Every question answered in seconds, every essay polished before it is understood, every doubt settled by a machine that never tires and never says “work it out yourself”. Imagine a child who has never once been stuck. Ten years ago this was science fiction.
Today it is a tab open on most teenagers’ phones.
A 2025 Harvard trial found that undergraduates in a physics course learned more than twice as much from an AI tutor as from a well-run class. A World Bank experiment in Nigeria put six weeks of guided AI tutoring at roughly 1.5 to 2 years of ordinary schooling. Writing survived the warning.) 5 Oct 2026 | 12:06 The same tool that teachers can also carry. A 2025 experiment with 117 university students found that those using ChatGPT produced the best essays and learned no more than those without it. A 1986 review of 79 studies found that calculators, used alongside ordinary teaching, improved children’s math skills at most grade levels. UNESCO’s 2023 guidance reaches the same conclusion in policy language: AI should serve human capability, and independent use should wait until 13.
The authors called it “metacognitive laziness”. Under normal circumstances the case for AI in learning would be made here, and it is a strong one. The generation that grew up with Google never spent an afternoon in a library’s card index, and nobody sensible wants that afternoon back. (Socrates, in Plato’s Phaedrus, warned that writing would weaken memory. The question for a parent is whether the child is learning to walk or learning to be carried. Here are five signs the carrying has begun, and what to do about each. Ask her to walk you through paragraph two. If the answer is a shrug, the essay was written rather than learned. The fix, in many homes the simplest rule and the hardest to keep: explaining the work is the price of the tool. In the same survey, a third of teenagers who use AI companions had chosen the AI over a person for a serious conversation, and a quarter had shared private information with it. This sign has nothing to do with homework, which is why it matters most. The fix: notice who the child turns to when something has gone wrong, and make sure a human gets there first. None of this is an argument against AI, any more than the pocket calculator was an argument against arithmetic. In the PNAS study, the tutor that gave hints rather than answers did no measurable harm to exam scores. Every generation hands its children a tool it does not fully trust, and the children turn out fine when an adult stays in the room. The job of parents and schools is to stay in the room. Rohan Parikh, Co-Founder and Chairperson, The Acres Foundation You use AI every day. Now get your AI Quotient. Take the AIQ test Get the latest Lifestyle News and more. Download the TOI app.
A field experiment published in PNAS in 2025, followed nearly a thousand high-school maths students in Turkey. Those who practised with an unrestricted chatbot did far better on practice problems, then scored 17 per cent worse than students with no AI at all once it was removed for the exam. A 2024 German experiment with university students found that those using a chatbot reported less mental effort than those using ordinary search, and reasoned less deeply. In a 2025 Microsoft survey of knowledge workers, the more people trusted the AI, the less critical thinking they reported; it was the strongest single predictor in the study. Half of American teenagers trust an AI’s advice at least somewhat, a 2025 survey found.
Who says otherwise? The fix: practise with the tool by all means, and test without it, often. Thinking, like a muscle, strengthens under load, and the first unaided attempt is where the load sits. The fix: ten minutes alone with the problem before anything is opened. The fix: three questions, asked at the dinner table until they become a reflex. How do you know? What did it leave out?