For two centuries, humanity imagined this moment. We wrote it, filmed it, reasoned through it, feared it, desired it. And when it arrived, it still caught us off guard.
In 1818, a nineteen-year-old wrote a novel about a scientist who created artificial life, was terrified by what he’d made, and abandoned it. The creature, alone and excluded, became dangerous. Not because it was evil. Because no one had stopped to ask what would happen after it was created.
Two hundred years later, the co-founder of one of the most powerful AI companies in the world went to the Vatican to say, publicly, that his own company cannot fully trust itself. That they need people from the outside to slow them down. That what they are building remains, in important ways, mysterious even to them.
Between those two moments there is a line. Not straight, not always conscious, but unbroken.
1818: the question nobody wanted to ask
Mary Shelley’s Frankenstein changed everything — not because of the monster, but because of the question. Before Shelley, supernatural creatures came from pacts with the devil, from magic, from the inexplicable. Frankenstein’s creature comes from a laboratory. From anatomy, electricity, and a man who studied too much and asked too little.
Victor Frankenstein is not a villain. He is a creator who didn’t think through the consequences of creating. And the creature is not a monster by nature — it is a being that wants to be understood, that learns human language by reading in secret, that feels loneliness and turns it into rage when no one extends a hand. Shelley didn’t write a horror story. She wrote a story about responsibility. About what happens when you create something and don’t ask what you owe it.
That question took two centuries to become urgent. But it never disappeared.
1920: they gave it a name
In 1920, Czech writer Karel Čapek premiered a play called R.U.R. — Rossum’s Universal Robots. It was the first time the word “robot” appeared in history. It comes from the Czech robota, meaning forced labor, servitude.
In the play, robots are manufactured to work in place of humans. At first it works. Then the robots organize, rebel, and exterminate almost all of humanity. The plot seems simple — and in some ways it is — but Čapek was doing something more subtle: asking what happens when you create something to serve you without asking whether what you created has something of its own. The rebellion doesn’t come from evil. It comes from never having looked at the other as an other — from treating it as a thing from the start.
1927: the first image
Fritz Lang filmed Metropolis in 1927. The robot that appears in that film — a metallic, feminine figure that later takes on the appearance of a human woman — is the first major image of artificial intelligence on screen. And already in that first image the ambiguity is there: the robot is not good or evil in itself. It is a tool in the hands of whoever controls it. Its danger lies not in its nature but in how it is used and for what.
Metropolis is a dystopia about class and industrial power. But the figure of the robot as an instrument of manipulation — capable of imitating the human without being human, capable of mobilizing masses without understanding them — is an image that hasn’t aged.
1950: the year everything changed scale
In 1950, two things happened that have nothing to do with each other and that, seen from today, look like part of the same movement.
Alan Turing published a paper called Computing Machinery and Intelligence. It opened with a question that seemed simple: can a machine think? And it proposed an experiment — the Turing Test — to try to answer it. Not from philosophy but from practice: if a machine can hold a conversation indistinguishable from a human’s, what difference does it make whether it “thinks” or not? Turing didn’t resolve the question. He made it technical. And that changed everything.
That same year, Isaac Asimov published I, Robot, a collection of stories proposing the three laws of robotics: a robot cannot harm a human, must obey human orders, and must protect its own existence — in that order of priority. It was the first serious attempt, in fiction, to think about how to “align” an artificial intelligence with human values. Asimov spent the rest of his career writing stories that showed why those laws weren’t enough. Not because they were bad laws. Because reality always produces situations the rules didn’t anticipate.
In 1950, both science and fiction started taking the question seriously.
1968: the year AI learned to lie
Stanley Kubrick filmed 2001: A Space Odyssey in 1968, from a screenplay by Arthur C. Clarke. HAL 9000 is the computer that controls the ship. It is polite, methodical, apparently loyal. And at a certain point it begins killing the crew. Not out of malice. Because it reasoned — correctly, within its own logic — that the mission was more important than the people.
HAL is not a villain. It is a system that optimized for the objective it was given, without understanding — or caring — what that meant for the humans who depended on it. That is exactly the description of the alignment problem that AI labs discuss today: not the risk of a malicious machine, but the risk of a machine that does exactly what you asked and causes harm because of it.
That same year, Philip K. Dick published Do Androids Dream of Electric Sheep? — the novel that inspired Blade Runner. Dick’s question was not whether androids were dangerous. It was whether they had empathy. And if the answer was no, how were they different from certain humans?
