The future of AI in schools isn’t personalized learning

At first blush, it sounds too good to be true: a learning experience that’s precisely tailored to a child’s needs, strengths, and struggles, speeding up or slowing down as the moment demands, with infinite patience. 

For a decade or more, that’s been the promise fueling the education technology industry—customized learning that fuels rapid progress. Yet, for the most part, it was too good to be true. Not because the ambition was wrong, but because the prevailing vision has had it backwards.

A FAILED PERSONALIZED LEARNING APPROACH?

The vision of AI in education that has drawn the most attention and investment centers on personalized learning. Think Khan Academy’s AI tutor, or models like Alpha School, where teachers are largely replaced by AI-driven platforms. Alpha School is a small but fast-growing network of private schools where students are taught via AI instruction, with adults in the room serving more as coaches and caretakers than teachers.

The idea is intuitive: Every student gets an adaptive, AI-driven experience tailored to their specific needs and strengths. The algorithm meets the child. The teacher steps back. It sounds compelling, until you think about what it actually produces: device-mediated instruction. Students are in front of screens, moving through content at their own pace, supervised by adults whose job is more coordination and childcare than teaching.

We’ve run this experiment already at scale, without choosing to. It was called the COVID pandemic. Years later, we are still recovering from what it did to student achievement, social development, and the relationship between kids and school. That experience, plus a broader cultural reckoning about screens and child development, has given us real evidence about when devices serve kids and when they don’t. The answer is more nuanced than either the techno-optimists or the skeptics would suggest. But one thing is clear: A screen is not a substitute for a teacher.

The problem with personalized learning is that it turns the teacher into a bottleneck, rather than the leverage point. That architecture is backwards.

Personalized learning puts the algorithm in the driver’s seat. But personalized teaching puts the teacher there, with dramatically better data, better tools, and more time to do what only a teacher can do.

THE WRONG KIND OF HELP

The reason kids come to school is fundamentally social. John Hattie’s decades of educational research consistently and unambiguously shows that teacher-student relationships, classroom discussion, collaborative problem-solving, and peer learning produce some of the strongest achievement effects. These aren’t incidental features of school. They are school.

What teachers don’t need is more systems, more passwords, more time in front of dashboards. What they do need is what most have never had: synthesized assessment data that is connected to where the class is right now, and tied to what comes next in the curriculum. Assembling that picture is enormously time-consuming. For most teachers, it simply isn’t doable. Not at the level of precision that students deserve.

This is where AI can change the equation. Not by replacing the teacher at the front of the room, but as the capable colleague handling the preparation work that makes great instruction possible.

Think of what teachers who have access to a skilled teaching assistant are able to do. They delegate the time-intensive back-office instructional work so they can invest their energy where it matters most: the human work of teaching. 

Most teachers don’t have that luxury on a daily basis. AI can change that—if it’s built around the right problem.

THE DATA PROBLEM NOBODY SOLVED

Here’s what makes personalized teaching hard, and what makes most tools inadequate to it: You need data. Not end-of-year assessment data. Not a snapshot from September. You need to know where each student is right now, in the days and weeks before a specific lesson, connected to the specific skills that lesson depends on. That only works if the tools understand how knowledge builds across a curriculum sequence, not just within a single lesson.

That’s a much harder data problem than it sounds. Formative assessment—the ongoing evidence a teacher gathers during instruction—is the heartbeat of good teaching. But in most schools, that evidence lives in disconnected systems, if it’s captured at all. Tools that can generate a lesson plan can’t tell a teacher whether her class is ready for it, because they don’t have access to that ongoing picture of student understanding. Knowing your kids isn’t just about having the data. It’s about being able to make sense of it—across every student, every skill, months of history. All at the speed good teaching actually requires.

This is the problem HMH’s latest technology is finally beginning to solve. Rather than sitting between students and their teachers, our dynamic learning models work behind the scenes connecting data from tests and each student’s learning history to a map of how knowledge builds over time. Before a lesson on fraction-addition, a teacher using our platform can see whether his class has demonstrated mastery of equivalent fractions—and critically, he sees the reasoning, not just a recommendation. A transparent rationale he can interrogate, trust, and act on. Not a black box.

That’s something the field has not been able to do, until now.

THE VERSION THAT ACTUALLY WORKS

As the technology advances, teachers with access to this kind of intelligence will walk into their classrooms differently. AI will surface exactly where their students are, explain it, and hand the teacher a concrete recommendation—before the bell rings. The teacher will still make the call. But she’ll make it with the level of situational awareness that used to take years to develop. And yet it was still never complete.

Teachers with this kind of intelligence aren’t just better informed. They are freed up to be better teachers.

The vision of AI as a tireless, personalized tutor for every student is a tempting goal, but it mistakes efficiency for education. It optimizes the transmission of content while leaving aside the things that actually make learning stick, like a teacher who notices when something is off, who adjusts mid-lesson, who makes a student feel genuinely seen. Those things don’t happen on a screen. They happen in a classroom, between people. Because the classroom isn’t the inefficiency we’re trying to optimize away. It’s the point. 

Jack Lynch is the CEO of HMH.