In an age where AI can answer almost anything in seconds, education systems still spend most of their energy testing whether students can produce correct answers. That mismatch is becoming harder to ignore. The capacity to arrive at the right answer is no longer the scarcest cognitive resource. The capacity to ask the right question might be.
Answers Have Become Cheap
The economics of knowledge have shifted fundamentally. A generation ago, memorising facts, formulating a response under time pressure, and organising information on paper were genuine skills that distinguished capable students from struggling ones. Those skills required sustained effort to develop and weren’t easily replicated.
AI changes that entirely. ChatGPT, Claude, Gemini, and similar systems can produce accurate, well-structured answers to most standard academic questions within seconds. They can summarise a text, explain a historical event, solve an equation, write an essay, or draft a lab report. The output-oriented tasks that make up the bulk of formal assessment — the ones that reward knowing and retrieving — are precisely the tasks AI does with the least friction.
This creates a structural problem for education that goes beyond academic integrity. When the answer is cheap, what is the point of optimising education around producing answers? And if producing answers is no longer the distinguishing skill, what is?
The Problem with AI-Friendly Questions
A UNESCO article — The Disappearance of the Unclear Question — identifies something more troubling than students submitting AI-written essays. It identifies a subtler shift in what students choose to ask in the first place.
The article, written by educators at Johns Hopkins University and Aalborg University, describes a student who spent a month working through sources on the Mozart Effect — a debunked theory about music and intelligence. She began broadly, hit dead ends, revised her question repeatedly, and eventually produced a nuanced argument. With AI, that process could compress into minutes.
The danger, as the authors argue, isn’t just that students get the answer faster. It’s that they learn to ask different questions. AI is optimised for resolution, for clear answers to well-formed prompts. Students who internalise this get better at producing AI-compatible questions — constrained, unambiguous, output-oriented. And they gradually stop asking the ones that don’t work: the unclear questions, the ones that lead nowhere at first, the ones where the struggle to articulate the problem is itself the intellectual work.
The article’s core observation is precise: we risk conflating better prompts with better thinking. The two are not the same.
Curiosity Comes Before the Question
There is something prior to the question, and that prior thing is what education most needs to protect. It is curiosity — the experience of not knowing something and caring about that. Curiosity generates questions. Questions generate inquiry. Inquiry generates understanding. AI can accelerate the middle steps. It cannot generate the starting condition.
The OECD’s framework on empowering learners for the AI age places curiosity, alongside creativity and ethical reasoning, at the centre of what AI literacy actually requires. The OECD’s formulation is notable: curiosity isn’t framed as a personality trait some students happen to have. It’s framed as a capacity that education should cultivate and that AI environments can either support or undermine.
The UNESCO authors describe this in terms of cognitive friction — the productive discomfort of not knowing, of following a line of thought that leads nowhere for a while, of revising a question because the first version was too broad or too narrow. That friction isn’t a bug in the learning process. It’s the mechanism by which genuine understanding develops. AI, by design, removes it.
As explored in Attention Is Becoming Education’s Scarcest Resource, the digital environment already undermines the sustained concentration that difficult learning requires. The UNESCO finding adds another layer: the same environment may be reshaping which questions students think are worth asking.
Schools Still Grade the Answer
The structural tension here is real. Most educational assessment systems — examinations, essays, problem sets, coursework — evaluate the quality of student output. They reward accuracy, completeness, and clear expression of knowledge. These are reasonable things to reward when producing that output requires genuine understanding. They are much less meaningful when the output can be generated without understanding.
What assessment doesn’t typically reward is the quality of the initial question, the willingness to sit with uncertainty, or the intellectual stamina to revise a line of inquiry several times before it becomes productive. The student who spends a month narrowing a research topic through struggle and failure develops capacities that don’t easily appear in the final paper. The student who uses AI to reach the same endpoint in an afternoon may not.
The OECD’s report on what teachers should teach in a world of powerful AI explicitly centres inquiry — the practice of formulating questions and pursuing them through complexity — as a core educational goal that AI should support rather than replace. The report argues that AI’s value in education depends precisely on whether it is used to extend inquiry or to shortcut it. That distinction is almost entirely a matter of pedagogical design, not technology.
What AI Cannot Build
The arXiv research on curiosity and metacognition in AI-assisted learning adds a useful nuance. AI environments can, when designed thoughtfully, scaffold curiosity — by surfacing unexpected connections, posing counterexamples, or generating questions the student hasn’t yet thought to ask. But the default use pattern, the path of least resistance, is the shortcut: use AI to arrive at answers faster, with less discomfort and less time spent in productive confusion.
The student who learns to use AI well is not the one who generates the best prompts. It is the one who uses AI output as raw material for deeper questioning — who reads the AI’s answer, identifies what it glosses over, and uses that gap to formulate a harder question. That is a genuinely sophisticated cognitive move. It requires the student to already have some sense of what they don’t know and why it matters. It requires, in other words, curiosity that pre-exists the AI interaction.
As explored in When Systems Decide for Us, What Happens to Human Judgment?, the risk in AI-mediated environments is not that humans lose access to information. It’s that they lose the practice of exercising judgment — the uncomfortable, iterative work of figuring out what matters and what to do about it. Curiosity and question-framing are the entry point to that work.
The Classroom That Rewards the Question
None of this argues for rejecting AI in education. The UNESCO authors are explicit on this: AI can assist, extend, and inspire thinking when used at the right stage for the right purpose. The problem is not AI per se. It is the educational design question of what we are trying to produce.
A classroom that optimises entirely for answer production will be outcompeted by AI almost immediately. A classroom that optimises for question formation, intellectual stamina, and the willingness to sit with uncertainty for long enough that real insight becomes possible — that classroom is building capacities that AI cannot replicate, because they have to be built through exactly the kind of friction that AI is designed to remove.
In a world of cheap answers, the most valuable thing education can do is make students hard to satisfy. Students who can articulate what they don’t yet understand, who know how to turn that gap into a question, and who can follow that question through uncertainty toward something genuinely new — those students aren’t made by any technology. They are made by the slow, uncomfortable process of learning how to think.
Key Sources
- UNESCO – The Disappearance of the Unclear Question
- OECD – Empowering Learners for the Age of AI
- OECD – What Should Teachers Teach and Students Learn in a Future of Powerful AI?
- arXiv – Curiosity and Metacognition: Towards a Unified Framework for Learning and Education in the Age of AI
Subscribe to EuroLuminant for independent European journalism.
