Are we teaching students to think or to match patterns?

For a long time, coursework has been used as a proxy for learning. Write an essay, answer a question, complete an assignment. The assumption is that the output reflects understanding.

That assumption is starting to break down.

AI tools can now generate essays that are coherent, structured, and often difficult to distinguish from student work. Not perfect, but good enough to pass in many cases. That has led to familiar reactions: concerns about cheating, attempts to detect AI-generated work, and new rules around what is and is not allowed.

But the deeper issue sits elsewhere.

Coursework has never just been about knowledge. It rewards structure, clarity, and the ability to present an argument in a recognisable way. In other words, it rewards patterns.

That worked when producing those patterns required effort. Now that they can be generated, the signal becomes harder to read. Is the student demonstrating understanding, or producing something that looks like it?

The same question is now appearing on the other side as well. Some schools are experimenting with AI to mark essays in subjects like English and history. Not multiple choice questions, but subjective work that traditionally relies on interpretation and judgement.

The appeal is obvious. Faster feedback, more consistency, less workload. But it also raises a different question. If a system is trained to recognise strong answers based on past examples, what is it actually rewarding?

Clarity and structure are easier to identify than originality or risk. A well-formed argument that follows familiar patterns may be easier to score than something more unusual. That does not necessarily make the system wrong. But it does shape what counts as a “good” answer.

When both sides use the same logic

This is where the tension becomes harder to ignore. Students use AI to produce structured responses. Institutions experiment with AI to evaluate them. It starts to look like a system where outputs are generated and assessed using similar logic.

Not quite AI versus human. More like pattern versus pattern.

It is easy to frame this as a problem caused by new technology. But many of the underlying issues were already there. AI has not broken coursework. It has exposed how much of it relies on outputs that follow recognisable forms. In that sense, AI is acting less like a disruption, and more like a stress test.

Some universities are already responding by shifting assessments towards more in-person work, oral exams, and formats that are harder to outsource. Not necessarily because those methods are new, but because they rely on something different: not just producing an answer, but showing how it was formed.

What education is actually measuring

This is not just about preventing misuse. It is about what education is trying to measure in the first place.

If coursework can now be generated and assessed through recognisable patterns, then the question becomes whether those outputs were ever measuring understanding as clearly as we assumed.

The signal starts to shift from thinking to matching. From demonstrating knowledge to producing something that resembles it.

And that raises a broader question: are we teaching students to think, or to produce something that looks like they have?