Homework We Never Finished

Before I get into it, a note. This is the first post under Notes, which is where I write about what I have been reading and what I made of it. It leans on one person's opinion rather than a position the whole team has signed off on. Please read it that way.

Let me start with what the review actually argues. Its core claim is that AI does not remain a remarkable tool. It turns into infrastructure, the way electricity and the internet did: astonishing at first, then convenient, then so ordinary that you stop noticing it is there. It works its way into drafting, diagnosis, financial analysis and teaching, until it is running somewhere out of sight.

So what is left for people? The review gives two answers. One is judgement. No matter how fast AI produces a plausible answer, deciding whether it is right still falls to us, which is why the expert does not disappear but matters more. The other is whatever a machine has trouble imitating: originality, and the trust that comes from meeting someone in person.

I agree with the broad shape of it. Go point by point, though, and there are three places where I see it differently.

They are not the same kind of infrastructure

The first is the comparison itself. You might fairly ask why I am picking at one analogy. I am picking at it because it leads directly into everything else I want to say.

Electricity and the internet sit further apart than they look. Electricity has already sunk into the foundations of daily life. Lighting and heating, obviously, but also the water that reaches your tap and the food that does not spoil in your fridge. When the power goes out, work does not slow down. Living stops.

The internet has not gone that far. There is still a great deal you can do without it. If you really wanted to, you could work with no connection at all. It is slower, and something feels missing, but it is not impossible.

That is exactly where I think AI sits. Take it away and the work slows down and something feels missing, but it does not stop.

The distinction matters because the two comparisons ask different things of us. You just use electricity. You never have to wonder what is coming out of the socket. The internet is different. Anyone using it has always had to judge whether what they are looking at is true.

AI did not invent this problem

The second is the warning the review draws out.

The danger is not that AI gets things wrong. It is that people fail to notice when it does.

Hyun Wol-an, review of Jason Schenker's book

I agree with the warning. But this is not a problem AI created. Deciding whether the top search result is actually true is a habit we should have built back when the internet was spreading through everything. Twenty years on, most of us still have not built it.

Back then, though, we had something working in our favour. We went to the sources ourselves. You scanned a page of results, opened a few links, noticed which site you had landed on, saw the comments underneath. You did not have to try to be suspicious. Some of it got filtered out along the way, without anyone intending it. Where vigilance was missing, the process made up part of the difference.

That process is gone now. AI does the gathering and the tidying, so what reaches us is not the source but a summary of it, already cleaned up once. The trouble is that the cleaning removes the very things worth being suspicious about. The passage that did not add up, the citation that looked odd, the detail that would have made you stop: all of it arrives smoothed over.

So when AI showed up we panicked as if the problem were new. AI did not create it. It took away the process that had been quietly covering for us, and made the output far more convincing.

Is expertise enough on its own?

The third is where the expert sits. The review concludes that the role grows rather than shrinks. That is true. But stopping there is a problem.

If sorting out what is true is left to expertise alone, then everyone without expertise stays dependent on somebody else's judgement. In that case the democratisation of information and knowledge that the internet and AI were supposed to deliver does not actually happen. We are left with the phrase and nothing behind it.

You can dig all the way down without being an expert. That means retracing each claim to its source, finding the original text and reading it, and refusing to skip the part where the pieces do not fit. That is stubbornness, and focus, and the decision to bother. It does not replace expertise. But for someone without expertise, I think it is close to the only way to get at the truth of anything.

There is a hole in that argument, of course. Telling everyone to be more stubborn is an easy way to hand a structural problem to individuals. That work belongs to whoever builds the tool: showing sources, exposing confidence levels, marking where a claim can be checked. But however well those things are built, they do nothing if nobody stops to check one more time at the end.

So I would add one line to the review's conclusion. I agree that what remains is the work of setting the standard for judgement. But if the only people who can set that standard are experts, then we have not gained a piece of infrastructure. We have gained one more thing to lean on.

I studied education, so my mind goes there first. I think this ends up being a question of teaching. Not teaching more domain expertise, but teaching the habit itself: the reflex to go back over whatever you were handed. Across every field, at every age, everywhere at once.

And it is not a new assignment either. It should have started when the internet was spreading through everything. We are roughly twenty years late, and we still have not finished it. That is why the title calls it homework.