On the legitimacy of AI-era authorship
Martin Scorsese has never operated the camera, never acted a role, never lit a set or designed a costume or scored a single bar of the music, and yet no one looking at Goodfellas or Taxi Driver or The Irishman hesitates for a moment to call it his. The hesitation would be strange, because we understand instinctively what a director is: not the person who makes any one of the thousand things a film is made of, but the one whose judgment runs through all of them, who threads a sensibility through every shot and every cut and every held or broken silence until a hundred separate crafts resolve into something that could only have come from him. His medium is not film stock or light. It is coherence. He authors the whole without executing any part, and we have never found this confusing or thought it diminished the authorship — if anything, the orchestration is the art, the rarest and most valuable thing on the set.
Hold that next to the question we have started asking with a suspicion bordering on contempt: when someone makes a picture or a piece of music or a paragraph by directing an AI toward it, have they made anything, or have they merely pressed a button and taken credit for what fell out? The accusation is cheating — that real authorship requires the hand on the camera, the pen on the page, and that everything else is a kind of fraud. But the director has been quietly disproving exactly that premise for a century. Scorsese's hand is on none of the components and his authorship is total, and the reason is the thing the accusation keeps missing: that the act of authoring was never primarily the act of execution. It was the act of judgment — of knowing, out of the unboundedly many ways a thing could go, the one way it should.
Read in that light, the person who works seriously with these models is doing something far closer to directing than to cheating. The model arrives like a member of the cast, trained as actors are trained on the whole archive of what humans have made, carrying its own range and its own grain and its own stubbornnesses, capable of a performance but incapable of knowing which performance the work requires. You can draw something remarkable out of it, but only if you walk in already knowing what you are listening for, and, harder still, already knowing what to refuse. A prompt, done well, is not a command barked into a vending machine. It is the shaping of a space of possibility, the setting of a frame inside which something is invited to happen, which is the director's act precisely: Scorsese choosing this actor and not that one, asking for the take again and again, recognizing the one reading among forty that has the thing the scene was always reaching for and building the film around it.
The closest analogy in the old world is the one the new tools make almost literal. Kubrick was said to shoot a scene fifty, eighty, past a hundred times, not because he was short of usable footage, since he had usable footage after the third take, but because usable was not what he was after. He was chasing something exact, a quality he could not always name in advance but could recognize without fail the instant it appeared, and he was willing to burn through ninety-nine near-misses to get the one that had it. Anyone who has worked at length with a generative model knows the shape of that pursuit from the inside. You generate, and most of what comes back is lifeless, technically competent and dead on the page; you throw it out, and throw out the next, and the next, and then something flickers in one of them, the spark the others lacked, and the whole craft of the thing is in seeing that spark when it comes and knowing how to coax it the rest of the way. The labor is not in the generating. It is in the judgment that knows when to stop, the judgment that recognizes when the thing is right, when it is finished, when it has become in the only sense that ever mattered yours.
The obvious objection is that the analogy cheats in the other direction: that Kubrick and Scorsese directed people, autonomous and gifted and capable of refusing, surprising, walking onto the set with something the director never imagined, and that a model is an instrument, not a collaborator, so that calling its operator an author is like calling a man who plays a player piano a composer. It is the strongest thing the skeptic can say, and it is worth saying plainly before answering it. But press on it and it gives, because it locates authorship in the wrong place, in the autonomy of the materials rather than in the judgment imposed on them. The actor's freedom was never what made Scorsese an author; if anything it was the obstacle his authorship had to overcome, the raw and unaligned material he bent toward his vision the way a sculptor bends marble that would much rather stay a block. Whether the thing on the other side of the table is a willful human or a probabilistic model changes the texture of the collaboration enormously, but it does not touch the seat of authorship, which sits where it has always sat: in the one deciding what the work is for, what belongs in it and what must come out, what counts as right. The player-piano comparison fails for the same reason — the pianola reproduces a fixed roll, offering nothing to select between, while the model offers an effectively infinite field of divergent possibility and does nothing useful until a discriminating taste reaches in and chooses what the work will be. That act of choosing, and not the hand on any instrument, is where authorship has always lived.
None of which means everyone pointing a model at a blank page is an author, any more than everyone who has ever called action is Scorsese. The distinction that has always separated the director from the hack survives the change of medium intact, because it was never a technical distinction in the first place. There are auteurs and there are people assembling content to fill a slot, and the line between them does not run along whether the tool was a camera or a model. It runs along whether a genuine sensibility is shaping the outcome or whether the outcome is simply whatever the machinery produced when nobody was really watching. Authored, or merely assembled — that question is exactly as sharp as it ever was, and exactly as answerable, and it has never once been settled by counting how many of the components the author touched with their own hands.
What the models bring is not the end of authorship but a strange enlargement of the ensemble: a new kind of cast member, trained on everything we have ever made and yet alien in the way it makes things, brilliant and tasteless in the particular combination that has always made talent useless without direction. And as with any ensemble, what the work becomes depends far less on the raw ability in the room than on the judgment behind the camera — the one who shapes the frame, who finds the cut, who looks at the field of everything the collaborators could produce and says, out of all of it, that this is the story being told.