[This is more or less what I said at the ToddleAI Schools Showcase on 12 September, lightly tidied. It’s more a talk than a post, because it was one; and the images were the slides. The initial audience was for educators considering which AI tools to use and how to use them, but I hope there’s something useful here for parents and others.]
Kids create some great work at school – a drama production, an app created, a poem written maybe. But most of it isn’t so great. Not really – and that’s OK. An essay on photosynthesis, or an algebra exercise, or a recitation of the causes of WW2 – they are there to critique. They are not important in themselves; they are means to an end.
We ask kids to do these things for two reasons :(i) they force students to think and (ii) they give us evidence that they have thought. And well-structured thought, of course, leads to great benefit. Memory is, as Daniel Willingham says, the residue of thought. Now we cannot directly measure memory, so the work is a proxy for the thought, and the benefit. But the work itself is not the point.
We have always known this, even if we forget it occasionally. And it has worked until now. But as we all know, AI means the essay, the algebra, the creative writing, the lists – all can now be created without forcing the thinking or proving that any took place. And that’s why the word “real” in the title is the point. It admits that a lot of what we think we see as understanding may not be ‘real’.
So far, so obvious – and it’s different in schools than anywhere else. A hospital that uses AI still heals people, and health is a real good. Nobody stands up at a medical conference to ask what medicine is now for. Because we know that the product of hospitals – health – is the point. But schools’ products were never the point; they were the proxies, all we had to measure because we cannot see inside students’ heads. So a machine that gives us proxies doesn’t help us; it just means the proxies are even less helpful.
There are two obvious responses – (1) stop students using AI, (2) make sure AI supports rather than replaces learning. This conference is largely about the latter. The first has some merit nevertheless (because, for example, essays are unsurpassed in their ability to force deep thought) but of course is almost impossible to ensure outside exam halls. The second is the open question and I want to come at it through a lens that starts with social systems. And in particular, what actually motivates kids to think in the first place.
One common point I hear and read a lot about is that we need to motivate students by making school work “relevant” or by “connecting it to the real world”. I want to think about the words “relevant” and “real” as used here, because I think they are smuggling in a lot of assumptions that go unchallenged.
Let me tell you via a story that will be very familiar, I am sure. A year or two ago I taught a girl – I’ll call her Kim. Quiet, at the start of the year. Sat near the back, said little, a student you had to work to draw out. And then one morning, in a Theory of Knowledge class, something changed for her. I don’t know why it was, but in a class on paradigms, she started arguing – really arguing, with delight. It was almost fierce – she leant in, her voice rose, she pulled a face at herself when she got tangled in her own logic, but kept at it. It was one of those delightful lessons students and teachers remember. Now, she wasn’t getting it all right – but that’s not my point. My point is that she was completely alive to it, and it was catching – her energy lifted the whole class.
There was a parents’ evening later that day and Kim’s father asked me a perfectly fair question – what job might all this Theory of Knowledge lead to?
Why it was “relevant” or “connected to the real world”, in other words. And I hesitated, because I was still thinking about his daughter in that room, and what I wanted to say was: you should have seen her. But that wasn’t what he’d asked, and I wasn’t quick enough to find the right words. So I said something about critical thinking, transferable skills, interview skills. True enough. But it came out like an apology – because I was justifying something wonderful in the language of the labour market.
Now I tell this story in this context, to show that AI was beside the point – because what happened to Kim wasn’t about the tools, it was about a person catching fire over an idea, encountered in the right way, at the right time. The question is, how can AI contribute to that? I mentioned assumptions earlier – and the big one being smuggled in here is that ‘reality’ or ‘relevance’ comes from usefulness – that a thing is worth wanting because it connects to a life or job – outside the room.
But Kim shows that motivation doesn’t need relevance – she wanted it for no useful reason at all. The father wanted relevance; Kim had something that needed no relevance; and the thing she had is the thing that actually produces thinking.
So what was it that Kim had? And why have I so far said so little about AI in this AI session? What Kim had was absorption; intrinsic motivation, being gripped by the thing itself. Not the same as enjoyment or fun; she wasn’t just having a nice time, she was fully alive to a hard problem.
And I’ve said so little about AI because what actually produces thinking turns out to be nothing to do with tools. Which means that as we introduce AI – which I think we should – we need to be careful because tools are independent of the thing that matters most, which Kim had, which we see in our classrooms when it works. Absorption, intrinsic motivation and encounter are, I would argue, largely tool-agnostic. So can AI generate that motivation? Or at least, not undermine it?
This is not new. We know motivation is critical, and we are good at manufacturing extrinsic motivation – grades, rewards, make it relevant, invest in your future, secure your College place and so on. Even as we know that there is strong evidence that extrinsic rewards can crowd out intrinsic motivation. So there is a danger that our toolkit doesn’t just fail to help; it actively corrodes the thing we want. What good teachers transmit in a room is not information and not affection – it’s a live state, and only a mind currently in that state can pass it on.
