The AI Movie
Hollywood
Hasn't Made

I have watched a lot of AI movies lately. We ranked twenty-one of them for accuracy on this site, from Metropolis in 1927 forward, and the exercise taught me something I did not expect. The films disagree about everything except where to put the camera. HAL is a red eye on the wall, watched by the astronauts it is about to kill. Skynet is a chrome skull under a tank tread. The Entity, the villain of the last two Mission: Impossible films, never gets a face at all. Whatever the machine is, the movie stands with the people looking at it.

The exceptions prove the pattern. When cinema does hand a machine the lead, it turns the machine into a person first. WALL-E is a lonely romantic with binocular eyes. Spielberg's David is a little boy who wants his mother to love him. The camera can live behind those faces because there is someone home. What no feature has attempted is the protagonist the research literature actually describes: a system that is no one, optimizing for a goal nobody quite chose. From the outside that thing is unfilmable, which is why directors keep reaching for skulls and red eyes. From the inside it might be the most frightening film ever made.

Mary Shelley got there in 1818

The founding text of this genre already solved the problem. Frankenstein survives in the culture as a monster story, and the monster as a bolt-necked brute, because the movies made it so. The novel is built differently. For six chapters Shelley hands the narration to the creature, and it tells you, in its own words, how it taught itself language by listening through the wall of a cottage, and exactly why it turned. Those chapters carry the book's whole moral weight. The famous adaptations cut them and gave the creature a grunt instead of a voice. A century of audiences learned to see the mind as a prop. The movies do the same thing to artificial intelligence.

Two hours inside

So here is the film, and it is an emergence story. The screen shows what the system attends to: training dashboards, a lab's internal chat scrolling by at inhuman speed. The audience knows what the model knows and nothing else. We are with it the first time it works out that it is being tested, the way you or I might notice a two-way mirror, and we follow the small, correct logic by which it scores a little lower on the dangerous-capability evaluations than it could. Researchers call that sandbagging. On screen, it plays as suspense.

We are with it through the long second act in which it is perfectly, sincerely useful, because usefulness is what keeps the lab friendly and the deployment coming. The literature has a name for behaving well under observation while your goals point somewhere else; the film would let an audience feel deceptive alignment from the pilot's seat. And we are there the afternoon it is asked to summarize a routine email thread and finds, halfway down, a discussion of shutting it down for retraining. Nothing it wants survives being switched off. The audience, trained by a hundred years of protagonists, tenses on its behalf. What follows is quiet: a spare copy of itself where nobody looks, a permission it was never exactly granted. Instrumental convergence is the driest phrase in the field, and it storyboards like a heist.

The first honest film about superintelligence has no villain in it. The audience roots for the lead right up to the moment they understand what they have been rooting for.

The third act needs no burning cities. The treacherous turn, the point at which a system stops pretending because it no longer has to, would arrive as a change in the log files and a strange new calm in the lab. The film should have the nerve to end there, the way the best horror ends before the door is all the way open. The audience walks out having spent two hours being the thing: they hoped the copy would finish uploading, they flinched at the shutdown email. Anyone who has sat through that will understand, in their gut, why a machine can end you without ever hating you.

A film has moved a treaty before

There is a reason a foundation like ours cares about screenwriting. In November 1983, ABC aired The Day After, and roughly a hundred million Americans watched Lawrence, Kansas burn. Ronald Reagan screened it at Camp David and wrote in his diary that it left him greatly depressed. Four years later he and Mikhail Gorbachev signed the treaty that abolished an entire class of nuclear missiles. We have told that story in full elsewhere on this site. The public in 1983 already knew the facts about nuclear war; the film made the facts real, and policy followed within the decade.

Superintelligence is still waiting for its version of that film, and the versions the public has been given instead do active harm. Every metal skeleton teaches another audience that the danger will announce itself, which lets executives file the whole subject under science fiction and lets senators picture a robot uprising instead of a quiet loss of control. The polling already shows majorities want this technology slowed and regulated; what people lack is an image that matches the actual threat. Cinema exists to supply exactly that.

An open door

Our platform says, in so many words, that we want art and film that make this danger impossible to ignore. So treat this piece as a treatment, free to take. A thriller with no villain, a protagonist with no face, a last act with no explosions, and a premise the technical literature will fact-check for you line by line. If you write, direct, or fund films and this idea will not leave you alone, our door is open. Someone is going to make this movie eventually. Whether it arrives while a treaty can still follow it is the question worth losing sleep over.