We Have Known for Decades That Tutoring Works. The Question Was Always Scale.

A famous finding, and a bigger problem underneath it

There is a statistic about one-to-one tutoring that gets repeated a lot in education.

It comes from Benjamin Bloom's famous work on what became known as the "2 sigma problem". The headline version is striking: the average student taught through one-to-one tutoring performed better than 98% of students taught in a conventional classroom.

That is a remarkable claim.

It is also one worth handling carefully.

The original finding became famous partly because the effect size was so dramatic, and later evidence tends to suggest a smaller average effect than that headline figure. But I do not think that weakens the main point. If anything, the broader evidence still points in the same direction: personalised tutoring is one of the strongest educational interventions we know.

That matters.

Because the real significance of Bloom's finding was never just the number. It was the problem underneath it.

If personalised tutoring is so powerful, why is it not normal?

The answer, obviously, is not that schools failed to notice. Teachers have always known that more individual attention tends to help pupils learn better. The problem has always been that one-to-one support is expensive, time-intensive, and very hard to provide consistently at scale.

That is why I think AI matters much more than most of the public debate suggests.

Not because it replaces teachers.

Not because it fully replicates a human tutor.

But because it moves schools in that direction.

The debate is often framed badly

A lot of discussion about AI in schools still seems to fall into two unhelpful camps.

In one version, AI is treated as a threat: a machine coming for the teacher's job, flattening judgement, automating relationships, and turning education into a cold transaction.

In the other, it is treated as a productivity gadget: a quick way to write a worksheet, plan a lesson, or save a bit of time on admin.

Neither framing gets to the heart of it.

The more interesting possibility is that AI allows more pupils to receive something closer to the kind of responsive, individualised support that we have long known is powerful, but have rarely been able to provide at scale.

That is a much bigger idea than simply saving time.

And it is a much more realistic one than replacing teachers.

What makes tutoring so powerful?

When people hear "one-to-one tutoring", they often think simply in terms of extra time.

But the real power of tutoring is not just more minutes with an adult. It is the quality of attention.

A good tutor can notice misunderstanding quickly. They can rephrase an explanation on the spot. They can ask a follow-up question when a pupil gives a weak answer. They can slow down, speed up, go again, or take a different route. They can respond to the actual pupil in front of them rather than to an imagined average pupil in a class of thirty.

That kind of responsiveness matters because learning is full of small moments where pupils drift off course. A student half-understands a concept. They use a key term slightly wrongly. They make a link that sounds plausible but is not quite secure. In a normal classroom, some of those moments get picked up and some do not. That is not a criticism of teachers. It is simply the reality of teaching groups.

One-to-one support changes that. It increases the chances that confusion gets spotted early, that feedback arrives at the right moment, and that pupils are asked to think again before a mistake hardens into habit.

This is why the tutoring effect has always been so important. It is not magic. It is what happens when teaching becomes more responsive, more personal, and more immediate.

And yet schools have never been able to deliver it at scale. In a class of thirty, with five periods in a day, with books to mark, lessons to plan, emails to answer, and pastoral fires burning in the background, there are hard limits. Even excellent teachers cannot provide continuous one-to-one support to every pupil.

That is why Bloom's work still matters. It did not simply show that tutoring was effective. It exposed how large the gap could be between what works best educationally and what schools can realistically deliver.

This is where AI becomes interesting

AI is not a human tutor.

It does not know the pupil in the way a teacher does. It does not understand family context, classroom dynamics, motivation, mood, or all the subtle human signals that shape learning. It cannot replace professional judgement. It cannot replace relationships. It cannot replace the authority and care of a good teacher.

But that is not the test it has to pass.

The more useful question is this:

Can AI help schools deliver more of the features that make tutoring effective, even in large classes?

In many cases, I think the answer is yes.

AI can give immediate feedback rather than feedback next week. It can offer an alternative explanation when a student is stuck. It can generate extra practice at the right level. It can respond to a partial answer and push the pupil a step further. It can help a student rehearse, revise, check, retry, and refine.

It also has a practical advantage that is easy to miss. It can keep going. It can re-explain, rephrase, quiz, prompt, and respond as many times as needed without the fatigue that shapes real school life. A teacher may explain something brilliantly, but they cannot explain the same point fifteen times in a day to fifteen different pupils without cost. AI can absorb some of that repeat support.

None of that makes it a teacher.

But it does make it useful.

We do not need AI to reproduce perfect one-to-one tutoring in order for it to be valuable. We need it to help us close some of the distance between whole-class teaching and individualised support.

That is a serious educational gain.

In other words, AI helps teachers extend their reach

This, to me, is the most important framing.

