Today a Year 9 student started building a tool to make her soccer coach's life easier. Team lists, training plans, the endless admin that sits behind a Saturday morning game. She'd watched the problem up close and decided she could fix it.
A Year 7 boy spent his session designing something to help him study. He struggles with organisation, he knows it, and he's building a system shaped precisely around how his own brain lets him down. It's software nobody needs. Nobody except him. He has complete control over building a tool that is ultra-custom to his own life, and there is no product on the market that could ever be that.
Another Year 7 student, that same morning, was building a motivation tool because he wants to become a better musician. Practice tracking, goal setting, some way of seeing his own progress that means something to him.
There was more. There usually is.
Here is the thing I keep turning over. Not one of those three projects survives a locked-down device. And right now, across staffrooms, board papers and newspaper columns, we are busily talking ourselves into locking them down.
The argument, stated plainly
A popular position has formed about children and technology. It runs roughly like this: screens are harming young people, the evidence is in, and the responsible thing for schools to do is restrict, filter, ban, and where possible remove the devices altogether. It is delivered with confidence, it is repeated by people who many educators respect, and in some form it is now the default assumption in a great many schools.
I think that position takes a real body of evidence and draws the wrong conclusion from it, and I think the cost of that wrong conclusion falls on exactly the students who would benefit most from the alternative.
Yong Zhao's recent paper in the ECNU Review of Education gives that alternative a name. He calls it the courageous minority: the small group of teachers, students and leaders willing to redesign learning inside the spaces they actually control, before the system is ready and without waiting for permission that may never arrive. That minority exists in most schools. It is where anything genuinely new in education has ever started.
It is also entirely dependent on being allowed to touch the tools. Which is why the headlines matter so much more than they appear to. Follow them, and we lock the devices down, or remove them, and we quietly foreclose one of the most interesting opportunities in the history of school learning before we have seriously tested what it can do. It would not be the first time.
It is the oldest move in education: meet change by restricting it, then wonder later why nothing changed.
What the students are actually doing
Start with what the alternative looks like in practice, because the popular position rarely describes it.
We use AI in these classes constantly. It is not the author of the work. It is the apprentice.
The student holds the intent. The student decides what the thing is for, who it serves, what counts as working. The AI helps them build faster and reach further than a twelve-year-old could reach alone, which means the ceiling on what they can attempt has lifted enormously. A student who would once have spent a term learning enough syntax to make a button do something can now spend that term on the harder and more interesting question of what the button should do, and for whom, and whether the answer is any good.
The framing is not mine. Ken Kahn set it out in The Learner's Apprentice: AI and the Amplification of Human Creativity, and the subtitle is the argument. Kahn began his doctoral work with Papert and Minsky in the MIT AI Lab in 1973, spent decades building programming environments in which children could construct with machine intelligence rather than receive from it, and then, when generative AI arrived, turned to what happens when a child can converse a complex program into existence. His book documents hundreds of these projects. Its premise is that the chatbot works best as apprentice, co-thinker, pair coder and illustrator, an intellectual ally that amplifies what a young person can do rather than a machine that does it for them. Amplification of a child's creativity is a different educational proposition from automation of a child's work, even when the same tool sits underneath both.
Seymour Papert drew the line fifty years ago in Mindstorms. He wanted classrooms where "the child programs the computer" rather than the computer programming the child. He was not romantic about it. He was worried, specifically, that schools would take the most open-ended object ever invented and put it to work doing exactly what they already did, only faster. Content delivery with better graphics.
AI makes his question sharper rather than obsolete. Point the technology one way and you get a more sophisticated teaching machine: students observed, measured, sorted, and moved along a path someone else designed for them. Point it the other way and you get a workshop for the imagination.
The popular position has an answer to Papert's question. It just doesn't realise it is answering it.
Where the evidence stops and the policy starts
The case for restriction deserves a fair hearing. Read closely, it actually opens a far more interesting conversation, one that points at something many of us are not acknowledging, or perhaps would rather not.
