Will AI replace teachers?

No. AI is becoming good at explaining things, and explaining is a minority of what a teacher does. The rest - managing a room, noticing a child has stopped eating lunch, deciding that this class needs a different approach today, being an adult a fourteen-year-old will listen to - is not close to automated, and some of it may never be.

That is not a comforting answer so much as an accurate one, and the honest version includes the parts that will change.

What the job actually consists of

Ask anyone who has taught, and the explaining is not what fills the day:

• Managing the behaviour and attention of thirty people
• Knowing which child is struggling for reasons unrelated to the subject
• Deciding, live, that the plan is not working and changing it
• Marking, and turning marks into what to do next
• Pastoral work: safeguarding, families, the child whose circumstances just changed
• Motivating people who do not want to be there
• Preparation, administration, meetings, reports

AI can help with marking and preparation. It can do the explaining. On the rest it has nothing to offer, and several of those are the actual job.

What AI genuinely does well in education

Worth being concrete, because vagueness in both directions is the problem with this debate.

Already useful: drafting lesson materials and worked examples; generating differentiated versions of a task; marking written work against a rubric with feedback; explaining a concept to one student on demand at any hour; adapting reading level.

Two of those are substantial workload reductions, and workload is the main reason teachers leave the profession.

Not useful yet: anything requiring knowledge of a specific child over time, judgement about what a class needs, or the authority and relationship that make a room work.

The part worth taking seriously

The strongest argument that something real is happening is not that models are impressive. It is that one-to-one attention is the most powerful known lever in education and schools cannot afford it.

Bloom's 1984 report that one-to-one tutoring with mastery requirements produced about two standard deviations of improvement is quoted too freely and rarely reproduced at that size. But modern tutoring research still finds a pooled effect of about 0.29 standard deviations, which is large. The constraint has never been that we do not know individual attention works. It is that individual attention costs an adult's hour per child.

If some of that can be delivered at near-zero marginal cost, the thing being addressed is the thing schooling has been unable to solve for a century. That is a serious claim, and it is about supplementing teachers rather than replacing them: the teacher's hour becomes more valuable, not less, when the routine re-explanation is handled elsewhere.

The failure mode that should worry people

Not teachers losing jobs. Students using AI to avoid learning.

A tool that answers a question removes the difficulty, and the difficulty was producing the learning. A student who completes every assignment with help and cannot reproduce any of it has done worse than one who struggled and handed in less.

This is not hypothetical and it is not rare. It is also mostly a design problem: a tool that decides what to teach, makes a student explain it back, and refuses to do their work produces a different outcome from one that answers whatever it is asked. The models are similar; the products are not.

What schools should actually do with it

The decisions worth making are narrower than the debate suggests.

Highest value: teacher workload. Drafting materials, differentiating a task five ways, first- pass marking with feedback. These are hours a week per teacher, and workload is the main reason teachers leave. This is also the least controversial use, which is why it is the one being adopted quietly while the public argument is about students cheating.

High value: individual attention the timetable cannot buy. A student who needs a concept explained a fourth way, at 8pm, cannot have a teacher. That is the gap worth filling.

Low value: bans. A prohibition produces unsupervised use rather than no use, and it moves the behaviour somewhere nobody can teach into. The schools getting this right are teaching students how to use it in ways that produce learning, which requires teachers who have used it themselves.

Negative value: replacing instruction wholesale. Nobody serious is proposing this, and it is the version of the argument that gets debated.

What a teacher's job might look like in five years

Less: preparation from scratch, first-pass marking, re-explaining the same concept to the fourth student who missed it, producing differentiated versions by hand.

More: diagnosis, small-group work with the students who need it, the pastoral and relational work that is already most of the job, and deciding what matters - which is the judgement nobody is automating.

That is a change in the composition of the work rather than a reduction in the need for it. Whether it is experienced as relief or as intensification depends almost entirely on whether the time saved is given back or filled.

For a parent reading this

The practical question is not whether AI replaces teachers. It is whether your child is using it in the way that produces learning or the way that prevents it, and those look identical from across the room.

Three questions that tell you which:

1. Can they explain what they handed in? Without the tool open.
2. Do they use it before or after attempting? After is a tutor; before is a ghostwriter.
3. Has anything got easier for them that was hard? Real learning shows up as new capability, not as finished homework.

What to be sceptical of

"AI tutors produce two-sigma gains." Bloom's figure, applied by analogy to software, with the caveats dropped.

"Teachers will be obsolete in five years." Usually from someone who has not taught, and has not accounted for what the job contains.

"AI has no place in education." Usually from someone whose colleagues are quietly using it to halve their marking time.

Any confident claim at all, including an optimistic one. The evidence base for AI teaching is thin. The honest position is that the design can follow what is known about human teaching while the outcomes remain unproven.

Where TruLearn fits

We build an AI teacher, so our interest is obvious, and we try to be precise about the claim. TruLearn is built to do the part of teaching that is deliverable - plan, teach, question, correct, remember - for one student at a time, at a cost that makes frequency possible. It is not built to replace a teacher, and in a homeschool setting it does not replace the parent either: somebody still has to decide what matters and notice when a child is not all right.

We have no outcome data of our own, and we say so rather than borrowing somebody else's. More: AI tutor versus human tutor, and is ChatGPT a good tutor.