The Fire Still Needs a Master

There’s a school in Texas selling a viral promise: “two hours a day, no teachers”. Look closely, though, and it makes the case for the human subject teacher better than I ever could.

In my last post there was a throwaway line about my Grade 11 Physics students: some are hungry to dive into quantum mechanics, while others mostly want to know whether AI is going to write their lab reports for them one day. It’s the second group I’ve been thinking about ever since, because behind the cheek sits a serious question: if a machine can handle the maths, the physics, and the lab reports, then what is school still for – and what’s left for the humans who work there? A school in Texas has now offered what it thinks is the answer. It has staked its whole model on machines handling the academics, and it says it doesn’t need teachers at all.

It’s called Alpha School, and the pitch is genuinely seductive. Students do their core academics – maths, science, English, the lot – through adaptive AI software for about two hours a day, and spend the rest of the day on life skills, sport and passion projects. The founder, MacKenzie Price, says her students learn roughly twice as fast as kids in a traditional classroom. Two hours. No teachers. Double the speed. If you’re a parent watching your teenager trudge through a six-hour school day to absorb what looks like half a textbook, that isn’t a threat. It’s the dream.

And I want to take that dream seriously, because anyone who read that post knows I’m no Luddite. It was a frank audit of six months of pushing generative AI into the heart of my actual teaching, and the verdict was that the upgrade is real – genuinely, measurably real – even when it’s also maddening. So I’m not here to wave a walking stick at the cloud. I’m here because I think the so-called teacherless school, of all things, makes the case for why students still need a human in the room – and a very particular kind of human at that.

What the brochure doesn’t say

Look closely, and Alpha hasn’t actually done away with the teacher. It’s taken them apart.

Start with the software, which is genuinely good at one thing: adaptive apps in the Khan Academy mould (plus a few of Alpha’s own) that work out what a student knows and serve up the next right problem. I won’t pretend that’s nothing – it’s real, and it’s useful. It can even flag when a student is struggling. What it cannot do is care that they are. Which is exactly why Alpha pays human beings.

It calls them guides, and their job, in the school’s own words, is to keep students motivated and on track. They aren’t physics teachers or biologists or economists; they don’t teach content. They build relationships, hold attention, and keep students wanting to carry on – full-time, in person. And when a student gets properly stuck on the content itself, the system books them a coaching call with a remote subject teacher (a struggling student can request one too). So the subject expert didn’t vanish either. Alpha just made them remote – summoned when needed rather than standing at the whiteboard.

Add it up, and the “teacherless school” runs on a quiet sleight of hand: take the ordinary teacher, split them into a motivator in the room and an expert on a screen, automate the sequencing in between, and call the result “no teachers” – their phrase, not mine. Even Alpha’s own FAQ gives the game away – asked whether great educators are now obsolete, it answers that they’re “more important than ever.” And MacKenzie Price herself told CBS that “teachers are not going to be replaced” – that they’re the most important part, she says, of making the model work.

One caveat about the headline numbers (learning twice as fast as their conventionally schooled peers, and testing in the top few percent in the country). Alpha measures with respectable, nationally normed tests, and I won’t pretend otherwise. But before crediting the model, look at who’s in the sample – because nobody ends up at Alpha by default. Just about every student there has parents who went hunting for an AI-first start-up school, believed the pitch, and paid fees in the tens of thousands of dollars a year to act on it. That’s not a cross-section of teenagers; it’s a very particular breed of family – sold on the mission, fluent in tech, and raising exactly the kind of curious, backed-to-the-hilt kid who would post impressive numbers at almost any school (wealth alone isn’t the filter, by the way: after having taught at several fee-paying international schools, I can report that motivation to learn hard things is not something money reliably buys). So the results tell you what already-motivated students do when you hand them brilliant tools and step back. What the model does for typical teenagers – the ordinary mix that fills most conventional classrooms – is still an open question. And they are the students this whole argument is about: a model that only works for kids who arrive motivated hasn’t solved teaching; it has streamlined school for those who least needed the help.

Flint and steel

Before I make the case for the human, I owe the machine its due – because if I pretend the AI is useless, nothing else in this post deserves your trust.

Some days, teaching feels like trying to light a fire in the rain – hunched over two damp sticks, rubbing away, coaxing a spark that mostly refuses to come. Some days it catches. Most days it doesn’t. But you keep at it, because the alternative is everyone stays cold.

