A Misplaced Anxiety
Much of the current debate about artificial intelligence in education begins in the wrong place.
We worry that students may use AI systems to complete tasks once treated as evidence of learning: essays, problem sets, summaries, and explanations. We ask how to prevent misuse, preserve integrity, and defend established forms of assessment.
These concerns are real. But they do not reach the deeper issue.
Artificial intelligence did not create the gap between performance and understanding. It exposed it.
What appears as technological disruption may reveal something older and more structural: a weakness in how education has often equated visible output with genuine formation.
The Proxy Assumption
For long periods of educational history, producing the answer usually required substantial intellectual effort. There was no easy alternative source. Under those conditions, visible performance often functioned as a reasonable proxy for underlying understanding. But a proxy is not an identity.
To produce a correct result is not necessarily to possess the capacities that the result appears to display. Correct outputs may arise from memorisation, imitation, procedural training, narrow rehearsal, or temporary pattern recognition. They may indicate learning, but they do not guarantee it.
What artificial intelligence has changed is not the truth of this distinction, but our ability to ignore it.
Tasks once sustained by the practical difficulty of imitation can now be completed fluently, instantly, and at scale. The proxy assumption, long tolerated, has become unstable.
Artificial Intelligence as a Diagnostic
Artificial intelligence functions less as an enemy than as a diagnostic. It reveals what our tasks were demanding of learners.
When a generative system can complete an assignment with little apparent loss of value, a serious question arises: what, precisely, was the task cultivating in the learner?
This does not mean the task was worthless. Many traditional exercises built fluency, discipline, familiarity, and background knowledge. But their educational value may not have been the value often attributed to them.
Activities treated as evidence of thought may have depended more heavily on execution than on judgement.
AI now performs execution with increasing competence. In doing so, it forces institutions to distinguish what can be produced from what must be formed.
Result and Formation
The present moment makes one distinction difficult to avoid: the distinction between obtaining a result and being changed by the effort of reaching it.
A student may submit a correct mathematical solution generated by AI. Yet when asked why a particular step was necessary, how the reasoning would change in a new case, or where the method would fail, uncertainty appears. The result exists. The underlying capacity may not.
This does not mean formation occurs only through solitary struggle from first principles.
Students often learn through examples, models, imitation, and guided solutions. A proof carefully studied can teach more than one poorly invented. An elegant solution, when examined attentively, may cultivate greater understanding than blind trial and error.
But in such cases, formation occurs only when the learner reopens the process: reconstructing reasons, testing alternatives, identifying limits, and making the logic their own.
Outputs may support formation. They do not automatically constitute it.
An Older Warning
In the Phaedrus, Plato recounts a warning about writing. Writing, he is told, may give the appearance of wisdom without its reality.
Historically, Plato was too pessimistic. Writing enabled forms of thought impossible without it. Philosophy, science, law, and memory itself were transformed by textual culture.
Yet the warning still contains an enduring insight. A technology may extend intelligence while also loosening the connection between possession and acquisition, between access to knowledge and earned understanding.
Artificial intelligence raises a related question today. It can generate forms associated with learning (explanation, argument, solution, summary) without ensuring that the learner has undergone the intellectual work those forms once implied.
What Education Must Now Clarify
If intelligent systems can generate answers instantly, education faces a structural choice.
It can concentrate primarily on the use of the policing tool. Or it can clarify what education was meant to cultivate all along.
Understanding becomes visible when a learner can reconstruct reasoning in a new situation, recognise where a method fails, and defend conclusions that are genuinely their own.
These capacities belong to judgement. They may be assisted by tools, but they cannot be substituted by them.
Conclusion
Artificial intelligence did not break education. It revealed where education had confused outputs with understanding, performance with formation, and visible success with the formation it was meant to produce.
The central task now is not to preserve familiar assignments merely because they are inherited.
It is to design forms of learning in which result and formation cannot be cleanly separated — where producing the answer already requires becoming capable of understanding it.