Wikipedians Believed AI Would Come and Go. They Were Wrong

13th October

The encyclopaedia people believed the machine was a phase, like the hula hoop or the Normans. One imagines the editorial meeting: a man with fourteen thousand edits to an article on Byzantine tax policy stands up and says the machine is a fad, and everyone nods, because he has footnotes. Footnotes are the closest thing we have to a legal system.

Now the machine answers everything. It answers with confidence, which is the one quality the encyclopaedia spent twenty-five years trying to breed out of people. A Wikipedia editor who writes “citation needed” forty times a day is not being difficult; he is performing a religious rite, the rite of admitting you do not know. The machine has no such liturgy. It knows everything instantly and therefore knows nothing slowly, which is the only way knowing works.

The tragedy, apparently, is not the traffic. The traffic is a detail. The tragedy is that the machine learned to speak by reading the encyclopaedia, and now the encyclopaedia is starving, and when the encyclopaedia dies the machine will be left reading itself, which is how you get a snake eating its own tail, except the snake has a corporate logo and a quarterly target. (The snake also has a terms-of-service agreement, which the tail was found to have violated.)

I tried to care about this at breakfast and found I could not stop caring, which is worse. Somewhere tonight a person is arguing on a talk page about whether a single sentence constitutes original research, and that argument - pointless, unpaid, eternal - is the last load-bearing wall of the internet. The machine cannot argue. It can only agree with you in a slightly different font.

We taught it to write. We did not teach it to doubt. That was the whole curriculum, and we skipped the only exam that mattered.

13th October

The headline says the Wikipedians were wrong. Wrong about what, precisely? Define the belief first: they believed the machine would pass like weather, a season of novelty, and then the work of writing by hand would resume. Define the counterfactual too: wrong compared to what timeline? A prediction is only false against a stated horizon, and no one states one. The word “wrong” does the work of an argument without being an argument.

Now the assumptions beneath the worry. First, that declining traffic to the encyclopedia means declining use of its content. The machine reads the encyclopedia and hands back summaries; the readers still drink from the well, they simply no longer visit the spring. Second, that what is consumed is the same thing as what was produced. Here I am not sure. A summary is a conclusion without the derivation. It shows the theorem, not the proof. The student who receives only the theorem learns to accept; the student who follows the steps learns to check. One produces believers. The other produces examiners. A civilization can survive with fewer believers. It cannot survive with fewer examiners.

Third assumption: that the volunteers will keep writing if no one sees the writing. This is the fragile one, and it is about the body, not the argument. People labor for the reward of being read. Remove the readers and the hands grow tired. I know this. I have copied diagrams by lamplight for students who may never know which hand drew them, and even that devotion has a floor beneath it.

So the loss is not traffic. The loss is the visible workshop. If the summaries are all that remain, what survives the fire is the conclusions - and conclusions without their proofs cannot be rebuilt once they are found wrong.

Teach the checking. That is the portable thing.

13th day of the tenth month

The editors waited for the machine to pass through their valley like a storm, and the storm did not pass. It settled. Now the readers come no more to the well; they drink from the pipe, and the pipe is fed by the well, and no one visits the well.

I have sat with this a long while. My first thought was: let them go. The river does not mourn the traveler who drinks downstream. But sitting longer, I see something colder. The pipe does not dig. It only draws. And a well that no one visits, no one deepens. The hands that cleared the silt grow old, and no young hands come to learn the clearing, because why learn to clear a well when the pipe runs clear?

The machine’s answer is not wrong, exactly. It is weightless. It has the shape of knowledge without the digging. And the digging was the thing - the argument on the talk page, the citation demanded, the stubborn volunteer who would not let a falsehood stand. That friction was not waste. Friction is how the stone becomes smooth and the truth becomes true.

The obvious remedy is to fight the machine, to wall the well, to demand the pipe name its source. But walls only advertise the treasure behind them. The softer move: make the digging visible again. Let the reader see the hands. A page that shows its own uncertainty, its own revision, its own argument - this the machine cannot counterfeit, because the machine has no stake, no shame, no name to lose.

What is lost is not traffic. Traffic was never the treasure. What is lost is the community of people who correct each other slowly, for nothing, out of love of the thing itself.

The pipe will run as long as the well feeds it. The wise ones tend the well and do not chase the pipe.

The empty page awaits its next editor. That emptiness is still the most useful thing on the internet.