AI Is Becoming Part of Childhood. Are We Teaching Kids to Use It Without Letting It Think for Them?

Artificial intelligence is moving into childhood faster than researchers, schools and parents can fully understand its consequences.

That warning is driving new research at The Australian National University, where psychologists are speaking directly with children aged nine to 12 to understand how they use AI chatbots, what they talk about, why they turn to them and how those interactions may affect friendships, wellbeing and decision-making.

The ANU study is particularly interested in children who treat AI as more than a homework tool. Researchers want to hear from children who have AI “friends”, discuss their feelings with chatbots or ask artificial intelligence for advice about social and personal problems.

Lead researcher Associate Professor Amy Dawel, from the ANU School of Medicine and Psychology, says researchers urgently need a much better understanding of what AI is doing inside children's everyday lives.

That concern is not theoretical.

Recent Australian eSafety Commissioner research found 78 percent of children aged 10 to 17 had used an AI assistant, while one in five users were interacting with AI daily or more often.

More than half of children who had used AI assistants or companions said they had used them for personal or social reasons, including advice about life situations, physical health, mental health or their feelings.

For New Zealand, the questions being asked across the Tasman are already highly relevant.

AI is not waiting until university.

It is already inside New Zealand primary schools, bedrooms, homework routines and conversations.

And the issue is becoming less about whether children will use AI and much more about what parts of their thinking we are comfortable letting AI do for them.

Children are talking to machines about more than homework

The Australian research shows how quickly the role of AI has expanded.

Most children still use chatbots for practical purposes.

Among Australian children who had used an AI assistant or companion, 91 percent used them for information, learning or translation, 70 percent for writing or editing, and 68 percent for entertainment.

But the more striking numbers sit elsewhere.

Among users:

The findings challenge the common assumption that a child using ChatGPT is simply trying to finish an assignment faster.

For some children, AI is beginning to occupy territory previously held by parents, siblings, friends, teachers or counsellors.

That is precisely why ANU researchers want conversations with children rather than relying only on usage statistics.

They want to understand what those interactions actually feel like from a child's perspective.

Why a chatbot can feel surprisingly human

Modern AI systems are unusually persuasive because conversation is their interface.

A child does not necessarily experience a chatbot in the same way they experience a search engine.

It answers directly.

It remembers context.

It responds instantly.

It rarely appears impatient.

It can produce reassurance in seconds and can imitate empathy even though it has no emotions, childhood experience or personal responsibility for the consequences of its advice.

That distinction may be obvious to an adult familiar with AI.

It is less clear how consistently a nine-year-old or 11-year-old understands it.

Australia's eSafety Commissioner has warned that AI is not a trusted adult and should not replace human relationships when children need support about their health, wellbeing or difficult situations.

Privacy can disappear without a child realising it

Another concern raised by the ANU team is the amount of information children may reveal in ordinary conversations with AI.

Children may understand that they should not type their address into a chatbot.

But privacy is not limited to names and addresses.

A child might upload a photograph to ask AI to turn them into a pirate.

They might describe a fight with a classmate.

They might reveal where they go to school, what sports team they play for, their family situation or details about their mental health.

Australian research found almost one-third of children who had used AI assistants or companions had shared personal data or potentially sensitive information.

That makes AI literacy partly a privacy issue.

But it is also becoming a thinking issue.

New Zealand children are already using AI extensively

This is where the Australian research becomes especially relevant to New Zealand.

The Education Review Office released a major national review in July examining how AI is already being used in New Zealand schools.

Its findings show this is not something New Zealand can treat as a future problem.

Around three in four students are already using AI for schoolwork.

Among primary students surveyed by ERO, 62 percent used AI to find information, 54 percent to generate ideas, 48 percent for spelling and grammar, 39 percent for solving problems and 30 percent for writing essays.

Among secondary students, usage was even higher: 72 percent used AI to find information, 63 percent to generate ideas, 50 percent for solving problems and 41 percent for writing essays.

Perhaps more importantly, 58 percent of primary students and 77 percent of secondary students said they were teaching themselves how to use AI.

