Johan sits at the dining table with a History assignment due the next day. Ten years ago, he might have opened a textbook. Five years ago, he would probably have searched Google and clicked through several websites. Today, he can type a question into a chatbot.
Within seconds, it explains the topic, suggests an introduction, produces five arguments and offers to rewrite everything in the style of a 14-year-old student. His assignment is done but did any learning take place?
This is becoming an increasingly difficult question for parents and teachers. Artificial Intelligence (AI) has entered children's education remarkably quickly. Unicef reported in June that an estimated 13 million children, across ten countries, were already using AI to help with learning and homework. Children were more than three times as likely as their parents to use AI.
We should probably accept one thing early: AI is not going away. Trying to keep children completely away from it may eventually make as little sense as telling them never to use a calculator or search engine. Used properly, generative AI can be an extraordinary learning tool.
A child struggling with fractions can ask for another explanation without feeling embarrassed. A student learning English can practise writing and receive immediate feedback.
A teenager revising Biology can ask an AI tutor to generate ten questions on photosynthesis, answer them and then request harder ones. The tutor never becomes impatient. That is quite a remarkable resource to have sitting inside a phone. The problem begins when the tutor starts doing the thinking.
The struggle we shouldn't remove
Learning has always involved a certain amount of discomfort. Remember being confronted with a Mathematics problem and not knowing where to begin? Or staring at a blank page before writing an essay? I remember approaching my Additional Mathematics teacher, Mr Chu, with a problem I could not solve. Instead of showing me the answer, he said, “Show me what you have done so far, and I’ll show you where you went wrong.
"That frustrating period before the answer comes is not necessarily wasted time. The brain is retrieving information, testing possibilities, recognising what it does not understand and trying again.
AI can remove that struggle in seconds. Researchers call this cognitive offloading: transferring mental work to an external tool. We already do it all the time. We store telephone numbers in our phones and use GPS instead of memorising routes. Offloading is not inherently bad. It frees our brains for other tasks.
But there is a difference between using a calculator after understanding multiplication and using one before learning what multiplication means. The same distinction matters with AI.
Recent research gives us some clues. A 2025 experiment involving high-school Mathematics students found that access to generative AI substantially improved performance while students were using it. But students given relatively unrestricted AI assistance subsequently performed worse when the AI was removed.
An AI system deliberately designed to behave more like a tutor, guiding students rather than simply supplying answers, could avoid much of this problem. The message is not that AI makes children smarter or that AI makes them lazy. How it is used matters enormously.
Answer machine or thinking partner?
Parents can often tell the difference with a very simple question: “What did you ask AI to do?” Compare these two approaches. A child types: “Write me an essay explaining why recycling is important.” AI produces an introduction, arguments and conclusion. The child changes a few words and submits it. Very little thinking was required. Now imagine the child has already written an essay and asks:
> “Here is my argument. Tell me three weaknesses in it, but don't rewrite it for me.” Or:
> “I don't understand this algebra problem. Ask me questions that will help me work it out. Don't give me the answer.” Or even:
> “Quiz me on this chapter. Make the questions harder whenever I get three correct.”
The technology is identical. The intellectual work is completely different. This may be one of the most useful lessons parents can teach children about AI: use it to extend your thinking, not replace it.
AI can sound very sure while being very wrong
There is another problem children need to understand. AI chatbots are remarkably good at producing convincing language.
Unfortunately, convincing language is not the same as correct information. They can give incorrect facts, invent references, misinterpret questions or confidently present an explanation that is partly wrong. Adults get caught by this too.
Children may be particularly vulnerable because authority often sounds like fluency. If an answer is beautifully written, immediate and delivered without hesitation, why wouldn't you believe it?
This creates an opportunity to teach something that will matter far beyond homework. Ask the child: “How do you know that's true?” Find another source. Check the textbook. Look for the original study. Compare what two sources say. Ask AI where its information came from, but don't assume the references it produces exist.
The goal is not merely to catch AI making mistakes. It is to develop a habit of verification. This is a universal principle, both for children and adults.
Don't police every prompt
Parents may be tempted to respond by monitoring everything.
“Did you use ChatGPT?”
“How much of this did AI write?”
“Show me your prompts.”
That can quickly turn AI into another battlefield between parents and children. A better conversation might begin with curiosity. “Show me how you used it.” You may discover that your child is doing something genuinely clever. You may also discover that an entire Geography project appeared during the five minutes you were making coffee. Either way, the conversation becomes about learning rather than catching someone cheating. Families can also agree on a few simple boundaries:
> Try the problem yourself before asking AI.
> Don't submit something you cannot explain in your own words.
> For important facts, check another reliable source.
> Don't enter private personal information into an AI system.
> Occasionally, do the work without AI at all.
That last point matters. A child who can cycle only while someone holds the bicycle has not quite learnt to cycle.
The question after homework
Perhaps our old way of judging homework also needs to change. When an assignment looks unusually polished, parents traditionally ask, “Did you do this yourself?” In the AI era, that question becomes harder to answer. A more useful question may be: “Can you explain what you wrote?” Close the laptop. Ask the child to tell you the argument and why they chose that answer. Ask what part they found difficult and whether they disagree with anything the AI suggested.
If they can defend their reasoning, correct mistakes and explain the subject in their own words, AI may genuinely have helped them learn. If they cannot, the beautifully formatted assignment tells us very little.
Our children will almost certainly grow up with tools that can write and calculate faster than they can and retrieve information faster than any human could.
Competing with the machine is therefore not the point. Children still need to learn how to question an answer, recognise when something does not make sense, tolerate the frustration of not knowing immediately and form an opinion that belongs to them. AI can become a very good tutor but someone still has to teach the child to think.
Dr Naveen Nair Gangadaran is a paediatrician and committee member of the Malaysian Paediatric Association and Perinatal Society Malaysia. This is the first of four articles by Dr Naveen on Growing up in the AI age. The information provided is for educational and communication purposes only, and should not be considered as medical advice. The Star does not give any warranty on accuracy, completeness, functionality, usefulness or other assurances as to the content appearing in this article. The Star disclaims all responsibility for any losses, damage to property or personal injury suffered directly or indirectly from reliance on such information.
