The Generation Generation is over

Creation used to be the hard part.

For most of my schooling, the prized skill was making something from nothing. Write the essay. Solve the problem. Produce the painting. Education rewarded generation, and for good reason: producing something forces you to understand it, maybe unconsciously and often deliberately.¹ Now, generative AI has arrived and made creation cheap. A passable essay, a plausible image, a first draft of almost anything is now a sentence away.

So the Generation Generation is ending, and something I have started to think of as the Curation Generation is taking its place.

When anyone can generate, the scarce skill is no longer production but judgement: knowing what is worth keeping. Curation is more than sorting the good from the bad. It is checking whether a claim is actually true, noticing where a confident paragraph has quietly gone wrong, catching the bias, and reshaping raw output into something that fits a real audience in a real context. Generative models produce at a scale no human can match.

Here is where it gets more complicated than that™. We never really taught curation on its own. We taught creation, and curation came along for free, because the surest way I know to learn to judge a piece of work is to have struggled to make one.² You edit your own essay well precisely because you remember writing it: you know which parts you laboured over, and which you were quietly unsure of. Outsource the making to a machine, and you risk producing curators who never built the critical thinking that curation depends on. Judgement was a by-product of the very effort we are now automating away.

I have been turning this over for a while, and I should say where I stand. I learned in the older way: taught by generation, slowly. That was not a virtue, just the environment I grew up in, and it is the environment that is changing. So I find myself thinking about the generation coming up, including the students I will one day teach. Not that they will be worse for it, but that they will have to build judgement deliberately, where I built it as a side effect. The ones I teach now are at Master’s level and, for the most part, still reach for these tools with care; my unease is less about them than about what younger students are growing up inside. And it is not only mine: several of my extended family are teachers, and they are seeing the same thing from the front of the classroom.

The thing I keep coming back to is attention. Generating something almost always took longer than taking it in. Writing is slower than reading; making a film is slower than reviewing one; coding is only a little slower than reviewing someone else’s code. It is a spectrum. That slowness was never only a cost. It was where attention had to live, and attention is where judgement quietly forms. Social media had already thinned that attention before the chatbots arrived. So those born roughly between 2000 and 2010 (loosely, and not only them) meet thinned attention and effortless generation together, just as judgement is still taking shape.

The way through, I think, is to turn the machine’s weakness into the lesson. AI’s confident errors are not only a nuisance; they may be the best whetstone for judgement we have been handed in a long time. The invented citation, the plausible-but-wrong proof, the statistic that was never real: each is a chance to practise catching what looks right but isn’t. Give a learner a stack of AI-generated work and ask them to find what is wrong with it. Few things sharpen a critical eye like catching a machine mid-mistake, especially in a subject you are still learning, where the errors are subtle enough to fool you.³ We used to build judgement by making things badly and then fixing them. We can rebuild it by watching a machine make things badly and learning to recognise it.

There is a catch I should be honest about: we do not yet have a proven way to grow judgement without some version of the making. The research on teaching “critical thinking” as a free-floating skill is sobering: it transfers reasonably well to problems that resemble the ones you practised on, and poorly to genuinely new ones.⁴ Judgement, it seems, needs something to be about. So the answer is probably not zero creation but a smaller, sharper dose of it: make a little, then take it apart, and take the machine’s version apart beside your own.

Which leaves the harder question: why would anyone bother, when generating is so much cheaper in the moment? For a while I assumed the reward was real but far off, the way a good habit is. I have started to suspect it arrives sooner. When generation is nearly free, the scarce thing is the judgement layered on top of it, and scarce things get paid.⁵ The obvious objection is that the machines will get better at curating too — and in part they will. But someone still has to choose what matters, to verify it, and to be answerable for the call; for now, that someone is us. This is not yesterday’s job done more slowly; it is the job that is left. There may be a quieter reward, too: in a feed of endless competent sludge, something that is unmistakably yours, thought through and chosen and meant, is rarer than it used to be.

None of this makes curation nobler than creation, or creation obsolete. The two were always entangled; we are only changing which one we start with. The real risk is narrower, and worth naming: that we hand people the power to generate without ever teaching them to doubt.

So if I were redesigning how we teach, I would bring the uncomfortable habits forward: scepticism, checking sources, the reflex of asking “how would I know if this were wrong?” Not because making things no longer matters, but because in an age of infinite drafts, the person who can tell the good one from the merely fluent is the one doing the real work. The Generation Generation handed us the tools. The Curation Generation gets to decide what we do with them.

For the same idea from the teaching side, see adaptive adaptive learning.