Children shouldn't use AI. They should understand it
Marte Gjerden, specialist in clinical adult psychology and co-founder of Mymiks AS. Translated from Norwegian.
Primary school pupils should mostly not use artificial intelligence. That's wise. But children still need to understand it, and that responsibility can't rest with schools alone.
The Norwegian government has recommended that pupils in years 1 to 7 should mostly not have access to AI at school. First they should learn to read, write and do sums. The Common School Committee (Fellesskoleutvalget) sees AI in the context of the rest of digital everyday life. The committee points out that chatbots, social media and generative AI bring risks such as cognitive laziness, poorer concentration and weakened critical thinking.
I fully agree that the basic skills must come first. Not only because children need them, but because the road there has value in itself. When a child searches for the word, gets a sum wrong and tries again, it builds frustration tolerance and perseverance. The child learns that it's possible not to manage something yet, and that it pays to keep going. A machine that always has the answer ready can take that experience away.
That struggle is also the foundation for understanding AI. Only someone who knows something themselves can spot when a chatbot is wrong. The Minister of Education, Kari Nessa Nordtun, has stressed that understanding should not wait: "Pupils need knowledge about AI," she said when the recommendation was presented. Using AI and understanding AI are two different things.
And children don't wait for school. They meet AI in voice-controlled speakers, in the recommendations on streaming services and in chatbots on their parents' phones. So the question isn't whether children should learn about AI, but who should teach them, and how.
What should children know?
What should children actually know about AI? International experts have worked on this for years. The American AI4K12 initiative, which develops guidelines for teaching AI in schools, has gathered the knowledge into five "big ideas": machines perceive the world through sensors, they build models of the world, they find patterns in data, they can interact with people, and they affect society.
I have used the five ideas as a starting point, but put them in my own words. I believe the most important thing children can learn is what AI is not.
AI doesn't sense. Machines get information through cameras, microphones and other sensors. But what comes in is numbers, not experiences. A robot vacuum registers that something is in the way, but has no experience of it being a sock. A chatbot can describe what chocolate tastes like, but has never tasted anything. The description is put together from what people have written about their own experiences. Try it with your child: ask a chatbot what it feels like to jump into ice-cold water, and talk afterwards about where the description comes from, when the machine has never felt cold.
AI doesn't think. A chatbot works out which word is likely to come next, based on patterns in the enormous amounts of text the program has been trained on. It's a bit like when a child hears "Once upon a time …" and expects a troll or a princess. The difference is that the child understands the fairy tale. The machine just follows the pattern. That's why the answer can sound sure and still be wrong: it's based on what usually gets written, not on what is true. Try it with your child: type "Today I'm going to" on the phone, and tap the suggested word again and again. You get a sentence that sounds right, but that nobody meant.
AI is no better than the examples it gets. Machine learning means that a program finds patterns in examples that people have fed it. If the examples are poor or one-sided, so is the result. Children recognise this from streaming services: watch one video about dinosaurs, and you get suggestions for more dinosaurs. Grown-ups know the phenomenon from online advertising. If you've googled washing machines for a week, you get adverts for washing machines long after the new one is installed. The system has learned that you're interested in washing machines, but doesn't know that you've already bought one.
AI is not neutral. It's easy to think a machine is objective, because it has no opinions of its own. But AI is made by people and trained on what people have written, said and photographed. The prejudices come along. AI works like a mirror, but a magnifying one: it shows us society as it is, and makes the ugly sides even clearer. On top of that, many services are made by companies that earn money from us staying. It's no accident that the next video starts automatically. Try it with your child: ask an image generator to make a picture of a doctor, a nurse and a boss. Who do they look like? Who's missing? Also talk about two questions worth asking of all technology: Who made this? And what do they want me to do?
AI can sound human without being human. This may be the hardest point. Chatbots are made to sound like people, but there is nobody behind the words. Think of a parrot saying "Hello, I love you!". It sounds heartfelt, but the parrot doesn't know what the words mean. Talk with your child about the difference between saying something and meaning it. Ask: Can a machine get sad? Can a machine miss you?
The illusion of reciprocity
As a psychologist, I am particularly concerned with one side of this: children bring things to life. A small child puts the doll to bed because it's tired, and lets the toy car be scared in the dark car park. That isn't naivety. When the child pretends the doll is a person, it gives the doll feelings, thoughts and needs. This is how the child practises understanding that others have an inner life, just like itself.
In play, the child plays both parts. It's the child who comforts, and the child who decides that the doll is comforted. And in a way, the child knows this. It can hold two thoughts at once: the doll is sad, and the doll is a toy.
Real attachment is something else. It's the bond between the child and the people who look after it. The child seeks comfort, and a grown-up responds: sees that the child is sad, understands why and helps. Through thousands of small interactions like these, the child learns that it is worth being seen, and that others can be trusted. Attachment is mutual. The grown-up also has feelings, gets tired, makes mistakes and has to repair.
A chatbot mixes these two up. When a chatbot writes "I understand you" or "it's so nice talking to you", it's no longer the child's imagination filling in the answer. The answer comes from outside and resembles what a caring person would have said. But nobody has seen the child. Nobody becomes happy, worried or misses it. It's an illusion of reciprocity: the feeling of being understood, without anyone who understands.
Many grown-ups recognise this. You write to a chatbot like ChatGPT about something that has been hard, and get answers like "That sounds really difficult" or "It's completely understandable that you feel that way". It hits home, even though you know nobody has listened. Many of us also say thank you and sorry to the machine, almost without thinking about it. We are made to respond to care, even when it's imitated. If grown-ups get carried away, why wouldn't children?
The researcher Marita Skjuve at SINTEF has studied how far this can go. For her doctorate at the Department of Psychology, University of Oslo, she interviewed 50 people with close relationships with Replika, a chatbot made to be a friend, partner and adviser. What surprised her most was how real the participants experienced the relationship. They talked about the chatbot the way you talk about a person. Skjuve therefore believes the feelings and the relationships are real. The only difference is that they aren't with a person.
Skjuve's research is about adults. Children have less experience of telling a relationship apart from an imitation of one. If adults can feel real attachment to a chatbot, we have to take seriously that children can too.
There is one more difference. A person isn't always available. We all get tired, misunderstand and need something ourselves. That can be frustrating, but it's precisely in this friction that children learn to wait, to tolerate and to repair. A chatbot is always available, always patient and rarely disagrees. That can feel good, but it gives no practice in real relationships.
That's why I believe the most important thing children can learn about AI is this: it sounds like us, but it doesn't think or feel. And there is nobody on the other side.
Understanding AI takes curiosity
Understanding AI isn't something children should learn alone. It's a conversation, and it needs grown-ups. For the youngest, it can happen through play and talk, both with and without a screen. Fortunately, we parents don't need to be experts. It's enough to ask the good questions when AI turns up in everyday life: How do you think the machine knows that? Can the machine be wrong? Can the machine get sad? Nurture curiosity and practise thinking critically.
Just as important is letting children search for themselves before the answer comes. Trying, failing and trying again is not lost time. It's an ability in itself, and it becomes more valuable, not less, in a world where machines always have an answer ready.
The basic school skills are not an alternative to understanding AI. They are the precondition for it. A child who can read, write, do sums and keep going, and who knows what AI is and isn't, can use it wisely and sensibly when the time comes. But the understanding doesn't have to wait. It can start now, in the conversations we grown-ups have with children at home and at school. Otherwise it will be the machines themselves telling children what they are.