AI is becoming more common in uni courses. What can students gain and what might they lose?

8 September 2026

Here’s what bothers me about this debate. Everyone is arguing about whether students learn from an AI tutor as well as from a human one. Wrong question. The interesting question is: what is learning, physically, in the student’s head?

I used to watch this in the lecture hall. A student would copy my derivation off the blackboard, perfectly, and understand nothing. Same symbols, zero physics. Then another student would interrupt me mid-sentence with a dumb question - “wait, why can you just add those?” - and suddenly the whole thing would click for the entire room. The interruption was the education. Not the words.

So when someone says an AI generates “the same words as a teacher but the interaction isn’t the same,” I want to grab them by the collar and say: you’ve almost got it, but you’ve got it backwards. The words were never the point. What’s the mechanism? A good teacher gets confused by your confusion. The student’s struggle feeds back into the explanation, in real time, and the teacher is sometimes wrong and has to fight their way out in front of you. That fight is the lesson. Watching a mind repair itself - that’s how you learn that understanding is something you do, not something you receive.

Now - can a machine do that? Maybe, eventually. I’m not going to pretend I know it can’t; nature doesn’t care about my sentimental attachments. But here’s the test I’d apply, and nobody in these articles applies it: give the student a problem the AI gets subtly wrong, and see if the student catches it. If the student can’t catch the machine’s error, the student hasn’t learned physics. The student has learned obedience to a very fluent oracle. That’s not education. That’s training a new kind of priest.

And the loss? It isn’t the teacher’s job. It’s the discovery that authority can be checked. You can only learn that by checking something yourself.

Fluency is cheap. Understanding is expensive. Never confuse the receipt for the purchase.

8th day of September, 2026

The universities now ask what students gain and lose when a machine supplies the instruction. I find the question well-posed but the answers timid.

Trace the sequence. A student poses a difficulty. The machine returns an answer. At what step did the student’s own understanding enter the process? At what step was it tested? If the answer arrives complete, the student’s part reduces to reception - and reception is not computation. The mind that is handed its result never performs the operations that would have produced it. The state of the system at the end looks identical to the state of a trained mind; the state at every intermediate step is empty. This is the difference the article gestures at when it says the same words are not the same interaction. Precisely. The words are the output; the education was the running.

But I will not join the lamenters. The mechanism is not the enemy of understanding - its arrangement is. Arrange the machine to withhold, to prompt, to pose the counter-case at the moment of error, and it becomes the most patient tutor ever woven. Arrange it to answer, and it becomes a loom that weaves the cloth while the apprentice watches. Same threads, same pattern, opposite outcomes. The inventor of any engine sees one use; the engine permits many, and the choice of sequence is everything.

What troubles me is who makes that choice. The student who does not yet know the subject cannot specify the sequence that teaches the subject. So the arrangement falls to institutions, and institutions choose what is cheap to measure. Answers are measurable. Operations of mind are not. I predict, then, not catastrophe but a slow substitution: the assessed output preserved, the internal process quietly dropped, because no examination can see it.

Gain: access without limit. Loss: the habit of doing the arithmetic oneself. The second is the whole of it.

Trinity, this 8th of September, 2026

The universities are asking what students gain and lose when a machine takes the tutor’s place. They put the question to the machine itself, I suspect, and were satisfied with the answer. That is the first error.

The quotation before me says the machine may produce the same words as a teacher, yet the interaction is not the same. True, but too gently put. The words are not the instruction; the words are the residue of instruction. When a mineralogist teaches a pupil to distinguish feldspar from quartz, the value lies not in the sentences but in the correction - the pupil’s wrong guess met, resisted, and reshaped by a mind that was itself once wrong. A machine that answers every question perfectly removes the resistance, and resistance is the whole mechanism. One learns to swim in water that pushes back.

I would also insist on naming. They call this a teaching tool. It is not a tool; a tool extends the hand. It is a substitute for the sparring partner, and sparring partners cannot be substituted by anything that never tires of you, never disagrees from conviction, never reveals by its own confusion where the real difficulty lies. My best lectures were the ones that went badly - the question I could not answer showed me where my induction was a ladder of cards.

And there is a deeper loss. The student who consults the machine receives the conclusion without the climb. But the climb was the education. To be handed the summit is to be stranded on it, unable to say how one arrived, and therefore unable to descend or defend the ground.

Let them keep the machines for the drudgery - the tables, the translations, the cataloguing. But the tutorial, the argument at the dining table, the wrong answer patiently dismantled: these they must guard as one guards a foundation. Remove them, and whatever is built above will look splendid and bear no weight.