“We recommend a minimum age for independent use of generative tools in education and require every use to be accompanied by human verification — a tool that produces text is not a source.”UNESCO · Guidance for Generative AI in Education and Research · 2023 · UNESCO, «Guidance for Generative AI in Education and Research», Paris, 2023 — redare a recomandărilor centrale, necolaționată pe document
The tool is not a source, it is an editor that can be convincingly wrong.
The UNESCO guidance asks two things of schools and both transpose to the home: a minimum age for independent use and a duty of human verification. What goes on the fridge. Line one, under thirteen: the tool is used only alongside an adult, screen in view, and never for homework. At that age the very capacity the tool replaces is being built, so the shortcut costs more than it brings. Line two, from thirteen: they may use it alone under three conditions — first they write what they think, then they ask; any figure or quotation is checked against an independent source; nothing personal about them or anyone else goes into the tool. Line three, for homework: they declare what they used and for what, exactly like a bibliography. Homework handed in as theirs but written by a model is not a technology problem, it is an honesty one, and is treated as such. Three clarifications both of you should hold: the tool is not a source, it is an editor that can be convincingly wrong; it is built to please, so it will agree with them more often than it should, and that is dangerous exactly when they need contradiction; and it is built by someone, with interests, on data with holes. And for you as a parent, the rule that carries the rest: use it in front of them, out loud, showing where you checked and where it was wrong. A child learns their relationship with a tool by watching yours.
The card gives you the rule. The next step is what you do when the rule is broken, because that moment decides whether it was a rule or just a poster. If you find an essay written by the model and handed in as the child's own, don't treat it as a gadget problem; treat it as a lie. So the question isn't "did you use the tool?" but "why did you claim it was yours?". A concrete example: your child submits an essay too good for them. Instead of punishing them for AI, show them the real issue is signing something they didn't think — just like copying from a classmate on a test. Then offer a way back: redo the assignment, this time with a use declaration. That turns a violation into a lesson about honesty, not about technology. But what happens when the child agrees with the rule only in front of you and does otherwise when you're not watching?
The previous passage ended on the hard question: the child who agrees to your face and does otherwise when you're not watching. This is where the house rule proves itself or remains a display. Surveillance won't fix it, because you can't stand beside him at every search. What works is changing the motive: the rule isn't because "the parent says so," but because signing someone else's work deceives yourself first. Example: the child asks the model before thinking, then copies the answer, without technically lying about anything. It breaks no written rule, yet it's exactly the shortcut the rule exists to prevent — he gave up his own thinking without telling you. The conversation isn't about punishment but about what he lost: the chance to find out what he actually thinks. The question opening the next step: what does real verification look like for a thirteen-year-old, not a token one?
Why it matters Because a blanket ban does not survive a single term and produces only hidden, unchecked use, while a rule written by age keeps what must be kept — their own thinking and honesty — without leaving them behind their class.