Nobody left to argue with

Judgement is exercised alone. It is not formed alone.

September 2026

Taste is the kind of judgement people most readily call a gift. Last month, a designer whose eye I trust told me that some part of it cannot be taught; it can be improved and nurtured, they said, but people do not start from the same place. I disagreed, and could not say why.

The usual account of working alongside capable machines ends before this question comes up: the tools produce the options, a person decides which are worth keeping, and judgement is the part that stays human. That account begins with someone whose judgement is already formed, and it says nothing about how it was shaped or what keeps it from going stale.¹ This is about what keeps it alive in people who already have it.

A couple of days later I was having a drink with a friend, and the topic was still bothering me, so the conversation turned to taste: what it is, how anyone acquires it, whether it can be taught.

My friend has built a successful career on decisions made without a rulebook, yet on taste he describes himself as always a step behind, and he does not treat this as a flaw in himself. It was never considered important in his life, so he sees it as a gap in his education, and he wants recommendations.

Afterwards, the image I reached for was a man running alongside a train, trying to keep up. When I went looking for where it came from, I found it in Neal Stephenson’s Cryptonomicon, where it comes with the reason attached: “it’s not that we’re smarter than he is, necessarily, but that he’s a farmer who didn’t get a ticket.”

Whatever part of taste is inborn, the part my friend was missing had a door, and he was already standing at it, asking to be let in. He was eager to buy a ticket, eager to be part of the conversation.

And the conversation is what the ticket buys. Nobody on the train was born aboard; they got there by having people to talk to about these things, and it is in the talking that taste gets made.

A preference kept private never has to survive someone else’s account of what matters; said aloud to a person who might disagree, it has to become reasons, and the reasons are where it gets tested. As Celine Nguyen puts it, “In articulating your tastes, you are participating in a shared world.”²

The strongest reply from experience is the editor’s. “All an editor has,” Graydon Carter wrote, “is confidence in his or her own judgment and taste,”² and at the moment of decision that is exactly right, because the call is made alone. But confidence of that kind is the residue of years of argument with people who had the standing to say no, and it looks like a private possession only from the outside, and only at the end.

The costliest objection is simpler: there is no longer any shortage of things to argue with. A capable assistant will make the strongest case against a proposal on request, at any hour, without tiring or taking offence. The obvious worry is that it only tells people what they want to hear, but in one recent experiment AI advice moved people away from their starting positions on average, even though the model showed a measurable tendency to agree with them.³ A machine can supply a real objection, if you let it.

What it supplies, though, is an objection somebody asked for. For most of working history a second opinion came attached to a second person, and that person brought objections nobody had asked for: about what the work was for, about who would live with the result, about whether it was the right question at all. Those are the valuable ones, because they arrive where the asker was not looking, and a challenge that has to be requested tends to go unrequested at exactly the moment it would cost most.

In one study of lending decisions, people were keen to see an AI’s risk assessment but more likely to turn down the explanation when it might show that race and gender had shaped the result.⁴ The study was about explanations rather than conversation, yet the pattern is the one that matters here. The reasons were available. They were not wanted.

None of this shows that a machine cannot improve anyone’s judgement. The problem is not the machine but an arrangement in which every objection has to be asked for, and in that arrangement it is possible to keep the human in the loop while quietly removing the other humans from it. The approval survives; the colleague who might have asked what the work was for becomes someone who has to be invited.

My friend wanted a ticket because of what it buys: people to argue with. It would be a strange moment for the rest of us to get off the train.