J. C. Horvath told a US Senate committee this year that children's cognitive development has stalled and in some domains reversed, alongside declines in literacy, numeracy, attention and higher-order reasoning. His sharpest point should stop any of us who lead technology in schools: swapping one type of screen for another solves nothing, because screen-based technology can undermine learning whenever it displaces forms of teaching and human interaction better matched to how cognition actually works. Alongside that sits Maryanne Wolf on the erosion of deep reading, Jonathan Haidt on attention and belonging in screen-saturated schools, Twenge and Campbell on wellbeing, and the MIT cognitive debt study, in which participants who wrote essays with a large language model showed weaker recall and a diminished sense of authorship over work they had supposedly produced.
I take all of that seriously. Every one of those findings describes the same thing: a machine standing in for the thinking. Displacement. Offloading. Consumption.
Now watch what happens between the evidence and the policy. The research measures what a screen does when it replaces cognitive work. The recommendation that follows is about devices, all of them, in all their uses, for all students. That is a jump, and it is doing an enormous amount of unexamined work.
Nothing in the cognitive debt study tells us what happens when a fourteen-year-old spends six weeks arguing with a machine about why her code will not do what she wants, failing, revising and failing again. Nothing in Horvath's displacement argument applies to a screen that is not displacing the thinking but is where the thinking is happening. If the device is replacing thought, remove it or redesign the task. If the device is the workshop, taking it away removes the work and leaves the child exactly where the research says we found them.
The evidence is about consumption. The policy is about possession. Those are not the same claim.
The failure the technology exposed
Zhao's paper puts a sharper edge on this. His argument is that the current alarm about AI in schools, the cheating, the weakened effort, the cognitive dependence, is a set of symptoms rather than the disease. Schools still teach and assess a great deal of work that AI can now do competently. Summarise this. Explain that. Produce an acceptable essay on a topic chosen by somebody else. When the destination is one a machine reaches in nine seconds, students will use the machine. That is not primarily a story about character. It is a story about the task.
Which is not to wave character away. Character formation is one of the most necessary conversations in education right now, and it becomes more urgent rather than less in an age of capable machines. My discomfort is with how thin our version of it has become. Framed around AI, character education too often amounts to compliance in nicer language: do what the system has told you is right, do not use the tools to cheat, sign the agreement. That produces students who follow a rule while it is being enforced.
The character I see forming in these classrooms comes from somewhere else. It shows up when a Year 9 student notices her coach drowning in admin and decides that is a problem worth her time, or when a boy builds a study tool around a weakness he has had to be honest with himself about. Empathy for a real person, an idea wrestled with rather than received, the discipline to keep going when the thing does not work. Integrity grows out of ownership. A student who cares about what they are making has a reason not to fake it that no policy document can supply.
AI did not create this obsolescence. It revealed it.
Which reframes the whole restriction debate. Tighter rules, better detection, device bans: Zhao calls this improvement change, making the existing model run more smoothly without asking whether the model still holds. He sets it against transformational change, which asks what is actually worth learning when much of the old output can be generated on demand. Improvement change is common because it threatens nobody. Transformation is rare because it threatens the settlement that keeps schools stable.
Restriction is the ultimate improvement change. It leaves every assumption about curriculum, task and assessment exactly where it was, and it feels decisive while changing nothing.
It also does not work especially well on its own terms. Research Zhao cites found students navigating AI through peer culture rather than official guidance, amid policy confusion and a kind of shame around use. Prohibition rooted in suspicion rather than redesign does not remove AI from a young person's life. It removes the adults from the conversation.
And there is an irony in the paper worth sitting with. Governments and school systems spent an enormous amount of money and political capital connecting every student to the internet, on the grounds that access was essential to a modern education. Many of those same systems are now spending comparable effort restricting student access to that same internet. Zhao reads this reversal as revealing. Less a considered judgement about a specific technology than a loss of confidence in our own ability to make anything useful of it.
Read the methods section
If the popular position is now the default, this is precisely the moment someone should be checking the working. Peter Gray has been doing that, at real personal cost.
