is that it succeeds just often enough to make us stop looking at what it actually does. New York City’s decision to ban generative AI for 600,000 students in kindergarten through eighth grade is not a Luddite tantrum. [1] It is a rare, institutional admission that the technology’s most impressive demonstrations are often its most misleading ones. Mayor Eric Adams put it bluntly: the tech industry wants us to believe AI-powered early education is inevitable and necessary. The city, after watching the evidence pile up, has concluded the opposite.
The Mirage of Measurable Progress
The core problem is not that AI tools are useless. The problem is that they are optimized to display competence rather than to build it. When a third-grader uses a chatbot to generate a paragraph about the American Revolution, the output looks polished. It has topic sentences. It has a conclusion. It has none of the struggle, the false starts, or the cognitive friction that actually creates a young writer’s neural pathways. The child sees a finished product and believes they have learned something. The teacher sees a finished product and believes they have taught something. Both are wrong, and the technology is designed to keep them wrong.
This is where the deception operates at its most subtle level. A language model does not know what a child does not understand. It cannot see the moment when a student’s face goes blank because a concept has not landed. It can only produce text that looks like understanding. Researchers who study human cognition have known for decades that retrieval practice — forcing a student to recall information without prompts — is vastly superior to passive review. AI tools, by their very nature, eliminate retrieval. They replace the act of remembering with the act of requesting. The screen becomes a prosthetic memory, and the brain, like any unused muscle, quietly atrophies.
New York’s new policy, which also restricts individual screens to 30 minutes for third through fifth graders and 45 minutes for middle schoolers, reflects a growing awareness that the device itself is part of the problem. The policy was announced in April 2025. [1] Prekindergarten through second grade students will have no screens at all. This is not a rejection of technology. It is a rejection of a specific illusion: that more access to AI tools translates into more learning. The evidence from cognitive science suggests the opposite. When children type a question and receive an instant answer, they learn that answers are cheap. They do not learn that questions are valuable.
The Politics of Digital Faith
The tech industry’s push into classrooms has always been less about pedagogy than about market capture. Google and Microsoft have spent the better part of a decade embedding their products into school districts, often at discounted rates that make refusal politically difficult for cash-strapped administrators. Once a district adopts a platform, switching costs become prohibitive. The AI features are then layered on top, not because educators demanded them, but because the companies need to show growth to investors. The classroom becomes a beta-testing ground for products that were never designed with developmental psychology in mind.
What makes New York’s move significant is that it breaks this cycle at the level of policy, not just individual teacher discretion. Schools Chancellor David Banks told Politico that the draft guidance had “missed the mark” and that principals were delaying software purchases while the Department of Education reviewed the situation. [1] This is the language of a bureaucracy that has finally noticed it was being played. The initial AI policy, released in March 2025, divided usage into red, yellow, and green zones. Red areas — student assessment, grading, special education plans — were off-limits. Yellow areas required caution. Green areas, like brainstorming and creating accessible materials, were deemed safe. The problem was that the green zones were precisely where the most insidious deception occurs.
Consider what “brainstorming” means in an AI context. A student who asks a chatbot for ideas is not brainstorming. They are outsourcing the generation of possibilities to a statistical engine that has no stake in their curiosity. The tool will produce ten ideas, all of them plausible, none of them owned by the student. The feeling of having generated something — the dopamine hit of creative agency — is counterfeit. The student experiences the pleasure of discovery without doing the work of discovery. Over time, this teaches a devastating lesson: that thinking is something you watch happen, not something you do.
The Hidden Cost of Convenience
The deception extends beyond the classroom into the very metrics that schools use to measure success. Educational software companies love to cite engagement data. Students spend more time on their devices. They complete more exercises. They click more buttons. These metrics are easy to collect and impressive in presentations to school boards. But engagement is not learning. A child can be deeply engaged in a video game that teaches nothing. The industry has conflated attention with absorption, activity with acquisition. This is not a bug in their reporting. It is the feature that keeps the contracts flowing.
New York’s ban is also a rebuke to the idea that AI can somehow close achievement gaps. The argument, repeated endlessly in industry white papers, is that AI tutors can provide personalized instruction to students who lack access to expensive human tutors. On its face, this sounds compassionate. In practice, it often means giving the most disadvantaged students the lowest-quality educational interaction available. A human tutor can read a child’s face, adjust pacing, offer encouragement, and build trust. A chatbot can only generate text. The students who need the most human connection are being offered the most automated version of it. This is not equity. It is a digital version of separate but equal, and it is no more defensible now than it was in the analog era.
The exceptions in New York’s policy are telling. Assistive technology for students with disabilities remains allowed. English language learners retain access to AI tools. These are cases where the technology genuinely augments human capability rather than replacing it. A speech-to-text tool for a student with motor impairments is not deceptive. It does what it claims to do. The same cannot be said for a chatbot that purports to teach a child to write. The distinction is not between good and bad technology. It is between tools that are honest about their limitations and tools that are engineered to hide them.
The Consequence of Clarity
What happens when a major city tells the tech industry that its flagship product is not suitable for children? The immediate effect is financial. New York City public schools serve roughly 1.1 million students. That is a massive procurement budget that will now be spent elsewhere. The longer-term effect is psychological. Other districts watch what New York does. When the largest school system in the country says that AI is not inevitable, that it is not necessary, that it requires study rather than adoption, it gives permission for other administrators to slow down. The tech industry’s greatest advantage has always been the narrative of inevitability. Every district wants to be forward-thinking. No administrator wants to be the one who held back progress. New York just made it respectable to ask questions.
The city will spend the 2025-2026 school year studying the impacts of the technology. This is the right pace. The industry wants speed because speed favors the status quo. A year of study, conducted by people who are not selling anything, has the potential to produce something the market has never delivered: an honest assessment of what AI actually does in a classroom. The draft policy that was criticized in March was delayed because it did not go far enough. The new policy, with its bans and screen-time limits, goes further. It recognizes that the burden of proof should be on the technology, not on the children.
The deepest deception of AI in education is that it presents itself as a neutral tool when it is actually a reshaping force. Every hour a child spends interacting with a chatbot is an hour they are not spending interacting with a teacher, a peer, or their own thoughts. The technology does not just add something to the classroom. It subtracts something. It subtracts the space for confusion, for boredom, for the kind of struggle that precedes genuine insight. Those uncomfortable moments are not bugs in the educational system. They are the system. The polished output of a language model is the enemy of the messy process that builds a mind.
New York’s decision will be called many things. It will be called reactionary by the industry. It will be called protectionist by those who believe the future belongs to AI natives. It will be called paternalistic by parents who want their children to have every advantage. What it will not be called, by anyone who reads the policy carefully, is uninformed. The city looked at the claims, looked at the evidence, and found a gap wide enough to drive a school bus through. For 600,000 students, the screens will go dark. In their place will be something far more radical: teachers, paper, and the slow, unglamorous work of learning how to think. That work does not generate engagement metrics. It does not produce polished paragraphs. It produces something the tech industry has never managed to package: adults who cannot be fooled by a confident-sounding machine.
Sources
1. New York City Department of Education
2. Google