A child sits down at the kitchen table with an essay due the following morning. The subject is the causes of the First World War. She could read a chapter, make notes, decide which causes mattered most, struggle with the structure, write a poor first paragraph and gradually work out what she thinks. Or she could ask an AI system to produce 1,200 words at the appropriate level and be upstairs ten minutes later.
The first instinct of many schools has been to make AI harder to use. Phones are banned, more work is supervised, detection software is bought, coursework is redesigned and, in some cases, students are sent back to pen and paper. Some of this is sensible, particularly while schools work out what AI means for assessment. But it risks becoming an elaborate effort to recreate the conditions under which the old system worked, rather than asking a more uncomfortable question: does homework still provide the evidence of learning that we once assumed it did?
Until recently, an essay seemed to offer reasonably reliable evidence of learning. It was the visible result of an invisible process involving reading, confusion, selection, argument, doubt and revision. If a student produced a decent essay, it was reasonable to assume that some of that intellectual work had taken place. AI has made the assumption much less secure. It can produce the finished piece without requiring much of the thinking that once lay behind it.
That is why the debate so quickly becomes one about cheating, but cheating is only part of the problem. Education has always relied on proxies for learning because they are convenient to collect. Essays, worksheets, problem sets and examinations can be marked, compared and graded. Actual understanding is much harder to package. While producing the proxy required the student to do most of the thinking, the distinction hardly mattered. Now it does.
There is another complication. AI is not a temporary classroom nuisance that schools can wait out. The children using it today are likely to enter universities and workplaces where AI is simply part of how work gets done. They will use it to research, analyse, write, code, design, translate, plan and solve problems. In many jobs, using these systems well will not be an optional extra but part of being competent at the job. An education system that responds by keeping AI away from pupils may therefore succeed in protecting an assessment while leaving them less prepared for the world being assessed for.
That does not mean giving children an AI assistant for every task. Writing remains one of the ways we discover what we think, and anyone who has tried to explain a complicated idea on paper knows how often a supposedly clear thought collapses when it encounters a sentence. Knowledge matters too. Someone who knows almost nothing about a subject is badly equipped to judge an answer that is fluent, plausible and wrong. As machines become more convincing, that judgement becomes more valuable rather than less.
The issue is therefore not whether students should use AI, because increasingly they will. It is when they should use it, when they should work without it, and whether they understand enough to know the difference. Someone learning arithmetic still needs to understand multiplication before relying on a calculator. In much the same way, a student needs enough command of writing, research or reasoning to recognise when an AI has produced something weak. But once that foundation exists, deliberately denying them the tools they will use after school begins to look rather artificial.
What changes is the assumption that producing an answer proves understanding. A student might be allowed to use AI to prepare an essay and then be asked to defend it. Why is this evidence stronger than that evidence? Which claim are you least certain about? What would a historian who disagreed with you say? Which part of the AI’s argument did you reject? A student who understands the subject will usually cope. One who has merely commissioned a competent answer from a machine will often reveal that fairly quickly.
Examinations become more interesting in this world because they offer something coursework increasingly cannot: some certainty about what a student can do unaided. Put thirty students in a room, remove their phones and hand them a paper and you have a reasonably secure measure of that. There is still value in knowing it. The mistake would be to assume that unaided performance is the only capability worth measuring when almost nobody will work that way once they leave education.
Exams also have distortions of their own. They reward recall under pressure, speed, exam technique and the ability to perform on a particular morning, none of which is quite the same as judgement, creativity, collaboration or working through a difficult problem over several days. A student can be excellent at exams and less good at the kind of thinking adult life requires, while another can understand a subject deeply and perform badly in a silent hall at nine o’clock on a Tuesday.
Assessment may therefore need to become more varied rather than simply more secure. We need to know what a student can do unaided, but also what they can do with AI, with other people, under uncertainty and over time. That could mean closed-book examinations alongside oral defence, collaborative projects, practical work and assignments in which AI use is expected and the student is assessed on the quality of their judgement. Perhaps they should sometimes be required to show how they used AI, what they rejected and what they changed. Knowing how to challenge a machine may become as important as knowing how to operate one.
There is another reason school matters that gets less attention in discussions about educational technology. Children do not go there only to acquire information. They go there to encounter other children. They learn how to join a group, contribute to one and occasionally survive one. They collaborate with people they would not necessarily have chosen as friends, argue, compromise, lead and follow. Somewhere along the way they discover that being right is not always enough if nobody wants to work with you.
Friendship, embarrassment, loyalty, boredom, rivalry, kindness and the occasional spectacular falling-out do not appear neatly on a curriculum, but they shape people nevertheless. So does discovering that values become harder when they encounter other people. Fairness sounds straightforward until two children both believe they have been treated unfairly. Respect is easy in the abstract and rather harder when somebody has just taken your place in the team. Debate involves learning not merely how to argue, but how to listen closely enough to understand why somebody sensible might disagree.
School also gives children repeated experiences of success and disappointment at a scale they can usually survive. They get picked and not picked, praised and criticised, win and lose, produce work they are proud of and work they would rather forget. They learn that effort does not guarantee victory and that failure, although unpleasant, is usually temporary. Resilience grows through those experiences rather more effectively than it does from a lesson about resilience.
AI can help with many parts of education, but it cannot reproduce this social world. Nor does its arrival make teachers redundant. Information is abundant and explanations are becoming abundant too. What remains scarce is the experienced human being who notices that a pupil has misunderstood something, asks the question that reveals why, or recognises that a child who appears lazy is actually lost, bored or frightened of getting something wrong. Good teachers create standards, model behaviour, resolve conflict and encourage effort. Much of that work has always been difficult to measure.
Perhaps AI is exposing something that was already true. Schools were built partly for a world in which knowledge was difficult to access and intellectual labour was expensive to perform. That world is disappearing quickly. A machine can now draft, summarise, translate, calculate and explain across a remarkable range of subjects. Keeping children away from those capabilities until they leave school would be an odd preparation for adult life, but allowing them to rely on them before they have learned to think for themselves would be equally misguided.
Education now has to do both. It has to develop independent minds and teach those minds how to work intelligently with machines. It has to produce people who can reason without AI when necessary, use it when useful, spot weak evidence and sophisticated nonsense, and still do the much older human things: collaborate, disagree, recover from failure, make friends and become someone other people can trust.
If AI can do the homework, school still has plenty to do. It may simply force us to become more honest about how much of the important work was never really the paper that got turned in after all.
