The Machine Doesn't Have to Be Good

July 27, 2023 · essay · 8 min · ai · labor · hollywood · automation


Before the writers walked out, a small group of them spent months trying to prove that the machine could take their jobs. They failed.

The union had pulled them together to study the thing, and several of them wrote science fiction, so they knew how the story was supposed to go. They fed the machine prompts, again and again. They gamed out how a studio might use it. And they came away agreeing on one dull, inconvenient fact: it was bad at the work. It could summarize. It could finish a list. One of them, a comedy writer, said it was not so great at the kind of thing they actually did. Another spent hours prompting it and found that squeezing out anything usable took nearly as long as writing from scratch, which, he admitted, made him panic a little less.

Then those same writers went on strike, and a set of rules for that same machine sat high on their list of demands. When the actors joined them that July, their leaders reached for a bigger word. Existential.

Hold those two facts side by side. The people who studied the machine most closely concluded it could not do their jobs. Then they put their income on the line to fight it anyway. Either that is a panic over nothing, or the thing they were afraid of was never the thing the rest of us were watching.

The easy reading is the one every strike gets. Machines are coming for the creative work. Soon the software will write the movies and perform the parts, and these people are making a doomed last stand against the future, the way the weavers once smashed the looms. It is a clean story. It has been told about every new tool in the industry's history. And in the summer of 2023 it was mostly wrong, because it misread what the fight was about.

Start with what the fight was not. It was not a simple fear of technology. The writers who studied the machine did not want it banned. Asked about a ban, one of them said you cannot ban a tool, and that the right approach was to use it as one. Their actual proposal let a writer open the software, lean on it, and keep full credit and pay. The union representing the crews, the electricians and carpenters and hair and makeup people, went further and said its members had adapted to new technology for 130 years and would rather be trained on this one than shielded from it. These were not Luddites; some of them were the machine's earliest users.

Nor was the fear that the machine could already replace them. The people closest to it had just spent months proving it could not. So the panic was not about what the software could do in July of 2023. It was about what a piece of paper could do.

To see it, leave the writers for a moment and go stand at the bottom of the business, where the leverage runs out, with the extras.

A day player on the superhero movie The Flash was working as an extra when he was sent to a tractor trailer and told to take some pictures. Inside were hundreds of cameras. He stood with his arms raised while the rig took a 3D scan of his body. He was told it was for continuity and special effects, the same words you would use for ordinary production work. He was also told that if he refused, he would be sent home without pay. So he did it. It was a day's work. The rate for that day, for the lowest paid people in the actors' union, was $187. About 32,000 people did that job at least once that year.

He did not walk out of the trailer thinking he had signed away anything. He only wondered later. We don't know what all our scans are being used for, he said. They could possibly be used in other movies and other shows. He saw background work the way most extras do, as a first rung, a way onto real sets and eventually into real parts. And it occurred to him, after the fact, that a studio might now hold a copy of his face and body, captured on an afternoon when his only choice was to raise his arms or lose the day.

Here is the paperwork that sat next to the cameras. Background actors are hired through casting agencies, and they get a voucher with the pay stub. On the back, in fine print, is a sentence. It grants the production all rights of every kind to the results of the person's services, to be used or not used in any manner the company chooses, throughout the universe in perpetuity in all media whether now known or hereafter devised. That sentence is not science fiction. It is boilerplate. A member of the union's negotiating committee, himself a background actor, said the union had always treated it as meaningless. Then he said the rest of it: in light of the advances of AI, it takes on a whole new meaning.

A body turned into data for $187, under a sentence that reaches to the end of time. Another extra, who said he always refused the scans, asked the obvious question. If they have my image, and they can do anything they want with it, why would they ever hire me again.

Now the strong version of the other side, because it is stronger than the strikers' opponents usually get credit for.

