In 1984, educational psychologist Benjamin Bloom published a study that should have changed everything about how we teach. He found that students who received one-on-one tutoring performed, on average, two standard deviations better than students taught in conventional classrooms. In plain terms: the average tutored student outperformed roughly 98% of classroom-taught peers on the same material. He called this the 2 Sigma Problem — two sigma being the statistical shorthand for two standard deviations — and the 'problem' was that nobody could figure out how to replicate those results at scale.
The reason the finding is called a 'problem' and not a 'breakthrough' is telling. Bloom wasn't discovering that tutoring is nice to have. He was documenting a yawning gap between what education could achieve and what it actually delivered to most students. The issue was never pedagogical — it was economic. Hiring a personal tutor for every child is, for most of the world, simply not possible. So the insight sat in the literature for forty years, widely cited and largely unacted upon, a gap the system quietly accepted as permanent.
AI changes the arithmetic. A well-designed AI tutor can do several things that make it structurally closer to one-on-one tutoring than to a classroom: it adapts to the individual learner in real time, it never loses patience, it can repeat an explanation in a different way without frustration, and it is available at any hour without additional cost per session. None of this requires the AI to be sentient or to 'understand' the student the way a human teacher does. It requires the AI to be responsive, adaptive, and correct — which is a much lower bar, and one that current systems are beginning to clear.
The real-world stakes are not abstract. UNESCO estimates that 273 million children are currently out of school, and that 44 million additional teachers would be needed globally by 2030 to meet demand. No recruitment drive reaches that number in time. What AI offers is not a replacement for teachers — it is a way of extending educational reach into the gap between what teachers can provide and what students need. Two sigma may not be the right benchmark for AI tutoring, and the research on AI-assisted learning is still maturing. But Bloom's finding set the ceiling for what personalised education can do. The question now is how close we can get to it.
