The Omnipresent Teacher

For over 40 years, we’ve known that one-to-one tutoring can dramatically change how students learn.

In 1984, Benjamin Bloom showed that students receiving individual tutoring could outperform those in conventional classroom instruction by a striking margin. He called it the “2 Sigma problem”: if we know individual attention works so well, how do we make it possible at classroom scale?

The answer has always run into the same constraint: a teacher’s attention is finite. In a classroom of 25 students, there may be 25 different questions, misconceptions, ideas and directions worth exploring, but a teacher can only respond to so many of them at once. AI Tutors could change that equation by giving educators a way to stretch that finite attention.

Tech to the Help

An AI Tutor is a learning experience designed by a teacher for a particular purpose and built around a clear learning goal. The teacher decides what the tutor should know, what role it should play, how it should respond and where it should draw the line.

It might be a writing coach that works from the class rubric, a debate partner that challenges unsupported claims, or a historical character that answers questions from the perspective of their time.

Most importantly, the tutor doesn't give students answers. It can ask a question, offer a hint, reformulate, increase the challenge, ask for evidence or tell a student to try again. The teacher can see those interactions and use what they reveal to decide what students need next.

Not Everything Needs AI

I’m sceptical of the idea that AI will personalise a child's entire learning journey anytime soon. Learning is too complicated for that. Motivation, confidence, relationships, prior experiences and interests all shape how a learner responds. A teacher who knows a child over months or years understands things an AI Tutor simply cannot.

But personalisation doesn't have to mean personalising everything. Sometimes it means giving 24 students the space to take the same idea in 24 different directions.

A teacher in one of the schools used an AI Tutor to let students explore economics through the lens of Lionel Messi. Instead of taking everyone through the same worksheet, students could follow their own questions and lines of inquiry while the tutor responded to them individually.

All students took the assignment in completely different directions - a level of personalisation nearly impossible to orchestrate manually. While the AI Tutor handled independent exploration, the teacher could focus her attention where her guidance was needed most.

That is a much more realistic promise for AI in education: not a perfectly personalized pathway for every child, but more moments in which students can work at their own pace and get a response that meets them where they are.

When that happens, something else changes: class dynamics.

Redirecting Attention

Consider what happens in a typical classroom when a student asks for help.

Often, the students who get the most teacher attention aren't necessarily the ones who need it most. They may simply be the most confident about raising their hands, walking up to the teacher or asking for another round of feedback. AI Tutors change that dynamic.

A teacher can build a tutor around the same expectations, rubric or strategies they have already taught. Students who are ready to move further can continue the conversation independently. Meanwhile, the teacher can spend more of the lesson with students who need explicit instruction or a closer look at where they've gone off track.

Tangible Thinking Process

There is another limitation of the traditional classroom: teachers often see the finished product, but not the learning that happened along the way. Another teacher in one of the schools saw this with her middle school students. They could produce polished work with AI in seconds, making it increasingly difficult to tell from the final product who had done the thinking.

So she redesigned a three-day project around the process as well as the product. Students used six different AI Tutors, including a Metacognition Coach that stopped them at key points and asked them to consider what they had done, what belonged to them and where AI had taken over.

The result wasn't simply better evidence of who had used AI responsibly. It gave the educator a window into how students were planning, reasoning, getting stuck and making decisions.

AI can give students answers. But a well-designed AI Tutor can also give teachers evidence of the thinking behind those answers. In our research, 85% of teachers said AI Tutors surfaced aspects of student understanding they otherwise would not have seen. Four in five said they helped them meet individual learning needs better than before.

Teacher – A Constant Catalyst

None of this makes the teacher less important. If anything, it makes their judgement more important. The teacher still decides the purpose of the experience, what the tutor should know, when it should challenge or hold back, what it should refuse to do and what to do with what the conversation reveals.

Perhaps AI won't solve Bloom's 2 Sigma problem by giving every child their own tutor. But it can change what one teacher is able to make possible. One teacher. 25 learners. More opportunities for each of them to be challenged and supported.

That is what we set out to understand when we looked at more than 100,000 teacher-built AI Tutors and spoke to educators using them in their classrooms.