Gray is a research professor at Boston College, author of Free to Learn, and a co-founder of the nonprofit Let Grow. He co-founded it with Jonathan Haidt. They are friends. Haidt sent him a pre-publication draft of The Anxious Generation, and Gray read it and told him he could not support it, then published a detailed critique explaining why.
His central complaint is methodological, and it happens to be the exact capability the ACARA digital literacy results say our students are losing. When a book cites research supporting its thesis, do not assume the studies show what the author says they show. Read the methods. Gray demonstrates it on one experiment Haidt offers as causal evidence, an undergraduate study that limited social media use for three weeks. Participants could easily guess the hypothesis, so demand effects were uncontrolled. There was no placebo condition, in a study measuring precisely the sort of subjective wellbeing that placebo strongly moves. And after all that, the findings were weak and inconsistent anyway, with no significant effect on overall psychological wellbeing, anxiety, self-esteem, autonomy or self-acceptance.
The broader evidence he assembles complicates the story further. Teen mental suffering rose sharply between 1950 and 1990, well before the internet, peaking at a level comparable to today. Cross-national comparisons are inconsistent, with the pattern appearing in some English-speaking countries and not across much of the rest of the world. Candice Odgers, among the leading researchers in this field, reviewed Haidt's book in Nature and concluded that the causal claim is not supported by the science, and that the focus on social media risks distracting us from the actual drivers.
Then there is the finding that ought to make every school uncomfortable. When teens themselves are asked what is causing their distress, the most common answer is school pressure.
Which raises a question I cannot fully answer and cannot stop asking. Part of the appeal of the smartphone story may be that it locates the problem outside the school gates. A device is a satisfying culprit. It can be confiscated. Blaming it requires no one to look at assessment loads, at narrowed curriculum, at what we ask students to do all day and how little of it they chose.
Gray is not claiming screens are harmless. He says plainly that there are dangers in the digital world as there are in the physical one. His argument is about our reflex.
As a society we treat the removal of one more freedom as the natural solution to any problem children face. Peter Gray
He has spent three years assembling the full case, and it arrives in September as Restoring Childhood: How to Set Kids Free in the Age of Anxiety.
I am not qualified to adjudicate between these researchers, and I am wary of reaching for whichever one supports what I already wanted to do. Haidt and others have answers to Gray. The honest position is that this is contested rather than settled, and that is the point. When a claim about children arrives presented as settled, and the policy that follows from it is restriction, the responsible response is to open the methods section rather than the compliance manual.
Zhao's courageous minority works inside schools. Gray is doing the same job inside the evidence, questioning a consensus he is personally and professionally embedded in because he has read the studies and does not think they say what everybody is repeating. Same instinct, different room.
Where this actually gets decided
None of this argument is settled in a debate. It is settled in a provisioning decision, usually made by someone who was not in the room when the pedagogy was discussed.
Whether a student can build the thing they imagined is determined long before the lesson starts, by how we provision the technology. A device, a platform, a set of permissions, a filtering policy: each is a teaching decision wearing a technical disguise. Provision with real thought, giving genuine access, adequate resources and enough flexibility for a student to build something you did not anticipate, and the workshop stays open. Default to locking everything down and Papert's question has been answered on the student's behalf, badly, before they walked in.
Zhao warns that AI is being adopted inside schools in ways that reinforce the controlling logic of traditional schooling rather than disrupting it: attendance tracking, behaviour monitoring, engagement analytics, automated pathways through predetermined content. A systematic review he cites found such use extensive, while ethical questions of privacy and bias received far less attention. Technology deployed to watch students is technology that has already picked a side.
Gray's point about the reflex applies with unusual force here, because restriction is the cheapest decision available to anyone in my role. It requires no pedagogical argument, no conversation with a head of department, no explanation to a parent. It photographs well in a policy document, and nobody ever gets called into an office to justify a feature they turned off. The alternative, working out what a particular group of students needs to be able to do and then building guardrails narrow enough to be safe and wide enough to be useful, is slow, contested and comparatively invisible.