Then, in the last two years, the tools really did get better. It’s like the two damp sticks have been upgraded to flint and steel – in other words, generative AI that can now reliably get the subject content right, where two or three years ago it couldn’t solve a Grade 9 probability problem. The rain hasn’t stopped – and I’ll get to why it never does below – but the spark comes far more easily now.

I’ve felt it. As I admitted in my audit of using generative AI in my teaching, having a thinking partner to work through a Lorentz transformation or pressure-test an explanation has measurably improved my teaching resources – even if there are still days when using AI to build them makes me feel like I’m Rambo smashing rocks in the prison quarry. The flint and steel is a genuine gift. The schools that pretend AI changes nothing will be left behind by the ones that pick it up. On that, the Alpha crowd are dead right. But flint and steel never lit anything on its own. Someone still has to kneel down in the rain and strike it – and someone still has to want to. And that part, as Alpha itself concedes, is still a job for humans. The question – the one that is the focus of this post – is which humans.

The motivation gap

Here’s the thing almost nobody selling an AI tutor will say out loud: the cleverest tutor in the world is useless to a student who won’t sit down with it. It’s the point Craig Barton’s AI-in-Education series keeps circling back to – most sharply in his conversation with Barbara Oakley, who argued that an AI tutor’s effectiveness rises or falls on the student’s motivation, and that fostering that motivation is a job for a human, not the tool.

Give a curious, self-starting sixteen-year-old an AI that can explain special relativity six different ways, and you’ve handed a rocket to someone already aimed at the sky. A few of my students are exactly this, and they’re a joy to teach. But they aren’t most students. Most are somewhere in the middle, and a good number decided years ago that they’re “not a maths person” – or a physics person, or any kind of “science person” at all – a belief we manufacture as a culture, as I’ve argued before, until it sets like concrete.

An AI tutor on its own is not going to help those students. Sure, a well-designed one won’t let them simply copy answers down – but it can’t stop them getting frustrated, tuning out, and going through the motions: clicking through attempt after attempt until the app all but hands over the answer, doing just enough to keep their parents and teachers off their back.

Alpha understands this perfectly – it’s the entire reason the guides exist. So let me grant the point in full: motivation is the bottleneck, and keeping a student motivated takes a human. We agree.

The interesting question is the next one. A guide can get a student to sit down, stay focused, and keep doing something that looks like progress. But getting them to want to engage their brain with the hard-to-learn thing in front of them – to earnestly climb up the mountain of the subject itself – is that the same job? Or does it take something more than a generalist motivator, however committed?

Why the rain never stops

To answer that, we first need to understand what the rain is, and why it doesn’t stop.

The psychologist David Geary draws a now-famous distinction between biologically primary knowledge and biologically secondary knowledge. Primary knowledge is the stuff we evolved to absorb effortlessly, just by being around it: speaking your mother tongue, walking, reading a face. Nobody runs an intervention to teach a toddler to walk. Secondary knowledge is everything school is for: reading, algebra, Newton’s laws, balancing a redox equation. It’s culturally recent, evolutionarily unnatural, and – here’s the crucial bit – we are simply not built to pick it up on our own. That’s the rain, and that’s why it never lets up: not a fault in the student, not a fault in the school – just the permanent weather of learning anything hard.

Which is why, as Carl Hendrick puts it, the academic learning we send teenagers to school for is “biologically unnatural”. Natural selection optimised us for the savannah, not the exam hall. You cannot hand a teenager a chatbot and tell them to go off and discover calculus; left to itself, the mind does not wander towards the quadratic formula. And this is why Geary and Hendrick matter here: if secondary knowledge won’t grow on its own, then every student needs two imports from outside their own head. A route: the subject laid out deliberately, rung by rung, by someone (or something) that knows the whole climb. And the will: the drive to keep hauling themselves up while their brain lodges objections. The route, I’ll concede without a fight – the machine now lays those rungs brilliantly. Alpha’s whole model is relentless, structured sequencing – the next right problem, then the next; mastery before you move on – tireless, adaptive, and increasingly better at it than an exhausted human juggling thirty other students. That’s not praise for praise’s sake; it’s narrowing the field. The better the machine gets at laying rungs, the more starkly the other import stands alone: the will. And no ladder, however well-built, keeps the rain off.

