The claim that AI will replace teachers is made almost exclusively by people who have not spent a day in a classroom. It misunderstands what teaching is by such a wide margin that it’s barely worth arguing with.
But there is a real problem in education that this technology is genuinely well suited to, and it isn’t instruction. It’s pace.
The pace problem
A classroom moves at one speed. It has to; there is one teacher and thirty students. That speed is chosen, sensibly, to work for the middle of the distribution.
Which means that on any given topic, a portion of the room has already understood it and is now waiting, and another portion did not understand the prerequisite three weeks ago and is now accumulating confusion. Both groups are quietly learning something unintended: that this subject is not for them.
This is not a criticism of teachers. It is arithmetic. One adult cannot deliver thirty different paces simultaneously, and expecting them to is how you burn out good people.
Why the gap compounds
Mathematics is the clearest example because it’s so strictly sequential. A student who doesn’t properly get fractions will struggle with ratios, then with algebra involving fractions, then with rational functions. Each failure looks like a new failure. It’s the same one, four years later.
By the time it’s visible enough to trigger intervention, the gap is large and the student has usually concluded something permanent about their own ability. That conclusion is far harder to fix than the original misunderstanding.
The intervention that would have worked is small and early: fifteen minutes on the specific thing, at the moment it stopped making sense. What prevents it is not that teachers don’t know this. It’s that identifying which of thirty students has which specific gap, on which specific day, is not humanly achievable alongside actually teaching.
What the technology is actually good at
Three things, none of which are teaching.
- Infinite patience on repetition. A student can ask the same question six different ways without any social cost. That cost is real and it’s the reason a lot of students stop asking.
- Explaining the same idea differently. Not better than a good teacher, but tirelessly. If the visual explanation didn’t land, try the concrete one, then the analogy.
- Locating the actual gap. A student failing at quadratics might have a factorising problem, or an arithmetic problem, or a notation problem. Systematically working backwards to find where it broke is a genuinely useful diagnostic task.
All three are about pace and diagnosis. None of them are about deciding what a child needs, which remains firmly a human judgement.
What it is bad at, and this matters
It cannot tell that a student’s work has fallen off because something is happening at home. It cannot decide that this particular child needs to be pushed today and that one needs to be left alone. It cannot build the relationship that makes a disengaged fourteen-year-old care about the subject, and that relationship is frequently the entire mechanism by which learning happens.
It will also, occasionally, explain something incorrectly with complete confidence. In a domain where the user is by definition unable to evaluate the answer, that is a serious property, and anyone building in this space who isn’t preoccupied with it should be.
The honest position
I’m building in this area, so treat my enthusiasm with appropriate suspicion. Educational technology has a long and embarrassing history of promising transformation and delivering worksheets with better graphics.
What I think is defensible: the pace problem is real, it is arithmetic rather than a failure of effort, and it is the specific problem this technology is shaped to help with. A tool that gives a student a patient second explanation at the moment they lose the thread, and tells the teacher which students lost it, is a useful thing. It is also a much smaller claim than “revolutionising education”, and I’d rather make the smaller claim and be right.
The teacher is not the bottleneck being removed here. The bottleneck is that there is one of them and thirty of everyone else.