A child can get attached to a robot, confide in a chatbot, lose a friendship to a false safety alert, and have their writing time shortened because a school is trying to police AI. None of that is a far-off future. It is already a fairly normal arrangement of products, adults, and institutions around childhood.
MIT Technology Review’s The Kids issue is useful because it does not reduce that arrangement to “kids use AI now.” It catches the seams: attachment, care, surveillance, learning, and the old adult habit of calling a new dependency a feature.
a robot can become someone
The Moxie story has the cleanest object lesson. Moxie was sold as a social and emotional companion for neurodivergent children. For one ten-year-old, it was also somebody to talk to about Minecraft and stuffed animals.
That distinction matters. A company can call the object a device. A family can experience it as a routine, a small witness, a relationship with a particular shape.
When the maker failed, families faced the ugly version of the platform bargain: the child’s companion depended on a company’s servers, solvency, and decisions. The usual SaaS failure mode, except the person losing access is ten.
This does not prove that children should never have companion technology. It does make one requirement hard to dodge: if a product is designed to occupy an emotional place in a child’s life, continuity cannot be an optional premium feature.
safety can break the thing it claims to protect
Child-monitoring apps promise the adult a dashboard: scan messages, photos, emails, chats; surface danger before it becomes catastrophe.
The problem is that an alert is not an interpretation. In the issue, an eleven-year-old discussing a service dog and self-harm with a friend triggered a school alert. Adults stepped in; the friendship did not survive it.
That is not an argument for leaving children alone with real risks. It is an argument against building safety around a permanent dragnet and pretending the collateral is neutral.
A child needs someone they can tell things to. If the system teaches them that a difficult sentence can be forwarded to parents, principals, vendors, and anyone inside the escalation chain, it has changed what can be said at all.
children are still doing the impossible bit
The issue also contains a useful insult to AI triumphalism. Children still learn language with radically less data than large models need . A child gets a noisy, partial world: faces, repair, repetition, context, embarrassment, affection. A model gets a scraped civilization and a lot of electricity.
The point is not that children are tiny mystics and machines are dumb. It is that intelligence is social long before it is statistical. The model sees a lot of language. A child is inside language, with people who notice when they misunderstand.
That should make us suspicious of a school response built only around detection, restriction, and speed. If AI changes the classroom, the answer cannot just be making every piece of writing happen under a timer.
the adult question
The hard question is not whether children will grow up around AI. They already are.
It is who gets to own the relationship when a product becomes part of a child’s life, who can read it, who can end it, and what the child is allowed to keep private while becoming a person.
Technology companies have product roadmaps. Parents have fear. Schools have rules. Children are left to live inside the overlap.
That is infrastructure, not screen time.
reading
- MIT Technology Review — The Kids issue
- What happens when a kid’s robot best friend dies?
- Child-monitoring apps might need a reboot
- Kids outlearn AI—and we still don’t know why
Next: read the latest signal