Essay

The Fool's Errand

The grievance about generative AI is real and it is structural. Saying the thing shouldn't exist is still a losing bet.

I make my living as a musician, and I know what the work costs. Years of it, most of them broke, the kind of learning you only do by failing in private for a long time and not quitting. So when musicians I know first heard a machine produce a passable version of what they do, I understood the thing that went cold in them. They did not say it was bad. They said it was bad that it existed. I think they are right about the feeling and wrong about the conclusion.

I want to walk all the way into that side of it before I say anything else, because it is not a tantrum. It is an argument, and the argument is good.

Here is the argument as honestly as I can put it. A person lives a life. The life is hard. Out of that hardness, the struggle and the poverty and the years of doing the work when nobody was watching, the person makes something. The thing they make carries the cost of the life inside it. Then a model reads it, along with millions of other lives, and learns the shape of what was made. After that, anyone can have a version of it for almost nothing. The cost is gone. The collective gets the fruit without the labor, or a company sells the fruit and keeps the money, and the person who paid for it in years gets nothing back. Ask whether that is fair and the honest first answer is no. It is not obviously fair at all.

I am not standing above this looking down. I have the same years in my own hands, the same nights of playing to a half-empty room for gas money and a bar tab. So I am not going to tell you the feeling is irrational, because it isn't. The feeling is tracking something true. It is just tracking it to the wrong door.

And there's a second wound underneath the money one, which is worse and which nobody names cleanly. It isn't only that you might not get paid. It's that the hardest, most private part of you, the part you earned alone, turns out to be legible. Patternable. A thing a machine can read off the surface of your work like weather off a barometer. The years felt like they made you singular. The machine implies they made you predictable. That is a real grief, and I am not going to insult it by calling it a business problem.

Where the offense actually comes from

The offense feels like theft. I think it is closer to a category error baked into the law a century before the machine showed up.

We built intellectual property and commercial ownership to solve a real problem. Creative work is expensive to make and cheap to copy. If anyone can copy your song the instant you release it, you can't eat, so we drew a fence around the copy and called the fenced thing property. That was a reasonable thing to do. It paid a lot of people, including me. But the fence was always a workaround. It treats a song like a plot of land because land was the thing we already knew how to own, and for a long time the metaphor held well enough that nobody had to look at it closely.

A generative model walks straight through the metaphor. It doesn't copy your plot of land. It learns the difference between your land and everyone else's and then grows something on a field that didn't exist before. The law has no clean word for that, because the law was written to count copies, and there is no copy to count. So the grievance is real and the offense is real, and both of them are landing on a system that was never built to handle the case. The musician feels robbed by a thief who, under the rules as written, mostly isn't one. That gap is not a failure of perception. It is the structure failing under a load it was not designed for.

I say this to take the moral weight off the people who are upset. You are not naive for being angry. You are standing on a floor that was poured for a different building. The people who tell you to just adapt are usually the ones who never had to lay a single board of that floor themselves.

I watched this coming for twenty years

In the mid-2000s I read Kurzweil on the exponential growth of these systems. I didn't take it on faith. I sat with what exponential actually means, which is the thing almost nobody feels in their body, because we are built to expect the next step to look like the last step. An exponential curve looks flat and harmless right up until it doesn't, and the part that looks flat can last a very long time. I made myself believe the math instead of the flatness.

Then I watched it arrive in public. There was the clip of Will Smith eating spaghetti, the AI video everyone laughed at because the noodles behaved like worms and the face melted. Around the same time, the reconstructed Nirvana song, "Drowning in the Sun," a machine guessing at a dead man's next record. Both crude. Both easy to dismiss if you only looked at the frame in front of you.

I didn't look at the frame. I looked at the slope. A system that can do a bad version of a hard thing, and that improves on the whole record of human work every time it runs, is not going to stay bad. It recursively improves on the corpus we already made. The spaghetti was the flat part of the curve. The gap between the joke and the thing it was joking about was going to close, and close fast, and anyone betting on it staying open was betting against arithmetic.

I made my own peace with it early, and told no one. I decided not to bet against the curve. Nothing about that decision made me happy. It just kept me from being surprised.

The word that does the damage

Which brings me to the word I keep hearing, and the word is shouldn't.

It shouldn't be allowed. It shouldn't exist. They shouldn't be able to do this. I feel the pull of that word in my own chest. And it is, as far as I can tell, a fool's errand, because shouldn't is an argument about a world that doesn't exist anymore, made to an audience that has already moved into the new one.

This is the hard part and I am not going to soften it. Saying that something which already exists shouldn't exist is not a strategy. It is a kind of grief that has put on the costume of a position. The thing is here. It is being used by millions of people every day, including, I'd bet, some of the same people who say it shouldn't be. You can argue about how it pays, who it pays, what the rules around it become. Those arguments are alive and worth winning. But the argument over whether the capability gets to exist was settled by the curve a long time ago, and we were all watching when it happened.

The honest move, the only one I've found that doesn't curdle into bitterness, is to look straight at what is. Including the parts that cost people something. Especially those parts. People lost something real. The years are still in their hands and the machine got the shape of them for free, and no clever framing makes that even. I am not going to pretend the books balance. They don't.

What I won't do is spend the rest of my life insisting the water flow uphill because I can describe, in moving detail, how much better the valley was when it was dry. The valley was better. The water is here. Both of those are true at the same time, and only one of them is going to be true tomorrow.

Musicians keep playing. They play in small rooms for people who came specifically to watch a person do a hard thing in front of them with no net. A machine can make the sound now. It cannot yet be the person in the room who paid years for the right to make it badly in public and then, slowly, not badly. I don't know how long that stays true, and I am not going to bet against the curve on it. For now, the rooms still fill to watch a person do it.