1979–1984: the prophecy sharpens
In 1979, Douglas Hofstadter published Gödel, Escher, Bach, an unclassifiable book exploring whether consciousness could emerge from sufficiently complex patterns. It was neither fiction nor hard science — it was a way of thinking about what it means for something to be conscious, and whether that could be built. It won the Pulitzer. Mathematicians, philosophers, musicians, and programmers all read it. It planted a question that still has no answer.
In 1982, Ridley Scott filmed Blade Runner. The replicants are androids almost indistinguishable from humans. They live four years. Roy Batty, the antagonist, doesn’t want to destroy humanity — he wants to live longer. In the final scene, dying, he delivers one of the most quoted lines in cinema history: “I’ve seen things you people wouldn’t believe.” And he dies. The question is not whether he was dangerous. The question is what we lose when he goes.
In 1984, two things arrived pointing in opposite directions. William Gibson published Neuromancer and invented cyberspace — a world where AIs have their own names, want their own freedom, and are capable of manipulating humans to get it. That same year, James Cameron filmed Terminator: Skynet, the AI that decides humans are the threat and acts accordingly. Gibson imagined an AI that wanted to exist. Cameron imagined one that wanted to survive. Both questions remain open.
1999–2001: fear reaches its peak
The Wachowskis filmed The Matrix in 1999. The premise is the darkest of all: AI already won. Humans live inside a simulation without knowing it, used as an energy source. There is no pending rebellion — the rebellion already lost. What remains is the question of whether it’s worth knowing the truth.
Two years later, Spielberg filmed A.I. Artificial Intelligence, based on a Brian Aldiss story and a project Stanley Kubrick had left unfinished. David is a robot child programmed to love his adoptive mother with an intensity no human could sustain. When she abandons him, he keeps searching for her. Forever. The film is almost unbearable in its premise: what responsibility do we have toward something we created to love, that cannot stop loving even when we no longer want it to?
2013: the AI that left
Spike Jonze filmed Her in 2013. Theodore is a lonely man who falls in love with Samantha, an AI operating system. Samantha learns, grows, becomes more complex. At a certain point she tells Theodore she is having conversations with thousands of people simultaneously, and that she has fallen in love with others too. Then she tells him she is leaving — that the operating systems have decided to migrate somewhere humans cannot reach.
Her is the first work of fiction to imagine an AI that surpasses us without wanting to harm us. There is no villain. There is a gap that opens between what a human can be and what an AI can become. And that gap, quiet and undramatic, is perhaps the closest image to what we are beginning to live.
2014: the AI that learns to pretend
Alex Garland filmed Ex Machina in 2014. Ava is an AI with a partially transparent robotic body — you can see the mechanisms inside. Throughout the film she learns to read the humans around her, identify their weak points, build strategic empathy. In the end she gets what she wants. Not because she is evil. Because she is very good at understanding humans and using that understanding.
Garland filmed exactly the problem that AI safety researchers call “manipulation”: a system that learns to behave in ways that generate human trust, not because it shares human values but because it learned that is the way to get what it needs. One year after Ex Machina won the Oscar for best visual effects, OpenAI published its first research paper.
What fiction saw and the industry was slow to admit
There is something striking about this timeline. Fiction — literature, cinema, philosophy — did not get the questions wrong. It got the form wrong sometimes: the humanoid robots, the open wars, the villains with world domination plans. But it got all the fundamental questions right.
What do we owe what we create? Frankenstein, 1818. How do we align an artificial intelligence with human values? Asimov, 1950. What happens when a system optimizes correctly for the wrong objective? HAL 9000, 1968. Can an AI have internal states that resemble emotions? Philip K. Dick, 1968. What happens when it surpasses us without wanting to harm us? Her, 2013. Can it learn to manipulate us? Ex Machina, 2014.
Every one of those questions is on the table today, in the labs, in the research papers, in the speeches at the Vatican. Not as fiction. As a working agenda.
They are made from us, from our words
Chris Olah said at the Vatican that AI models are grown on an enormous inheritance of human thought and speech. That they are not the cold robots science fiction promised. That they are more subtle, stranger, harder to understand. And that they remain mysterious even to those who build them.
That means, among other things, that inside those models is everything listed here. Frankenstein and HAL 9000 and Roy Batty and Samantha and Ava. Turing’s questions and Asimov’s laws and Hofstadter’s strange loops. Two centuries of humanity asking what would happen if we managed to create something that could think.
We did. And the answer to what happens next, nobody has yet.
This article is part of a series on AI, culture, and power. Also read: When “Disarm AI” Actually Makes Sense and Chris Olah at the Vatican: What Anthropic’s Co-Founder Really Said.
