The wrong question
When generative AI arrived in classrooms, the dominant institutional response was a management question: how do we stop students from using it to cheat?
It was the wrong question. And the fact that it was so universal says something important about the state of pedagogical thinking in our institutions.
The cheat reveals the contract. When a student uses ChatGPT to pass an assessment, they are not committing sophisticated fraud. They are discovering that the contract we offered them — study this, reproduce it here, get this grade — can be fulfilled without learning anything. That is not a technology problem. It is a crisis of meaning.
What AI has made visible
Educational assessment has spent decades measuring what is easy to measure, not what matters to measure. Information retrieval, content synthesis, structured writing, solving problems of a known type. All of it reproducible. All of it delegable — first to textbooks, then to Google, now to language models.
AI did not create this problem. It made it impossible to postpone.
There is something profoundly revealing about the fact that the very tool alarming teaching teams is simultaneously the one exposing which assessments had genuine meaning and which did not. If an assessment is indistinguishable from a task that AI can resolve in two minutes, the question is not what to do about AI. The question is what we were actually assessing.
Learning cannot be delegated
There is something language models, however powerful, cannot do: learn in place of the student. They can generate text on any subject. They cannot build understanding, develop judgement, or transform experience into knowledge that belongs to the person who lived it.
This seems obvious. And yet our assessment models have spent decades measuring precisely the outputs AI can replicate — not the processes that make learning something change the person who learns it.
Experiential learning — in the deep sense that Kolb, Dewey or Freire gave to that concept — is not an alternative pedagogical method. It is the only honest description of how human beings construct knowledge that transforms how they act in the world. Experience provokes, reflection elaborates, abstraction generalises, experimentation confirms or contradicts. That cycle cannot be substituted or accelerated without breaking it.
An assessment that ignores this cycle is not merely vulnerable to AI. It is, in a more fundamental sense, an assessment that ignores how learning actually works.
The illusion of new instruments
Faced with this diagnosis, the temptation is to change the instruments. Portfolios instead of exams. Oral defences instead of written papers. Real projects instead of invented case studies.
None of these changes is bad. Some are excellent. But if they are adopted as a technical solution — as a tactical response to AI — without revisiting the model of learning that underpins them, they quickly become new ways of fulfilling the same empty contract.
A portfolio can be as shallow as an exam if the student has not had to genuinely reflect on anything. An oral defence can become a more sophisticated memorisation ritual. A real project can produce invisible teamwork with individually irrelevant outcomes.
The problem is not the instruments. The problem is what conception of learning lies behind them.
What it means to assess with meaning
Assessing with meaning in a world with AI requires something more uncomfortable than changing formats: it requires asking what it actually means for someone to have learned something.
Not in the sense that they can reproduce it. In the sense that their way of seeing, doing or deciding in that domain has changed. That they can navigate ambiguous situations with more discernment. That they can justify their decisions with their own arguments, not borrowed ones. That they can detect when a generated response — from AI or any other source — is insufficient, incorrect or irrelevant to the real problem in front of them.
That is not assessed in a two-hour exam. Nor in a well-presented portfolio. It is assessed in conversation, in performance under the unexpected, in the capacity to question, in the willingness to not know and keep thinking.
The institutions doing this well have not found the perfect instrument. They have built assessment cultures where what is valued — explicitly, with shared and debated criteria — is the thinking, not the answer.
The role of AI in a transformed assessment model
There is an interesting paradox here. The same AI that challenges traditional assessment models can, when well integrated, be an extraordinary tool for deep learning.
Not as an oracle that provides answers. As an interlocutor that compels thinking. A student who uses AI to explore the implications of a design decision, to test their hypothesis against perspectives they had not considered, to identify the limits of their own reasoning — that student is using the tool in a way that does not replace learning. It amplifies it.
The difference between the use that destroys learning and the use that strengthens it is not in the tool. It is in whether the student is thinking or delegating the thinking. And that distinction can only be made visible from within a pedagogical relationship where thinking has value, where its absence is noticed, where the educator knows each student's process well enough to recognise whether what they present actually belongs to them.
That is not solved with plagiarism detection. It is solved with pedagogical presence.
What this demands of institutions
Everything above has implications that go far beyond the individual classroom. Assessment is not each teacher's private matter. It is an institutional declaration of values — about what the institution understands as learning, what deserves to be recognised, and what does not.
If an institution says it values critical thinking but assesses with multiple-choice exams, it does not value critical thinking. If it says it develops reflective professionals but gives no protected time in the curriculum for reflection — real time, with genuine accompaniment — it is not developing reflective professionals.
AI has placed a question on the table that institutions can answer in two ways. The first is defensive: how do we protect our assessments from AI? The second is transformative: what model of learning do we want our assessments to sustain, and what would need to change for that to be real?
The first response leads to a technological arms race that institutions are not going to win. The second leads to something worth building, regardless of what the technology does next.
AI has not broken educational assessment. It has revealed that assessment has been broken for a long time. The question is whether we now have the courage to fix what is actually failing.
