Teaching with AI: When the Answer Becomes the Lesson

Artificial intelligence can explain almost anything. But the ability to explain something is not necessarily the ability to teach it.

A medical trainee presents a patient with fever, hypotension and an elevated lactate. The diagnosis seems obvious. The trainee proposes sepsis, recommends antibiotics and intravenous fluids, and begins outlining a management plan.

The supervising clinician asks a different question.

What else could explain the hypotension?

The trainee hesitates.

The question is not intended to demonstrate superior knowledge. Nor is the supervisor necessarily suggesting that the initial assessment is wrong. The purpose is to expose an assumption the trainee has not yet recognised making.

This is one of the less appreciated elements of clinical teaching. A good educator does not simply know the answer. They recognise what the learner needs to confront before the answer becomes meaningful.

Artificial intelligence is becoming remarkably capable of explaining medicine. It can simplify complex physiology, construct differential diagnoses, generate clinical scenarios and adapt its responses to the apparent level of the learner.

These capabilities could transform medical education, particularly where access to experienced teachers is limited. A student struggling with an unfamiliar concept no longer needs to wait for the next tutorial or hope that a busy clinician has time to explain it.

That is a considerable achievement.

But explanation and teaching are not interchangeable.

An explanation usually begins with a question and attempts to resolve it. Teaching sometimes involves recognising that the question itself is inadequate.

The trainee assessing the hypotensive patient may not need a more detailed explanation of sepsis. They may need to recognise how quickly they committed to a diagnosis, which competing possibilities they overlooked, or why the available evidence does not yet justify their certainty.

An AI tutor capable of recognising those needs could be enormously valuable. One that simply generates increasingly sophisticated answers might reinforce the very habits an educator is trying to challenge.

Research into AI-assisted medical education is encouraging. A 2026 systematic review of randomised trials reported improvements in measured knowledge and clinical skills, although the studies varied considerably and longer-term effects remain uncertain.

But better performance after an educational intervention does not, by itself, tell us what made the teaching effective.

Was it the clarity of the explanation? The opportunity to practise? The immediacy of feedback? Or the learner's need to retrieve and apply knowledge before receiving assistance?

These distinctions matter when we decide how educational AI should behave.

A novice learning the fundamentals of shock may benefit from direct, structured explanations. A more experienced trainee may need conflicting information, incomplete clinical data and questions that force them to defend their reasoning.

The same technology could support both. But it must recognise that effective assistance changes as expertise develops.

There is also a dimension of teaching that extends beyond cognitive performance.

Clinical educators demonstrate how to respond when a diagnosis is uncertain, how to acknowledge mistakes, when to seek advice and how to communicate decisions that carry consequences for patients. Much of this is learned through watching how another professional behaves under pressure.

AI may provide useful simulations of those encounters. It cannot fully reproduce the professional relationship, shared responsibility and lived consequences through which such lessons acquire their meaning.

This is not a reason to resist educational technology. It is a reason to become more deliberate about what we expect it to achieve.

The best use of AI in medical education may not be to reproduce everything a teacher does. It may be to make explanations and practice more accessible while allowing human educators to concentrate on the judgement, uncertainty and professional development that demand their particular attention.

We should judge these systems not simply by the quality of the answers they provide, but by what learners become capable of doing afterwards.

Sometimes the most valuable contribution a teacher makes is not knowing the answer. It is recognising the question the student has not yet learned to ask.