This is why educators need to be leading the conversation about student technology access rather than leaving it to procurement, compliance or risk management alone. Those functions matter, and I do a fair amount of that work myself. They are not equipped to answer the question of what a device is for. If the people who understand learning are not shaping what the tools are permitted to do, someone will make that call on other grounds, and those grounds will almost always favour restriction, because restriction is the option that never has to be explained.
Provisioning is the infrastructure of the courageous minority.
Every teacher willing to redesign their classroom is relying on a decision someone in IT made months earlier about what the machine in front of their students is allowed to become.
The gap we are quietly building
Tonight I am heading to a community event run by industry, a free evening where young people from schools right across our region come together to learn about AI and build things with it. Volunteers giving up their time, kids arriving curious, the whole thing designed around making rather than watching.
Some of those students will not be able to participate properly. Not for want of ability or interest, and not because anybody in the room wants to hold them back. They will arrive with school devices configured so that the tools on the table cannot be installed, the sites cannot be reached, the code cannot be run. Different schools, different platforms, different restrictions, same outcome. My students will build. Others will watch someone else build.
I have been thinking about that room for a while now, because of what it suggests about the shape of the disadvantage we are creating.
We have spent two decades worrying about the digital divide as a question of access to hardware, and that work mattered. The divide forming now is a different one. It runs between students who are allowed to use the machine in their bag and students who are not. Same device in many cases. Same capability sitting dormant inside it. One child permitted to find out what they can make with it, the other permitted to consume on it.
That gap will not show up in any equity audit, because on paper both students have a laptop. It compounds quietly, in the projects one of them attempts and the other never gets to attempt, and in what each of them comes to believe about their own relationship to technology.
One learns that a computer is a thing you can bend to your intentions. The other learns that a computer is a thing that tells you no.
Every one of those restrictions was applied by someone acting responsibly, with the wellbeing of children in mind, and usually in response to precisely the evidence and the headlines discussed above. That is what makes it worth arguing about rather than mocking. Reasonable people, reasonable intentions, and a generation of students on the wrong side of a line nobody meant to draw.
What we stand to lose
Put it back together.
The evidence tells us that machines standing in for student thinking produce hollow learning. Agreed. Zhao adds that the reason so much student work is offloadable is that we kept setting work worth offloading. The popular position responds to both of those findings by removing the machine, which leaves the second problem entirely untouched and takes the workshop away from the students who were using it properly.
Zhao's case is that transformation never arrives as system-wide reform. Large reforms get diluted, reinterpreted and absorbed back into the existing grammar of schooling. What actually moves is a courageous minority building alternatives in protected spaces, generating evidence the wider system can eventually notice. He and his colleagues now have around twenty schools worldwide doing this deliberately, comparing notes at whatever hour the time zones allow.
That minority is not asking for much. Access, resources, flexibility, and a device that says yes.
If we follow the headlines instead, we will lock the devices down, and in some places remove them entirely, and we will do it in the name of protecting children from a harm the research describes rather more narrowly than the policy assumes. We will forfeit the chance to find out what a generation of students can build when the tool amplifies them instead of replacing them. We will sort young people into the allowed and the not allowed, school by school, without ever calling it that. And we will have done what schools have done with every significant technology for a century, which is absorb it into what we already did, or refuse it, and then express surprise that nothing improved.
One finding in Zhao's paper stays with me above all the others. When researchers examined the barriers school leaders named as obstacles to innovation, only around a third turned out to be real. The rest were imagined: things that could have been worked around, waived or reinterpreted, but were treated as walls.
David Loader made the same point long before any of this, and he had earned the right to. As a principal he led the first one-to-one laptop programme in the world, at a time when there was no precedent to point to and every reason to wait for one.
While those of us in schools may wish to blame others for imposing constraints, many of our constraints are self imposed. Are we able to escape from these and respond to what we believe is best for our students?
— David Loader, The Inner Principal
I suspect that number holds for school technology too. Most of what we tell students they cannot do is not a limit we have hit. It is a choice we made and stopped examining.
A Year 7 boy is building software that nobody needs except him. He'll finish it, and it will be his. Every wall between him and that is one somebody chose to build.