The part nobody calls teaching

Most of what anyone knows was not learned at a desk. Speech, cooking, driving, a trade, how to argue, how to use the phone in your pocket: the classroom had little to do with any of it. "Average Americans spend less than 5 percent of their life in classrooms," John Falk and Lynn Dierking wrote in 2010, adding that a growing body of evidence shows "that most science is learned outside of school."¹ The arithmetic of the first claim holds. An American public school year in 2011-12 was 179 days of about 6.7 hours, 1,203 hours in all.² Thirteen such years come to about 15,600 hours, roughly 2% of the hours in an 80-year life.

The 95 percent in their title, though, is a share of hours, not of learning.¹ Nobody can count learning in a unit that adds up a language, a trade and quadratic equations, so "most learning happens outside school" is plausible and unmeasured. If it is true, it sharpens the question. Explanation of almost anything is now free and instant. If school was never where most learning happened, and is no longer where explanations come from, what is it for?

An explanation is the cheapest part of learning. The rest is attempting the thing, being corrected, and coming back to it later, often enough that it stays. Outside school, life supplies those for whatever it demands: a job makes you practise the job. What life does not demand, nobody practises. That is the part a school is built to supply: effort on things no one needs today, on a schedule, with someone checking.

Free explanation has already been tried at scale. From 2012 to 2018, MIT and Harvard put courses on edX for anyone; 5.63 million learners registered 12.67 million times.³ In 2017-18, 3.13% of participants completed their courses, while among learners who had paid for a "verified" certificate, 46% did.³ The explanations were the same for both groups. Paying did not improve them; it marked, and bound, people who had already decided to finish.

The pandemic ran the reverse test: schools closed or taught remotely while every explanation stayed online. Bastian Betthäuser, Anders Bach-Mortensen and Per Engzell pooled 42 studies from 15 countries and found an average deficit of 0.14 standard deviations, about 35% of a normal school year's learning, which arose early and persisted.⁴ The deficit was larger in maths than in reading.⁴ One likely reason is that maths is the subject almost nobody practises at home unless someone requires it.

The answer on demand

Generative AI makes explanation free and personal, and the sharpest test so far shows what that does when nothing else changes. In autumn 2023, nearly 1,000 students at a high school in Turkey did four 90-minute maths review sessions, some with a ChatGPT-like assistant built on GPT-4, some with a version told to give hints rather than answers, and some with only their notes and textbook.⁵ With the plain assistant, practice scores rose 48%, and on the exam that followed, taken without it, they were 17% below the control group's.⁵ The most common message students sent the plain assistant was "What is the answer?", and they did not perceive that they had learned less.⁵ The hint-giving version raised practice scores 127% and left exam scores no worse than the control's.⁵

The difference between the two versions was not what they knew but whether they let a student skip the attempt. Put inside a school's frame, a similar tool has gone the other way. In Edo State, Nigeria, a randomised World Bank pilot in 2024 gave students six weeks of after-school sessions with a generative AI tutor, with teachers supporting them.⁶ Participants gained about 0.3 standard deviations, and each additional session attended brought more.⁶ The comparison group was not in the programme at all.⁶ The gain therefore counts the extra supervised time as well as the tool, and cannot separate them. What it does show is that the tool did not undo learning when a teacher, a timetable and an attendance register kept the attempt in place.

What a school is now for

If this is right, cheap explanation does not make school less necessary. It strips school down to the part that was always doing the work: the obligation to try, be corrected and return, on things nobody would choose to practise. That part becomes more valuable as answers get easier, because the easier the answer, the more tempting it is to skip the try. A good school may now be worth most for what it withholds.

It also becomes harder to prove. A diploma has always certified, among other things, that years of required work were done by the person holding it. When the work can be handed to a model at home, that record means less unless the work happens where someone can see it. Some professors seem to have noticed. Roaring Spring Paper Products, which makes the blue exam booklets, says its sales rose in 2023 and 2024, and that customers tell it professors "want students to be forced to use their own work."⁷

Explanation was never the scarce part. The scarce part is a reason to do the work when the answer is one question away, and for most people, most of the time, that reason is somebody else's requirement.