AI needs shame, not taste

Contents

The most flattering answer in the room

Taste never stopped anyone

What shame is for

Being seen

But isn't this just guilt?

The cheaper virtue

You have seen the posts. The LinkedIn announcement that opens "In a world where", four hundred words long and not one of them the author's. The comment that thanks you for "this incredibly thoughtful piece" and then describes a piece you did not write. The company email so completely handed over to a machine, start to finish, that no human could have read it back without wincing, which is the tell, because plainly no human did. Bad work is not the part that gets to me. Bad work has always been with us. It is that nobody making it appears to feel a thing. Not a wince, not a flicker. Nobody gives a shit.

I spend my days in it. Most of my working life has gone on lifecycle marketing, so it reaches me by the hour: the forty-slide deck a colleague forwards having plainly written none of it, the Slack reply that took a machine to compose and no one to mean it, the brief that turns up with the placeholder text still in it. Whoever sent it felt nothing. I still do, which increasingly feels like my problem, not theirs.

There used to be a flinch. A small involuntary stop somewhere between making a thing and sending it, the moment you pictured the person on the receiving end and felt, before the thought had finished, that this was not good enough to put in front of them. It came faster than judgement and did its work while judgement was still clearing its throat. That is what has gone missing from the people filling the channels, who appear to have stopped feeling it.

And the industry, asked to name the human thing the machines still cannot manage, has reached for exactly the wrong word. The word everyone has settled on is taste.

The most flattering answer in the room

You can more or less date the consensus. In February Paul Graham told a couple of million followers that taste would only get more important as AI improved, because once anyone can make anything, the question that's left is what you choose to make. He linked an essay he'd written back in 2002. Two days later Greg Brockman, OpenAI's president, followed with a tidier version: taste is a new core skill. From there it hardened into a genre. Taste is the moat. Taste is the last defensible advantage. Taste is the soul of the thing the clones can't copy. Search the phrase and you'll find a hundred variations, each one explaining that AI has driven the cost of competent execution to roughly nothing, so the only edge left is knowing what's worth making in the first place.

I understand why it caught on. It is the most flattering answer available. It takes the person hammering the generate button and recasts them as a curator, a discerning hand, a tastemaker picking from the machine's offerings like a buyer working a gallery wall. Suddenly everyone on LinkedIn can pat themselves on the back about how much taste they have, unlike, of course, everyone else. It only works as flattery if other people are short of the stuff, and it asks nothing of you beyond the conviction that you are not one of them, a conviction in generous supply. Best of all, it lets you carry on shipping at the new volume with a clear conscience. Taste is the one virtue you can award yourself while doing exactly what you were already doing, only faster and in greater quantity.

Taste never stopped anyone

Taste is real. Graham is right that it is not just a matter of opinion: get better at making things and your tastes shift, and your old preferences turn out not to have been merely different. They were worse. Good work has properties you can name and learn to see. It is a genuine skill, it sharpens with miles on the clock, and a person who has it can look at two things and be correct about which is the better one. Fine. Granted, all of it.

But what is taste even for? The work, and your relationship to the work. It answers two questions, is this any good and what should I make, and both of them run entirely between you and the thing on the screen. The person who eventually receives the thing never really enters into it, except as an abstraction, a market to be gratified. Taste is you, alone, marking your own bloody homework.

Which is exactly why it has never once stopped anyone doing the thing it is now being sold as the cure for. I have sat in the meeting where everyone round the table can see that the campaign is beneath us, can say out loud why it is cheap, and we ship it regardless, because the boss is carrying an AI mandate from upstairs, and stopping it was never taste's job. Taste tells you the slop is slop. It has not, in the entire history of the species, stopped a single person pressing send.

What shame is for

The faculty that did the governing was shame, which is awkward, because shame is the emotion we have spent the last couple of decades training ourselves to disown.

A group of evolutionary psychologists at UC Santa Barbara have spent years making the case that shame, far from being a malfunction we would be better off without, is one of the more carefully engineered systems in the human mind.1 Their information threat theory holds that shame evolved as a defence against social devaluation: the intensity of shame you feel about a possible act tracks, closely and across wildly different cultures, how much other people would think less of you if they knew you'd done it. It is a pricing mechanism. It takes the private payoff of doing something, weighs it against the cost in other people's regard, and when the second outweighs the first it stops you. Later work in the same programme found the trigger is the devaluation itself rather than any actual wrongdoing; shame climbs with the sheer publicity of an act, whether or not you did anything wrong.

That is why shame, and not taste, is the thing that used to keep the volume down. It is the only one of these faculties that puts the audience in the room before you act. Taste is you and the work. Shame is you and the face of the person about to receive the work, pictured clearly enough to change what you do next.