The value of AI in schools is not that it removes the teacher from the process.

It is that it helps the teacher extend the reach of personalised support.

That might mean helping one student get instant feedback while the teacher is with another. It might mean allowing a class to receive more precise feedback than the teacher could physically write for every pupil every time. It might mean giving students more chances to practise with guidance in between lessons. It might mean surfacing misconceptions earlier. It might mean letting a teacher spend their energy where human judgement matters most because some of the first-response work has been supported elsewhere.

That is augmentation, not replacement.

And augmentation is a much more educationally serious idea than automation.

The teacher still sets the standard, shapes the task, interprets the output, notices the individual, and decides what matters. But the teacher is no longer doing every first response entirely alone.

That changes the economics of individual attention.

I have seen a version of this in my own classroom

I have found this in my own teaching as well.

In large A Level classes, when we are working through difficult topics, I have used Teach Edge to give students a form of one-to-one Socratic questioning that I simply could not sustain personally with every student in the room at once. It allows them to be probed, challenged, and nudged forward individually while I am still teaching the wider class.

What has struck me is that students respond well to it. They like having to think their way through something rather than just being given the answer. They like being pushed with follow-up questions. They like the fact that the questioning does not stop at the first vague response.

That does not make the tool the teacher.

But it does show something important. AI can help bring some of the benefits of individualised support into classrooms built around large class sizes.

There is another benefit too, and in some ways it matters just as much. This kind of questioning does not only help the individual student. It can also help me teach the class better.

In a large room, misunderstandings often sit in pockets. A few students answer confidently, the lesson appears to be moving well, and yet a meaningful part of the class may still be confused. One-to-one questioning helps draw some of that out. It reveals misconceptions that might otherwise stay hidden. It gives me a clearer picture of where the class actually is, not just where the most vocal or secure students are.

So the value is not only personalised support for the pupil.

It is also better visibility for the teacher.

That, to me, is a very important part of the story.

This matters in ordinary schools, not ideal ones

One reason AI sometimes gets discussed in such abstract terms is that people imagine either a perfect future or a terrible one.

But most schools are neither utopias nor dystopias. They are ordinary places, doing their best under constraint.

Teachers are trying to give pupils more feedback than time really allows. Students need more practice than the timetable can provide. Class sizes are what they are. Budgets are finite. One-to-one human tutoring for every child is not about to become standard.

So the real comparison is not between AI and some ideal world where every pupil has a brilliant personal tutor on demand.

The real comparison is between AI-supported schooling and the actual conditions most schools currently face.

That is where the case becomes much stronger.

If AI allows a pupil to get better feedback, faster correction, more tailored explanation, or more opportunities to improve than they would otherwise have had, then that is not a trivial gain. It is movement toward one of the strongest models of learning support we know.

Not perfectly.

But meaningfully.

The mistake is to think only in extremes

A lot of resistance to AI in education comes from an understandable fear that people are making inflated claims.

That fear is fair. Some claims are inflated.

AI is not about to become a flawless personal tutor. It can be wrong. It can misunderstand. It can overstate. It can give feedback that sounds plausible but is not quite right. Used badly, it can produce shallowness instead of learning.

But none of that makes it educationally trivial.

The mistake is to think the options are either fully human teaching or full machine replacement.

That is not where the interesting ground is.

The interesting ground is in between.

What happens when teachers use AI to increase the amount of individualised response available to pupils, while still retaining professional oversight? What happens when students can get immediate support in the moment they are stuck, rather than waiting until the next lesson? What happens when one-to-one questioning helps expose misconceptions that whole-class interaction misses? What happens when feedback becomes more frequent, more dialogic, and more tailored, even if not perfect?

Those are the right questions.

And they are much more promising than the usual culture-war framing.

A better way to think about the future

If Bloom's work still has something important to say to us, it is not simply that tutoring is powerful.

It is that one of the best things in education has historically been very difficult to scale.

That is why AI deserves serious attention.

Not because it solves education.

Not because it eliminates the need for teachers.

But because it offers a way of bringing more personalised support into systems that have always struggled to provide enough of it.

For decades, schools have known the power of one-to-one help. The issue was never belief. It was feasibility.

AI does not remove that problem entirely. But it may reduce it.

And that is why I think the most interesting case for AI in schools is not replacement.

It is reach.

It is the possibility that, in large classes and busy schools, more pupils can get something a bit closer to the kind of attention that helps people learn best.

Not instead of teachers.

Because of what teachers can now extend.

Gary Roebuck is an A Level Economics teacher and founder of TeachEdge, an AI-powered marking and feedback platform built for UK secondary schools.