For centuries, schools have operated on a relatively simple assumption: the things pupils practise at school will later prove useful in life. The Latin phrase non scholæ, sed vitæ discimus (“we learn not for school, but for life”) captures the ideal neatly enough, even if educational systems have often struggled to live up to it in practice. Yet artificial intelligence is now forcing us to ask an uncomfortable question: what exactly are we preparing young people for?
Recently, Danish radio featured a discussion about AI in the gymnasium. Much of it was sensible enough on its own terms. Teachers and educational experts discussed oral examinations, shorter written examinations conducted without computers, and other methods intended to ensure that pupils still genuinely understand what they are doing rather than merely outsourcing the entire process to ChatGPT. While listening, however, I could not help feeling that part of the debate already belonged to a disappearing world. For decades, schools have trained pupils to produce essays, reports and analyses under controlled conditions because coherent written production was itself a difficult and economically valuable skill. But large language models can now generate respectable essays, summaries and reports in seconds, and within a few years they will almost certainly do so even better. One therefore has to ask what exactly is being tested when a pupil is required to write a long essay without AI assistance. In some respects, it resembles testing navigation by confiscating GPS devices, or insisting that accountants calculate square roots by hand. Such exercises may still possess some pedagogical value, but it becomes increasingly difficult to pretend that they directly reflect how most intellectual work will actually be carried out.
At the same time, the obvious counterargument should not be dismissed too lightly. The fact that machines can perform a task does not necessarily imply that humans no longer need to understand it. Nobody seriously argues that architects should stop learning geometry because CAD software exists, or that musicians should abandon music theory because synthesisers can generate harmonies automatically. Writing still trains thought. Organising an argument still matters. Linguistic precision still matters.
The deeper problem is not that schools continue teaching traditional skills, but that educational systems historically evolve slowly while technological revolutions arrive abruptly. Schools are extraordinarily good at preparing pupils for the world of twenty years ago. The industrial revolution gradually reduced the importance of physical labour while increasing the importance of clerical and technical skills. The AI revolution may in turn reduce the importance of routine intellectual labour, yet much of contemporary education still implicitly assumes that pupils are primarily being trained to become human information processors. Increasingly, however, that is not where human value may lie.
A few years ago, I had a conversation about what meaningful examinations in the age of AI might look like. At the time, a particularly good example seemed to be something like the following: “Research the worldwide decline of honey bees and write a fifty-page report with sources. You have four hours.” Back then, large language models possessed relatively small context windows and far weaker planning abilities than they do today. A pupil would still have needed to structure the project manually, divide the task into subproblems, manage sources, verify claims and stitch the whole thing together into something coherent. The AI could assist with fragments, but the orchestration remained fundamentally human. Today, however, frontier models are increasingly capable of carrying out much of the process themselves in a single interaction. The old test has therefore become obsolete with astonishing speed. Yet this does not imply that education itself becomes obsolete. Rather, it suggests that the valuable human skills move upwards into a different cognitive layer. The future may belong less to those who can mechanically produce information and more to those who can evaluate, direct, interrogate and contextualise it. A mediocre pupil can already generate polished-looking prose using AI. What becomes scarce is no longer fluent text, but judgement.
This changes the nature of intellectual competence itself. The crucial question is no longer simply whether a pupil can produce an answer, but whether they can recognise the difference between a deep answer and a shallow one, between genuine understanding and plausible-sounding sludge. Can they detect fabricated sources? Can they identify ideological framing hidden beneath apparently neutral prose? Can they compare multiple AI-generated analyses critically rather than accepting the first one presented? Can they notice which important questions were never asked at all? Can they steer an AI system towards originality rather than merely eliciting the safest consensus response statistically available on the internet? Perhaps most importantly, can they explain and defend their reasoning orally under pressure? These are not trivial skills. In some respects, they may prove substantially more intellectually demanding than the traditional essay-writing exercises they are gradually replacing.
One possible future examination format would therefore involve radical transparency rather than prohibition. Pupils might be granted full access to AI systems, while the examiner evaluates the entire interaction history rather than merely the final product. The key evidence would no longer be the essay itself, but the intellectual process revealed through the dialogue. What questions did the pupil ask? Which assumptions did they challenge? Did they test counterarguments? Did they verify sources? Did they recognise contradictions or gaps in the AI’s reasoning? Did they understand when the machine was bluffing? The distinction between weak and strong pupils may increasingly lie not in whether they use AI, but in how intelligently they use it.
Similarly, future education may place far greater emphasis on epistemic filtering and synthesis. Pupils could, for example, be presented with ten texts on the same topic: several written by humans, several by AI, some subtly biased, some factually unreliable, and perhaps one technically correct yet profoundly misleading. Their task would not necessarily be to produce another essay, but to assess the reliability, usefulness and hidden assumptions of each text, possibly through oral examination. Such exercises would reflect a world in which machine-generated information becomes effectively infinite and the bottleneck shifts from information production to information evaluation.
Under such conditions, speed-reading and speed-listening may themselves become crucial intellectual skills, since educated citizens, journalists, scientists and civil servants may increasingly need to process enormous volumes of machine-generated material simply in order to identify the small fraction genuinely deserving sustained human attention.
Yet there is also a danger here. If AI systems become increasingly capable of summarising, filtering and generating information, humans may gradually become excellent skimmers but poor thinkers. Rapid probabilistic assessment is valuable, but deep reading, sustained concentration and intellectual patience remain essential for genuine understanding. The challenge for education therefore lies not merely in adapting to AI, but in deciding which human capacities remain worth preserving even when machines can imitate them convincingly.
Ironically, this technological future may bring parts of education back towards older and more human forms. Oral examinations, debates, seminars and live intellectual performance become more valuable precisely because they test integration, flexibility and understanding rather than mere production. The future classroom may in some respects resemble a medieval disputation more than a twentieth-century exam hall filled with silent pupils producing handwritten essays under supervision. Schools, after all, have never existed merely to train workers. They also exist to shape citizens, cultures and minds.
The inscription above the former headmasters at my own old gymnasium, Horsens Statsskole, expressed the ideal succinctly enough: ut discat juventus sapere et fari — “that the young may learn wisdom and eloquence”. The world may no longer require millions of humans capable of producing competent five-page essays on demand, but it may still desperately require humans capable of recognising truth, resisting manipulation, sustaining attention, exercising judgement and asking worthwhile questions. Those are much harder capacities to automate. At least for now. Humanity does, however, possess a remarkable tendency to build machines that gradually encroach upon its own definitions of uniqueness, only to react with collective astonishment when this unexpectedly alters civilisation itself.