Being seen

What AI removed was not taste. The machines are, if anything, drowning in the stuff; they have read everything and can do any register you care to name. What they removed was exposure: the moment when whoever made the thing had to stand there while someone took it in.

What little stopped it lived in the labour. The needless email, the report nobody asked for: putting one in front of someone used to mean making it first, and being, for a moment, the person who made it. That cost something. Not much, often nowhere near enough, but now and then the small private embarrassment of being its author was what killed the thing before it shipped. Now no one makes it. The model makes it, a human forwards it, and the embarrassment never lands, because nobody was ever quite the author.

The cost of the slop has not gone anywhere, though. It has moved. A team at Stanford and BetterUp gave the workplace version of this a name, workslop, and the useful part of their argument is who pays: AI-generated filler shifts the burden downstream onto whoever receives it, who then has to work out what it's for, what's missing, and what's quietly wrong with it.2 The people who receive it also come away rating the sender as less capable, less creative, and less worth trusting. The devaluation that the shame system exists to anticipate is real, and measurable, and it duly happens. It just happens to the wrong person, in a room the sender is not standing in. The verdict gets handed down and the defendant never hears it read out.

Shame only works when the people who would think less of you are actually there, when you have to face them while they take it in. That is what has gone. Taste was never going to do that work; it never involved those people at all. So you end up with the worst arrangement available: the judgement still gets made, the regard still drops, the relationship still quietly corrodes, and the one person who could act on it, the one who pressed generate, is the only one who never has to hear it.

The machine has no shame. That part is fine. It is a machine, and we never expected it to have any. The damage is that, by removing the moment of being seen, it has quietly relieved us of ours as well.

But isn't this just guilt?

Anyone who has spent time near the psychology of this will, by now, be wanting to correct me. You mean guilt, not shame. There is a well-worn distinction, drawn most influentially from June Tangney's research: guilt is feeling bad about what you did, shame is feeling bad about what you are, and the settled view is that guilt is the healthy, reparative emotion while shame is the corrosive one that drives people into defensiveness, withdrawal, rage and worse.3 Cultivate guilt, fight shame. As advice to an individual human being, it is good advice.

Only it is about a different thing. The corrosive shame in that literature is a trait, a standing disposition to read every failure as proof that you are a bad person. The mechanism the Santa Barbara group describes is something else: the reflex that fires in the moment before you act and asks what this will cost you in other people's eyes. They share a name and not much else.

And the reason we are taught to prefer guilt is the giveaway. Guilt is the private one. You can feel it, make your amends, and move on without anyone else ever being involved. That privacy is the loophole. Guilt is the thing you can talk yourself out of at three in the morning. Shame needs an audience; it is enforced from the outside, by the live prospect of other people actually thinking less of you. Which is why it is the one of the two that has ever reliably moved an institution, a profession, or a person working to a quota. Guilt is deniable. Shame is the bill turning up.

The cheaper virtue

We were sold taste because taste is the part of all this we were only too happy to keep. It flatters us, it costs us nothing, and it lets the volume stay exactly where it is. The thing we actually lost is duller and far more useful: the small dread of being caught not having bothered, in front of someone who would notice. The machines were never going to feel that on our behalf. That part was always ours to do.

And we have lost the flinch.

1

Daniel Sznycer, John Tooby, Leda Cosmides, Roni Porat, Shaul Shalvi and Eran Halperin, "Shame closely tracks the threat of devaluation by others, even across cultures," Proceedings of the National Academy of Sciences 113(10), 2625-2630, 2016. https://www.pnas.org/doi/10.1073/pnas.1514699113 The follow-up on devaluation as the trigger is Theresa E. Robertson, Daniel Sznycer, Andrew W. Delton, John Tooby and Leda Cosmides, "The true trigger of shame: social devaluation is sufficient, wrongdoing is unnecessary," Evolution and Human Behavior, 2018. https://www.sciencedirect.com/science/article/abs/pii/S1090513817303872

2

Kate Niederhoffer, Gabriella Rosen Kellerman, Angela Lee, Alex Liebscher, Kristina Rapuano and Jeffrey T. Hancock, "AI-Generated 'Workslop' Is Destroying Productivity," Harvard Business Review, 22 September 2025. https://hbr.org/2025/09/ai-generated-workslop-is-destroying-productivity The underlying numbers come from a self-reported survey of around 1,150 US desk workers run by BetterUp with Stanford's Social Media Lab rather than from peer-reviewed work, so treat the specific figures as indicative. The mechanism holds regardless of the figures: the cost moves to whoever receives the slop.

3

June Price Tangney, Jeff Stuewig and Debra J. Mashek, "Moral Emotions and Moral Behavior," Annual Review of Psychology 58 (2007), 345-372. https://pmc.ncbi.nlm.nih.gov/articles/PMC3083636/