Will Kids Learn to Think, or Ask AI?

The research on cognitive development is reasonably consistent about the conditions under which genuine thinking capability is developed: engagement with problems that are slightly beyond current capability, the experience of making errors and diagnosing them, repeated practice that builds automated competence that can then be applied to more complex problems, and the specific kind of consolidation that follows from working through problems rather than being given solutions. These are not the conditions that AI-assisted learning typically provides: AI typically provides the solution first, which shifts the student's task from generating the solution to evaluating or understanding a provided solution, which is a different and generally less cognitively demanding activity.

The generation-before-evaluation sequence matters in ways that are specific to how cognitive capability develops. When students generate their own attempts at solutions, they activate the prior knowledge and reasoning processes that the new problem requires, creating the connections between existing and new knowledge that consolidate into genuine understanding. When students receive a provided solution, they evaluate it against an understanding they may not yet have, and the evaluation that is most cognitively available is the surface-level evaluation of whether the answer looks right rather than the deep evaluation of whether the reasoning that produces it is sound. The difference between these two evaluation tasks is the difference between developing the reasoning capability and checking the answer.

The specific cognitive capabilities that are most at risk from AI shortcutting are the ones that require extended practice to develop: the ability to sustain attention on a difficult problem over time, the ability to manage the frustration of not knowing and to use that discomfort productively, the ability to generate multiple possible approaches and evaluate them against each other, and the ability to recognise when a solution is right for reasons rather than just recognising that it is the expected answer. Each of these capabilities is developed through repeated experience of the productive struggle that AI shortcuts, and each is foundational for the kind of independent intellectual work that genuinely capable adults do.

The honest case for AI in children's education is real and deserves acknowledgment before the concern about cognitive shortcutting makes it seem like the argument is for eliminating AI from classrooms. There are specific uses of AI in education that genuinely support the development of thinking rather than substituting for it, and identifying them clearly is more useful than either blanket prohibition or uncritical adoption.

The most clearly beneficial use of AI in learning is as a responsive explanatory tool: a resource that can explain concepts in multiple ways, that can respond to the specific confusion a student has rather than providing a standard explanation, and that can provide examples calibrated to a student's level and interests. This use of AI is genuinely different from looking up the answer: it is a form of scaffolded explanation that good teachers provide and that AI can extend to contexts where a good teacher is not available. The student who uses AI to understand a concept they are struggling with, and who then applies that understanding to solve problems themselves, is using AI in a way that supports rather than bypasses their cognitive development.

AI is also genuinely useful as a feedback tool: providing rapid, specific feedback on writing, on problem-solving approaches, and on the reasoning behind answers in ways that traditional educational contexts cannot always provide. The student who receives immediate, specific feedback on why an approach did not work is in a better position to learn from the failure than the student who gets a wrong answer marked without explanation. The caveat is that the feedback loop that produces genuine learning is the one that includes genuine effort before the feedback: feedback on an AI-generated answer does not produce the same learning as feedback on the student's own attempt.

The response to the challenge of AI in children's education that is most likely to produce genuinely thoughtful adults requires schools to develop a level of pedagogical sophistication about the role of AI in learning that most current educational systems do not yet have. This sophistication involves being specific about which cognitive capabilities are being developed in which activities, and about which AI uses support and which undermine that development. It involves designing assessments that are genuinely resistant to AI-assisted answers, not as an anti-AI measure but as a way of preserving the contexts in which students develop the capabilities that assessments are supposed to develop. And it involves explicit education about AI, not just how to use it but what it does and does not do, so that students can make informed choices about when to use AI assistance and when to do the work themselves.

The most important pedagogical shift is likely a shift away from content acquisition as the primary goal of education and toward capability development, explicitly. If the goal of education is to produce students who have accumulated specific knowledge, AI has genuinely disrupted the value of that goal because AI can provide that knowledge on demand. If the goal is to produce students who can think, reason, create, and engage with genuinely difficult problems, AI has not disrupted that goal but has created a specific challenge: ensuring that the capabilities are genuinely developed rather than being substituted for by the AI that will eventually assist with their exercise.

The generation of children who grows up with AI as a native part of their educational environment will figure out some of these challenges themselves, in the same way that previous generations figured out how to use calculators without losing the mathematical intuition that calculators seem to threaten. But that figure-it-out process works better when the adults around them have thought carefully about what they need to develop that AI cannot develop for them. The answer is not to keep AI out of schools. It is to be deliberate about keeping genuine cognitive challenge in them.