None of this technology is new, and most of it is dull. Scanning actors, by one account, is now as routine as doing their hair and makeup. It exists for good reasons. A director explained that if a stunt is too dangerous, he can drop a scanned double into it instead of risking a person, or add an actor to a shot he could not make that day. Crowds have been faked for decades by a trick called tiling: film a clump of extras in one part of a stadium, move them, film them again, and stitch the copies together in the edit. A visual effects supervisor said that in his experience a scan is one day of work and stays inside the scene it was captured for, and that only the person scanned should ever control where the likeness goes. He was not defending abuse. He was describing his Tuesday.

And the studios' alliance said, flatly, that the union's nightmare was false. Its proposal, it said, let a background actor's digital replica be used only in the movie he was hired for. Any other use would need his consent, a fresh negotiation, and a minimum payment, and that rule covered background actors just as it covered stars. The writers' own demand, remember, was not a ban but a boundary: the machine could be a tool, like a pencil, so long as it did not take the writing credit, because credit is what drives the residual checks. Even the voucher had a boring pedigree. The casting agency pointed out that the union had reviewed that in-perpetuity language many times over the years and never once complained. This was normal paper. Everyone had signed it for decades. On this reading, the strike was a panic, and the cure was simple: get consent, pay for reuse, and go back to work.

That case holds, completely, on one condition. It holds only if consent means something.

Consent at hiring is not the same as consent at use, and the difference is not a technicality. When you sign at the start of the day, you cannot know the thing you are agreeing to. You do not know the movie, or the scene, or the year, or whether the face on the screen will be doing something you would never do. The union's negotiator had a name for agreement gathered that way, as a condition of getting the job, with refusal meaning no work: he called it fictional consent, one sentence on page seven of a twelve-page document. The studios proposed that a background actor's consent may be obtained at the time of employment, and said any later or different use would still need its own consent, its own bargaining, its own pay. The union wanted the consent that counts gathered at the moment of use, when the worker could finally see what he was agreeing to. That gap, small enough to sound like lawyers arguing, was the fight.

And notice what had actually changed. The in-perpetuity sentence was old. What was new was that the sentence had finally become worth cashing. The same director who described the safe, dull uses of a scan said the other part out loud: building a digital character used to be so expensive that the cost itself protected the human, and now the computer can do it cheaper, and in some cases better, than a person. The expense was the fence around the job. The fence was coming down. The machine did not need to be good yet; it only needed to be getting cheaper, while the raw material, the scans and the drafts, got signed over now, at yesterday's price, before anyone had to admit what it was worth.

You do not need a finished machine to make a worker optional. You need the worker's material captured while it is cheap.

This is why the machine's badness was beside the point, and why the writers were not being irrational to fear a tool that could not yet do their work. You do not need a finished machine to make a worker optional. You need the worker's material captured while it is cheap. For the extra it is a body scan. For the writer it is the draft. One of them described the real fear precisely: not that the software would write a good script, but that a studio would generate a bad one, hand it to a writer for a few days of cleanup, and pay him as a fixer instead of an author. Same work, a fraction of the price, because the expensive part had been quietly moved to the machine and the human had been repriced as a finisher.

And it does not stop at the studio gate. That spring, the outside estimates were already in. One bank put the number of jobs exposed to the new software at 300 million, and figured it could reach nearly half of all office and administrative tasks and a large share of legal work. Those were tables and percentages, the kind of thing that scrolls past. Hollywood did what the tables could not. It put a body in a trailer and a voucher on the back of a pay stub, and it showed you the mechanism at human scale, at the one place it always shows first: the bottom rung, where the worker has the least power to say no.

The extra with the scanned body kept pushing. He wrote to the leaders of his union with a single request. Whatever deal they reached, he wanted it to force the studios to delete the scans they had already taken. He still did not know what was on them, or where the copies had gone, or what they might one day become. He only knew he wanted them back, and that he had no way to get them.

The question was never whether the machine could do your job. It was whether, on some ordinary afternoon, for a day's pay, you had already handed over the thing it would need to.