This is usually where a parent puts up a hand – at a parent-teacher interview or parent evening, just after one of my rants about why continuing to study maths until senior high school or even first year university is so important – and asks the obvious question: if the machine can already do all of this, why make my kid grind through it at all? The education writer Daisy Christodoulou has the cleanest answer I’ve seen, and it comes in two halves. First, a machine outperforming us has never been a reason to stop learning the thing – we didn’t stop teaching children to draw when the camera arrived, or to run when the car did. And second, doing it is how you learn it: the problems a machine can already solve are the rungs a student climbs to reach the ones it can’t, and you can no more outsource that climb than outsource your own press-ups. The machine can do the maths. It cannot do the learning of the maths – and that has to happen inside a student’s own head, no matter how clever the software gets.

So: the climb can’t be skipped, the terrain fights you the whole way up, and the machine now lays the route better than most humans ever could. The machine covers the first import – the route. The second – the will – is still missing. A generalist guide can get a student to the ladder: on time, focused, hands on the rungs. But hands on the rungs and wanting to climb are two different things – and when the brain itself is resisting the climb, the wanting is everything. Which leaves exactly one question standing: where does the wanting come from? From a generalist cheering at the bottom of the ladder – or from a subject expert who can make the climb alongside the student? 

Why you still need a Jedi master

Here’s my answer. It isn’t that a human plans the route better than the machine – I’ve just conceded that bet, and anyone still making it is going to lose it soon. What I’ll stake everything on is this: a machine has never climbed the rungs itself.

Picture Yoda training Luke on Dagobah – the swamp planet where the small green master drills a doubting student who is quite certain he can’t. What makes Yoda irreplaceable isn’t that he could draw up a well-sequenced Jedi Academy scheme of work – that could be outsourced to a droid. It’s that he has actually lifted the X-wing. He has stood exactly where Luke is standing, felt the precise thing Luke is feeling – I can’t – and come out the other side. So when he says it can be done, the words carry the weight of someone who has done it.

And because he has made the climb himself, Yoda knows something the most devoted generalist cannot: what the next rung looks like from the inside. That’s what lets a master set up the punchline – arrange things so the student makes the discovery themselves, at the exact moment they’re ready to make it. Sagan knew why that moment matters more than most:

“When you make the finding yourself – even if you’re the last person on Earth to see the light – you’ll never forget it.” – Carl Sagan

Again, credit where it’s due: the software can set up a punchline too – sequencing discoveries is precisely what it’s built for. But what it cannot do is be there when the punchline lands. Anyone who has taught knows the moment: the student’s face changes, the pieces lock together, and your own enthusiasm for the subject starts bubbling over – and the student catches it, the way enthusiasm cannot be caught from a progress bar. That contagion between two people in a room – or on a video call, for that matter – is the closest thing teaching has to magic (a term I use loosely – this remains a proudly superstition-free blog). No app can reproduce that.

A student wrestling with special relativity doesn’t just need the next right problem. They need to see, in a real human who has wrestled the same beast and won, that the beast can be beaten – and what it looks like, from the inside, to beat it. That isn’t information. It’s testimony. And testimony is not something you can sequence. When the beast finally goes down – when the thing clicks – that same history pays off a second time: the master doesn’t just watch the student’s joy from the outside; they share in it, because they know exactly what that moment feels like. It’s the difference between being happy for a team that’s just won, and being part of the team that won it. Given an engaged student, a well-designed AI tutor really can produce the understanding, and a caring generalist can stand in their corner all the way through. But only the Jedi Master can convince a student the beast is beatable – and share in the win when it falls. 

Would Batman quit here?

So that’s the first thing only a master can give: the testimony of someone who has been there. Here’s the second, and it’s the one I’d defend as doggedly as Rambo going back through the jungle for Colonel Trautman.

A machine can sequence the struggle. An Alpha guide can cheer you through it. But neither can hand a frightened student the thing they need most: a braver version of themselves to borrow. It sounds like a line from a superhero movie, but it’s actually a measurable effect. In 2017, the psychologist Rachel White and her colleagues ran what’s become a rather famous experiment. They gave young children a dull, repetitive task and measured how long each child persevered. The ones who lasted longest weren’t told to “try harder” – they were invited to become someone else while they worked; to pretend to be a capable, determined character (Batman, in the original study). They called it the Batman Effect, and the mechanism is called self-distancing: step out of the cramped, frightened, I want to quit mindset and into a braver persona, and you reach reserves the first-person self couldn’t.