That should make parents and educators pay attention.

Children are not waiting for formal AI education.

They are developing their own habits now.

Even primary school children are using it

A separate New Zealand Council for Educational Research study found roughly nine in ten Years 5 to 8 students surveyed had heard of AI and more than half had already used generative AI tools.

Some used AI for drawing images.

Some used it for homework ideas.

Some were already chatting with AI “like a friend”.

NZCER also found most students did not recall receiving adult guidance about when or how they should use AI.

That leaves an uncomfortable gap.

Children have access to highly sophisticated systems capable of writing, reasoning, explaining, persuading and simulating conversation, while many are still learning through experimentation and peers.

The concern is not that AI makes children stupid

That would be an easy headline and a poor reading of the evidence.

AI can help children learn.

It can explain a difficult concept differently.

It can generate examples.

It can provide language support.

It can help a child who is stuck get started.

It can make learning more accessible for students who need information presented in different ways.

The concern begins when assistance becomes substitution.

There is a major difference between:

“Explain this maths problem another way because I don't understand it.”

and

“Give me the answer.”

There is a difference between:

“Read my paragraph and ask me questions that might improve my argument.”

and

“Write the paragraph for me.”

There is a difference between:

“Give me three ideas and I will decide which one makes sense.”

and

“Tell me what I should think.”

That distinction may eventually become one of the most important educational questions of the AI era.

What happens when children outsource the struggle?

Learning is not always efficient.

Sometimes it is frustrating.

A child has to stare at a maths problem.

They have to remember information.

They have to construct a sentence and realise it sounds wrong.

They have to argue with themselves.

They have to sit with uncertainty.

They have to make mistakes.

Much of what we call learning happens inside that struggle.

Generative AI can remove it almost instantly.

That is useful when AI removes unnecessary friction.

But it becomes more troubling if the system repeatedly performs the mental processes that children are supposed to be developing.

Researchers describe this as cognitive offloading, moving part of a thinking task to an external tool.

Cognitive offloading itself is nothing new.

Calculators do it.

GPS does it.

Search engines do it.

Writing something in a notebook means we no longer have to remember it.

The question is which functions we outsource, at what age, and how often.

Research is starting to identify a critical-thinking problem

The evidence is still developing and much of the strongest research involves university students rather than children, so claims about long-term effects on young brains need to be made carefully.

But emerging studies are raising legitimate concerns.

A 2026 study involving 353 university students found that deeper cognitive offloading to generative AI was associated with greater relinquishment of cognitive autonomy.

Researchers warned that sophisticated use of AI did not automatically produce better thinking and argued that education should concentrate on metacognition, meaning awareness and control of one's own thinking.

Another 2025 study involving 580 university students found greater AI dependence was associated with lower critical-thinking levels, with cognitive fatigue partly explaining the relationship.

Research published this year involving postgraduate student workflows also identified what researchers called an “efficiency trap”, where students believed AI was strengthening their thinking while in practice bypassing parts of foundational learning.

These studies do not prove that ChatGPT causes cognitive decline in children.

The populations are different, some findings rely on self-reporting, and long-term childhood evidence remains limited.

That uncertainty is exactly why the ANU work matters.

We need research before habits become deeply established.

New Zealand teachers are already seeing the problem

New Zealand does not need to rely entirely on overseas studies.

ERO found teachers are already concerned that some students use AI in ways that bypass learning.

Thirty-eight percent of secondary students and 19 percent of primary students reported using AI for schoolwork in ways they knew they should not.

Three-quarters of secondary teachers had encountered inappropriate student AI use at least occasionally.

ERO's report explicitly identified concerns about potential effects on critical and creative thinking and warned that confidence using AI does not necessarily mean students know how to use it critically.

That should be a warning against equating technological fluency with intellectual development.

A 12-year-old may become excellent at prompting an AI system without becoming better at solving a problem independently.

The danger of losing logical persistence

One particularly important skill at risk is not simply “intelligence”.

It is persistence.

