What Excellence Was For

In March 2016, in a hotel in Seoul, Lee Sedol played five games of Go against AlphaGo and lost four of them. He won the fourth, with a wedge at move 78 that the program's own policy network had rated as a move a human would find about once in ten thousand tries. In November 2019, at thirty-six, he retired from professional play, and he was direct about the reason: something had arrived that could not be beaten, so being number one would no longer mean what number one had meant.

Nothing had happened to his ability. Every shape, every joseki, every endgame count he had spent three decades loading into his head was still in there, and he could still have beaten more or less every human being alive. What changed was not how good he was. It was what being that good was worth.

The answer this site already gave

An earlier essay here, Meaning After Work, has an answer to this, and it is a good one, so it is worth stating at full strength rather than in whatever form is convenient to argue with.

The fear of being automated into irrelevance, that essay argues, rests on a premise almost nobody would defend out loud: that a person's worth is a function of what they can produce that nothing else can. Drop the premise and the fear loses its object. Meaning lives in the doing rather than the done, in relationships that nothing in a data centre touches, and in contribution — effort meeting a real need. Chess is offered as the proof. Deep Blue beat Kasparov in 1997, engines left every human comprehensively behind in the decade after, and more people play chess now than at any earlier point in the game's history. Nobody was ever playing to supply the world with moves it lacked.

All of that is true. On Lee Sedol's particular problem, it is also an answer to a question he was not asking.

Watch the chess example do its work, though, and you can see the join. For the club player the engine changed nothing, and it changed nothing for exactly the reason given: the value was always in the playing. But Lee Sedol was not a club player, and neither is the staff radiologist, the working translator, or the illustrator who spent twenty years getting fast. You can still enjoy it is addressed to the amateur and handed to the master. It is true, it costs nothing to say, and it does not touch the injury.

Two reasons to have been good at something

The injury is easier to locate if you ask a blunter question: what was the excellence for? There are two honest answers, and they behave completely differently when a machine shows up.

• It made you needed. Skill was scarce, so having it stood you where a need was and the need did the rest — you were the one who could set the bone, find the line, render the likeness, read the film. That is the productivity theory of meaning in better clothes, and it is not a delusion: being depended on genuinely is one of the ways a life comes to feel load-bearing. It is also precisely what a capable machine removes, and it removes it whole.
• It placed you among people who knew what it cost. A rank among peers, a teacher who finally nodded, an audience that could see the difficulty. This standing is made of other people's attention rather than of the scarcity of the output, and a machine does not reach it — not because attention is sacred, but because the machine is not in the relationship.

Nearly everyone who loves their work has both, braided so tightly that the loss arrives as one undifferentiated thing. Pull them apart and you can predict what survives. Professional chess survived, and not because engines politely stopped improving: the audience was never paying for the best available moves, which are free on any laptop, but to watch particular people, under pressure, whose earlier games they knew. Correspondence chess did not survive in the same sense at all — its whole point had been unhurried human analysis, and at the top it is now a contest in who prepares best with an engine. One game, two goods, and the machine took exactly one of them.

What a machine takes is not your skill and not your enjoyment of it. It takes the part of your skill that consisted in other people having no alternative.

The part that is really gone

This is where the consoling version has to stop. That first good is real, and when it goes it does not come back, and no argument returns it.

For two centuries, if you wanted to carry your wife's face around with you, you commissioned a miniature: a portrait a few inches across, painted in watercolour on ivory by someone who had trained for a decade to work at that size. The daguerreotype was announced in 1839. Within roughly a generation the trade was finished. It would have been no comfort whatsoever to tell those painters that the pleasure of painting remained available to them and that they might concentrate on process rather than product. They had not lost a pastime. They had lost being the ones who could.

A good number of them became photographers, or hand-coloured other people's photographs. That is a real answer and not a small one. It is a different answer from the one that says nothing was lost.

The announcement and the fact

There is a second thing the economic argument leaves out, and it is not consoling either. Being replaceable is a condition. Being told you are replaceable is an event, and the event arrives years before the condition does, when the condition arrives at all.

Machine translation is a fair case. It has taken the instrumental end of that trade nearly whole — the manual, the menu, the support ticket, text that exists to be understood once and then thrown away — and it has not taken the end where the value is a particular ear making several thousand particular choices across three hundred pages. But the announcement did not arrive in two parts. It arrived as one sentence, translation is solved, and it landed on both kinds of translator at once, and the second kind has spent years explaining a distinction that the people repeating the sentence had no reason to care about.

The gap matters because the two injuries call for opposite responses. Where the capability really is here, the honest task is to find what has become short — and what a flood of cheap competent output makes scarce is the judgement to tell competent from good, which is the half of any craft that was never visible in the artefact. Where the announcement has merely outrun the capability, the task is to not believe it, which is far harder than it sounds once your clients already do.

What to say to someone in it

Do not tell them their worth was never in their output. It is true and it is useless: it answers a question about worth to a person asking a question about grief, and a correction is not what anyone needs in the week they lose a standing they spent a life earning.

What helps is narrower and less flattering to the person saying it. Name what actually went — usually being needed, a rank, and a future that was being built on the assumption the rank would hold. Then name what did not go: the judgement, which a flood of plausible output makes more valuable rather than less; the people who understand what the work costs, who are still there and still understand it; and the doing, which was never anyone else's to take in the first place.

That does not add up to no loss. It adds up to a loss smaller than the one panic reports, and correctly located, which is the most that is honestly on offer.

Lee Sedol's last official games were against a machine. In December 2019 he played a Korean program called HanDol in his retirement match, won the first game with a two-stone handicap, and lost the next two. It is not a triumphant image and it is not meant to be. It is a man who had lost the thing his excellence was for, taking the handicap, and playing anyway. The question the machines are putting to one profession after another is not whether being surpassed hurts — it does. It is which of the two things you had is actually gone, a question you have to answer yourself, at speed, while someone with no stake in the answer explains that nothing has really changed.