Longtime readers will know I’ve been running my own unscientific version of this for about a decade – a trick I once wrote about without ever quite being able to explain why it worked. Well, here’s the answer, five years late: it was the Batman Effect all along. The lime-green bandana I used to tighten before my own university exams? That was self-distancing; I just called it “channelling Rambo” at the time. And these days, when I christen my extension classes – the Physics Warriors, the Further Maths Samurai, the Higher Level Jedi – it’s the same effect running at class scale, but with tongue firmly in cheek. The names are an invitation to take on a persona and treat the hard problems as training – so that a student stuck halfway through an IB exam question, ready to give up, finds the question quietly changing from do I understand this? to would a Jedi quit here? I’m handing them a cape, because the caped version keeps going when the Clark Kent version would have thrown in the towel.

But here’s the fine print on the trick, and it’s the whole argument in miniature. When I call my Higher Level maths class the Jedi, it works only because they know I have walked the path they’re on. The anointing means something because of who is doing the anointing. A generalist guide can call a student a Jedi all day long, and it lands like a participation certificate at a sports carnival. Only a Jedi can make a Jedi. Only someone who has lifted the X-wing can look a student in the eye, tell them they are capable of lifting it too, and be believed – because they know exactly how heavy it is, and they can see, from having been there, that this student can do it. The motivation and the mastery were never two separate things. The motivation works because of the mastery.

And that is the piece that Alpha can’t buy off the shelf. Its subject expert is real – but they’re a voice on a booked call, with no promise it’s the same voice twice. I’ve seen the difference from the other side of the webcam: what actually shifts a struggling student in online tutoring isn’t a rota of brilliant strangers, it’s the one mentor they’ve come to trust – the one they book again, and again, because something clicked. Luke didn’t file a support ticket and wait for the next available master – he went straight to Yoda when he needed help, and stayed with him until the X-wing was out of the swamp. The expertise matters enormously – but it does its work through a relationship, and a relationship is the one thing you cannot summon on demand and dismiss when the call ends.

Re-tightening the bandana

So I’m tightening the bandana again, not hanging it up.

The flint and steel is real; I’ll use it, my students will use it, and I’ll keep singing its praises. But the fire is still a flame that has to be lit inside one particular human head, against the grain of biology, through effort that doesn’t feel good – and someone still has to crouch in the rain beside that student, sheltering the spark, believing it’ll catch long after the student has stopped believing it themselves. In other words, someone still has to stand there while they hate it, and refuse to let them off the hook.

That someone is not an app. It isn’t a generalist guide with a motivational playbook, however professional. And it isn’t a brilliant stranger on a scheduled call. It’s a master of the subject who has climbed the thing themselves, who knows this student, and who has decided their climb is a cause worth fighting for.

Sagan called science a candle in the dark – a fragile, human-carried light in a demon-haunted world, and one that goes out by default, because nobody has to work at superstition; the dark is the resting state, and the rain never stops. The candle stays lit only because, in every generation, somebody deliberately passes the flame to the next pair of hands. An app can’t do the passing – it might detect when the flame is guttering, but it doesn’t truly care. A generalist can care, and care matters; but caring about a flame is not the same as understanding how it burns. The passing takes the master, because the master understands the flame – they’ve kept their own burning through years of the same rain, and they know what feeds it, what smothers it, and what it needs at the moment it starts to dim. Not coincidentally, that master is also exactly the kind of adult a teenager will willingly go through the “training from hell” for.

So, the tools got sharper, but the job didn’t change – if anything, the machine hands the job back to us stripped to its essence. Every hour the software buys back from sequencing content and marking quizzes is an hour the master can spend on the thing that really matters: the student in front of them.

Students have never had more powerful help than they do right now. But the easier it becomes to let the machine think for them, the more they need the one person it can’t replace: a master who has climbed the mountain of their subject, who won’t let them take shortcuts, and who can see that they are capable of climbing it too.

Are you on the subject expert’s side of this, or do you think the Alpha model – AI out front, humans as generalist coaches – is where school is heading, and I’m just a teacher defending his own relevance? Have you watched an AI tutor genuinely light a fire under a student who’d given up – or watched one wait politely while they drifted away? And what’s the “would Batman quit here?” trick that works for you or your students? Leave a comment below.

Acknowledgements: this post was co-written with Claude, and unless specified otherwise, images were generated with Gemini Pro – though, as ever, the bandana-tightening was mine.