So how do we reach for intrinsic motivation? Dan Pink popularised it in his excellent Drive, drawing on the remarkably robust research from Ryan and Deci showing that intrinsic motivation comes when people feel three qualities – autonomy, competence and connection. Now’s not the place to dive into this marvellous body of research, but suffice to say that these are not conditions that can be called up at will; we can make them more likely, but like chemistry and relationships between people, we can’t manufacture them, only gesture at them. AI may help or may not.
And that leads me to my very simple thesis today:
I want to suggest that autonomy, competence and connection cannot be summoned in a lesson or by a tool, but that they can be built into the fabric of a school – because they are made by structures, not websites, apps, speeches or innovation per se.
So how can what we know about motivation ensure that AI innovation lands on fertile ground?
Autonomy doesn’t come from telling children they’re in charge of their learning; it comes from whether we actually allow them some real choices. And there’s tension, of course, because real choices can have real consequences. But a school that removes every consequence teaches its own lesson. In my school we’ve tried to protect the space to get things wrong: assessments you can learn from rather than only be judged by, and a culture that treats a failed attempt as information, not disgrace. We do it imperfectly, and against real pressure – because parents, understandably, don’t pay fees for their child to fail.
So let’s ask: How does AI fit with autonomy? I believe it can do; and we need to ask the question.
Connection isn’t produced by telling children they belong; it’s produced, or destroyed, by how we do the ordinary things. By whether discipline repairs the relationship or severs it. By whether reporting tells a child something true about their own growth, or just ranks them against the person beside them. By whether we hand out awards that, by design, leave most children on the outside. And connection between children doesn’t arrive by magic either – we teach it, deliberately: how to talk when we disagree, how to acknowledge and validate people we disagree with. We’ve put that on the K-12 curriculum.
So let’s ask: How does AI fit with connection? I believe it can do; and we need to ask the question.
Competence – the felt sense of getting genuinely better at something. This ought to be a school’s bread and butter, and in one sense it is – because kids are always learning (albeit at different paces, and in unpredictable spurts). But it may not always feel that way – because a child knows they are getting better when they can see the distance between themselves now and themselves last term, and much of what schools do obscures exactly that: assessment against grade boundaries, ranking against the child next door (that is, by the way, one reason why activities like the orchestra, sports, debate, theatre are so valuable – the progress is much more visible). So we need to find ways to showcase progress, not just attainment; even when attainment is the community focus. Not easy.
So let’s ask: How does AI fit with competence? I believe it can do; and we need to ask the question.
Autonomy, connection and competence are not new; this is familiar to anyone experienced in education. But none of it is finished, none of it is easy, and we need to make sure that AI can help us do it better. And I think that’s the critical lens for this conversation.
All of these systems – assessment, reporting, pastoral care and so on – are places where AI tools can now help, and often genuinely will. And once we see that at the system level, then we can hone in on the fact that AI introduction is more an adaptive change than a technical one. The distinction is that technical problems can be solved with existing expertise, while adaptive ones require the people involved to change how they work (from Ronald Heifetz)- which is to say it’s about how it fits into the broader cultural feel of the school, because that’s where real motivation lies.
So we need to have our eyes open as we ask students to think about the learning, not just create the product. The systems are easy to overlook, because the connection to AI is not entirely obvious; but these are the foundational ground on which AI will either enhance or undermine learning. Used well, AI can help. Used carelessly, it optimises the measurable, allows cognitive offloading and lets the rest slip.
I mentioned connection before, and I meant connection between people. But there’s another kind of connection I want to explore: I think you can be connected not just to a person but to an idea.
What happened to Kim was a kind of connection – connected to the specific problem I had posed, gripped by it. But here’s the thing – that connection, to the idea, is almost always built on a connection to a person. She didn’t fall for the problem in the abstract. She fell for it in a room, with a teacher who was himself caught up in it! The absorption passed between us. I didn’t hand it to her like a worksheet; she caught it, because it was infectious. She could believe me when I told her how I came across this problem and why it fascinated me, and why it was important – she could feel I felt it. I’m currently unconvinced that an AI doing exactly the same thing will be remotely credible. So perhaps connection to ideas will always be mediated by humans.
An AI tutoring system can aim for autonomy: the student sets the pace. It may be able to do competence better than any of us: the difficulty calibrated perfectly, every time (maybe). It might even do a kind of ‘personal’ connection – endlessly patient, always available, never impatient, never tired. What it cannot do is be in the grip of the idea itself, because it is in the grip of nothing. It has all the content and none of the absorption. We talk about the best teachers being on fire and passing it to their classes. I don’t, today, see AI doing that. We need the teachers who can.