Before AI, when students did not immediately know an answer, they generally had several choices.

Ask a teacher.

Read.

Search.

Discuss.

Experiment.

Think.

Now another option sits permanently in their pocket:

Ask AI and receive a polished response within seconds.

If that becomes the default response to difficulty, children may get less practice in the uncomfortable but essential process of working something out.

Logical reasoning develops partly through repeated exposure to problems where the answer is not immediately available.

So does judgement.

So does creativity.

So does the ability to recognise that your first idea may be wrong.

If every blank page is instantly filled, we should at least ask what happens to the ability to begin with nothing.

New Zealand's policy is already moving, but families matter just as much

The Ministry of Education currently tells schools that generative AI can enhance learning, but that student work must remain authentic.

It warns that AI outputs can contain plausible but incorrect information, advises users not to enter personal information and notes that many AI systems may reproduce dominant cultural assumptions and perform poorly with mātauranga Māori, te reo Māori, Pacific languages and Polynesian cultures.

Schools are also required to have policies covering acceptable AI use in NCEA assessment, and generative AI is not permitted in NCEA external assessments.

But rules at school cannot control everything happening at home.

And that may be why the most useful lesson from the ANU researchers is also the simplest.

Talk to children.

Parents do not need to become AI experts

A parent trying to keep up with every AI platform is fighting a losing battle.

The technology will change too quickly.

A better approach is understanding how the child is using it.

Ask:

• What do you use AI for?

• What was the last thing you asked it?

• Have you ever asked it for advice about friends?

• Have you told it anything private?

• How do you know when its answer is wrong?

• Do you ever check what it tells you?

• What schoolwork do you think AI should not do for you?

• Would you trust the same answer if a stranger online gave it to you?

These questions are more useful than simply banning a particular app.

They teach judgement.

Perhaps children should answer before AI does

One practical rule families and schools could consider is remarkably simple.

Think first. AI second.

Ask the child to attempt the problem before consulting AI.

Write the first paragraph themselves.

Form their own opinion.

Predict the answer.

Explain their reasoning.

Then use AI to challenge, compare, explain or improve it.

That changes AI from a replacement for thinking into a tool for thinking.

A child who asks AI to critique their reasoning may learn something.

A child who repeatedly asks AI to produce the reasoning may gradually practise less of it themselves.

That difference matters.

We should not repeat the social-media mistake

There is a wider lesson here.

Society embraced smartphones and social media at enormous speed.

Only much later did parents, researchers, governments and technology companies begin seriously debating what constant connectivity, algorithmic feeds and addictive design meant for children.

AI is moving faster.

We have an opportunity not to repeat exactly the same sequence.

That does not mean frightening children away from technology.

AI will almost certainly be part of their universities, workplaces and adult lives.

Young New Zealanders will need to know how to use it.

But the goal should be AI literacy, not AI dependency.

The most technologically prepared child of the future may not be the one who asks AI for everything.

It may be the one who knows when not to ask.

The conversation New Zealand needs now

The ANU researchers are trying to understand a childhood experience that barely existed a few years ago.

New Zealand should pay close attention.

Our own evidence shows primary-age children are already using AI, many are learning how to use it largely by themselves, and schools are still developing consistent approaches.

There is enormous potential here.

AI could give children extraordinary access to explanations, creativity, language support and personalised learning.

But there is also an increasingly obvious risk.

If children routinely outsource writing, remembering, analysing, deciding, problem-solving and even emotional advice, they may be handing over precisely the abilities childhood and education are supposed to develop.

The right question is therefore no longer:

Should children use AI?

They already do.

The better questions are:

What should AI help them do?

What should they still learn to do themselves?

And perhaps most importantly:

Are we raising young people who know how to use artificial intelligence, or young people who gradually stop trusting their own?

That conversation needs to begin while the habits are still being formed.

Sources: The Australian National University; Australian eSafety Commissioner; New Zealand Education Review Office; New Zealand Ministry of Education; New Zealand Council for Educational Research; peer-reviewed research on generative AI and cognitive offloading.