AI & childhood

Tech companies want you to put AI in front of your kids. Schools are being sold AI tutors. Tablet platforms are quietly adding AI companions to apps designed for toddlers. Homework helpers are now homework writers. The pressure to fold children into the AI economy as quickly as possible is enormous, and most of it is well-meant.

We think it's wrong.

This piece is what we believe about AI and childhood, why we believe it, and where we'd push back on ourselves.

What "AI in childhood" actually looks like in 2025

A short tour. Most parents we talk to underestimate the surface area:

• AI tutors marketed to elementary schools as "personalized learning."
• Chatbot companions inside kids' apps that converse with children for hours and remember everything.
• AI image generators on the same platforms eight-year-olds use to share school photos.
• Homework helpers that started as Q&A and quietly became ghostwriters.
• Algorithmic feeds whose ranking systems decide what every child sees, all day.
• AI-generated content — videos, music, "stories" — produced for children at industrial scale on the platforms they live on.

"AI in childhood" is not a future deployment decision. It is already the default, and most of it was rolled out without anyone asking what childhood is actually for.

What we believe

01Under 15: not yet.

Children under 15 should not use generative AI tools as part of their daily lives.

This is not about whether AI is "safe." Plenty of AI products are nominally safe. It is about what gets crowded out when a tool that produces fluent text, images, and answers in three seconds becomes available to a developing mind.

Childhood is where humans build the substrate that adulthood runs on:

• The capacity to sit with not knowing.
• The friction of sounding out a word, finishing a sentence, drawing a hand badly six times before drawing it well.
• The discomfort of being bored — the precondition for almost every act of imagination.
• The slow, embodied learning of paying attention to another human face.
• The understanding, built over thousands of small frustrations, that you are the author of what you produce.

Generative AI is a frictionless answer machine. Used early enough, often enough, and casually enough, it doesn't supplement those capacities. It substitutes for them. The child who never had to struggle to write a paragraph or sit with a blank page is not getting a head start. They are getting a body of foundational capacities that nobody bothered to build.

We are not going to learn what the long-tail effects of this are by running the experiment on millions of nine-year-olds in real time. The default position should be no.

02Ages 16–18: ethics first. Use second.

By 16, most teenagers can reason about systems, incentives, and consequences. That is the right age to teach AI — and the wrong age to hand it to them without context.

The ethics of AI cannot be retrofitted. Once a teenager has used a chatbot every day for a year, "and by the way, here's how the data was sourced, who got hurt, what bias means, and what this is doing to the labor market" is not a conversation. It is an interruption.

So the order of operations should be reversed. Before high-school students are encouraged to use AI in school or at home, they should learn:

• Where the training data came from, and who consented.
• How model outputs can be confidently wrong.
• What bias is, what it looks like in real systems, and who pays for it.
• The labor of the data labelers, content moderators, and creative workers whose work was used.
• The environmental cost of large-scale training.
• The difference between "the model produced this" and "you produced this."
• How AI companies make money — and what role the user plays in that economy.

Only after that ethics foundation is in place — call it a semester, call it a year — should AI tools be introduced as a supervised, deliberate part of the learning environment.

03Across all ages: lean into being human.

Most of the conversation about kids and AI is framed as a deficit conversation. They'll fall behind. They won't be ready for the future. Other countries are doing it.

The deficit conversation has the arrow pointed the wrong way.

A child who reaches 18 without strong handwriting, sustained reading attention, the ability to talk to a stranger, the ability to be bored, embodied play, the experience of making something poorly and finishing it anyway, real friendships built without an algorithm in between, and the practiced muscle of forming their own opinion — that child is the one who is behind. AI fluency is the easy part. It can be taught in a season. The other capacities take a childhood, and you cannot bolt them on later.

Our advice — to parents, to schools, to ourselves — is not "delay AI" as a defensive crouch. It is "lean into being human" as a positive program.

That means more time outside, fewer screens. More books, fewer feeds. More boredom, less stimulation. More handwriting, more drawing, more music made badly with hands. More conversations with adults. More chores. More reading aloud. More noticing.

These are not nostalgia. They are the things AI is the worst at and humans are the best at, and they are the things that make a person a person. Childhood is where you build them, or you don't.

What we'll grant the other side

We believe what we just wrote. We will also push back on ourselves.

There are children for whom AI is a genuine accessibility breakthrough — text-to-speech, speech-to-text, communication aids, language scaffolding for English learners, visual support for blind students. A blanket "no AI under 15" written by people who have never needed assistive technology is not the rule those students need.

There are also children in under-resourced schools where a thoughtfully deployed AI tutor may, in fact, do more good than the alternatives. The honest answer is not "never" — it is "rarely, deliberately, and with adults in the room."

And the line between "using AI" and "using a product that contains AI" is already blurred and will get worse. Spelling correction is AI. Search ranking is AI. The line we draw is between generative, conversational, time-consuming AI — the kind that competes with the activities of childhood — and the long tail of AI that quietly improves background tools.

We are not arguing for abstinence. We are arguing for the reordering of priorities so that the most important developmental years are reserved for the things that build a human, and AI is introduced — when it is introduced — slowly, ethically, and with eyes open.

Why this matters now

The companies building these systems have an incentive to put them in front of children as early as possible. Schools, under pressure to look modern, have an incentive to adopt them quickly. Parents, under pressure to look like they aren't holding their kids back, have an incentive to consent.

Almost no one in the system has an incentive to say wait.

Childhood is not a runway for AI adoption. It is where you build the human who will one day decide what AI is for.

What we're committing to

The AI Impact Foundation will not build child-facing AI products and will not partner with companies that target generative AI at children under 15. Our high-school and college education tracks will lead with AI ethics — provenance, bias, labor, environment, harm — before we put a tool in any student's hands.

We think that is the floor, not the ceiling. We think the rest of the field should meet us there.

That's the work. Come help us do it.