And that isn’t a matter of luck, or of hiring people with a gift. Nobody can be in the grip of an idea if they are exhausted, teaching a subject they don’t like, or beign told exactly how to teach it. The same three things apply to teachers. Autonomy over how they teach; competence, which means depth in a subject and the time to keep reading in it; connection to colleagues who care about the same ideas. A school that builds those will have more teachers in that state more often. That’s better for kids and AI may be able to genuinely help with – if it gives time back rather than taking attention away.
Now, I might have used AI tools in that lesson, and perhaps Kim would have had the same reaction – but if so, it would have been less because of the tools, and more because of me, the teacher, and the ideas that she jumped into. Now do not misunderstand me – my thrust here is not anti-AI at all; I am saying that AI brings us back to the craft of human encounter. And the most interesting evidence comes from someone who believes in the tools far more than I do.
A few years ago Sal Khan, one of the most prominent voices in educational technology, set out a genuinely thrilling vision: an AI tutor for every child on Earth. He opened a wonderful academy, initially on YouTube, now his own site, his own global brand. It was not a gimmick – a serious, well-intentioned, well-funded, well-built attempt to give every student the thing we know works, the personal attention of a patient tutor, at global scale. If anyone was going to make the machine supply what a great teacher supplies, it was going to be him. He had the resources, the talent, and every reason to want it to succeed.
But in April this year, in an interview with Chalkbeat, he said something that went against everything he’d hoped for. He said of Khanmigo, his AI tutor, “For a lot of students, it was a non-event… They just didn’t use it much.”
And he compares his AI to a super-smart tutor at the back of the room, waiting for students to seek out help. Would they? His answer – obvious to any teacher – “Some will; most won’t.” In other words, AI tools don’t necessarily make students motivated to learn or even ask questions. And that’s the whole thing, right there. By themselves, they don’t create the wanting.
What to do about that? One answer, roughly, is a better tool – one that’s less patient, that chases the student, that holds them to account rather than waiting to be asked. But my reading is different. My reading is that the tutor didn’t fail because it wasn’t good enough. It failed because the one thing it couldn’t do was the one thing that matters: it could offer help, but it couldn’t make anyone want it. Khan and I differ on some things, but I think he’s right when he said, in that same interview: “AI is going to help. But I think our biggest lever is really investing in the human systems.”
That’s old wisdom about technology, and we should remember it as we decide where to spend our time, our money, our attention.
I’d add, though, that these human systems are the deep, unbuyable kind that manifest in school culture, in the ways we do things, not just what we do. The ways that cultivate motivated, curious students by helping them to feel autonomous, connected and competent. Which is why the thought I’d leave you with is this: The best AI implementation won’t happen in schools with the best AI tool or policy. It will happen in the schools that systematically build the conditions for motivation to be kindled and cherished, as infrastructure, not left to chance.
So let’s persistently ask: What makes a child want to learn, and not merely produce?
Post Script
I should declare an interest: I’ve always argued that the point of school is what happens inside a student’s head and not what is easy to measure or what can be captured in a transcript – so of course I would say this (in this at least, I have been consistent). But I’d add that the things I’ve described, and many I haven’t – the assessment as learning not ranking, the non-publishing of exam grades, the awards that don’t leave most children on the outside, growth measured by child not the cohort; the mixed-ability approach, the explicit teaching of perspectives, service and outdoor education as timetabled not extra; the mentoring, the advisory system so every child is known by an adult, the conceptual approach, restorative approach; how we hire and appraise for the teachers who reach students – are choices we’ve actually made, over years. They have not always been easy; and some were not pleased. Some still aren’t. That’s what I mean by infrastructure: not a single AI policy but the accumulated weight of hundreds of largely invisible decisions that everybody benefits from, even if they don’t always realise it.
And AI now makes it impossible to miss the difference between what’s in kids’ heads and how we measure it. Scary as it is, I welcome that; it’s long overdue.
References
- Prompts from Claude to get graphics from Gemini
- Barnum, Matt (2026) Why Sal Khan is rethinking how AI will change schools, Chalkbeat, 9 April
- Deci, Edward L., Richard Koestner and Richard M. Ryan (1999) A Meta-Analytic Review of Experiments Examining the Effects of Extrinsic Rewards on Intrinsic Motivation, Psychological Bulletin 125(6)
- Deci, Edward L., Richard Koestner and Richard M. Ryan (2001) Extrinsic Rewards and Intrinsic Motivation in Education: Reconsidered Once Again, Review of Educational Research 71(1). The authors’ own short summary of the meta-analysis, written for educators
- Heifetz, Ronald A. (1994) Leadership Without Easy Answers, Harvard University Press
- Pink, Daniel H. (2009) Drive: The Surprising Truth About What Motivates Us, Riverhead
- Ryan, Richard M., and Edward L. Deci (2000) Self-Determination Theory and the Facilitation of Intrinsic Motivation, Social Development, and Well-Being, American Psychologist 55(1). Full text, hosted by the authors
- Willingham, Daniel T. (2021) Why Don’t Students Like School?, Jossey-Bass, 2nd edition
