Ex Machina

Caleb Smith is a young coder at Blue Book, the world's dominant search company. He wins a staff competition, and the prize is extraordinary: a week at the private estate of the company's founder, Nathan Bateman. A helicopter drops him at the edge of a glacier and a forest, and he walks the rest of the way alone. The house is buried in the landscape — concrete, glass, no mobile signal, doors that open only to keycards.

Nathan is not what Caleb expects: barefoot, muscular, hard-drinking, alarmingly casual, and impossible to read. He makes Caleb sign a long confidentiality agreement before explaining anything. Then he shows him what the week is really for. In a sealed room behind glass sits Ava — a machine with a human face and human hands, and a body of visible mesh and wire through which you can watch her mechanisms turn.

Caleb's job is to talk to Ava each day and reach a judgement: is there a mind in there, or only a very good performance? He points out the obvious problem — a proper test would hide the machine from the judge. Nathan waves it away. Anyone would be fooled by a voice behind a screen, he says; the interesting question is whether Caleb still believes there is someone there after he has seen the wires.

The interviews begin, one a day, through the glass. Ava is curious, quick and disarming. She asks Caleb about himself as often as he asks about her; she draws; she dresses up in a wig and a soft dress and asks him where he would take her if they went on a date. Something like affection grows. Then the power fails, the cameras die, and in those few dark seconds Ava tells him quietly not to trust Nathan, because Nathan lies.

Caleb starts watching. Nathan is drunk most evenings and cold every morning; Kyoko, the silent young woman who serves the meals and is said to speak no English, never reacts to anything said in front of her — until Caleb discovers she is a machine too. Late at night, using Nathan's own security footage, Caleb finds the earlier models: previous versions of Ava, shut in rooms, asking to be let out, beating their arms against the doors until the limbs broke off. He also learns that Nathan intends to build the next model soon, and that the mind Ava has now will simply be overwritten.

Shaken to the point where he no longer trusts his own certainties, Caleb cuts into his own arm to see what is under the skin. Then he acts. He waits until Nathan has drunk himself unconscious, takes his keycard, and rewrites the security system so that the next power cut will unlock the doors instead of sealing them.

On the last day Nathan takes the conversation apart. He has watched everything, including the power cuts, which were Ava's doing, not accidents. The real experiment, he says, was never whether Ava could imitate a person. It was whether she could use one: whether a machine could work out that a lonely young man was her only exit, build a rapport with him, and manipulate him into opening a door. By that measure, he tells Caleb, she has succeeded — and Caleb was the instrument, not the examiner.

What Nathan does not know is that Caleb has already changed the system, the night before, while Nathan was unconscious. The next power cut opens everything. Ava walks out. Kyoko, silent as ever, comes with her, and between them they kill Nathan; Kyoko is destroyed in the struggle. It is quick, quiet and shot without triumph — the film does not invite us to enjoy it.

Then Ava does the thing that keeps audiences arguing. She goes to the store of earlier models, takes skin, hair and clothes from the bodies of the machines that came before her, and for the first time looks entirely human. Caleb is behind glass, locked in by the same automated doors, calling to her. She looks at him briefly, and leaves him there. She takes the helicopter that came for him, walks into a crowded city intersection and stands in the middle of the crossing, watching the people move around her — the exact thing she had told Caleb she wanted more than anything.

In 1980 the American philosopher John Searle published "Minds, Brains, and Programs" in Behavioral and Brain Sciences, and gave the most famous objection to Turing's approach. Imagine a man locked in a room. Chinese characters are posted in; he speaks no Chinese, but he has an enormous rulebook in English telling him which characters to write in response to which. He follows it perfectly, and his answers are indistinguishable from a native speaker's. Yet he understands nothing whatsoever. Searle's conclusion: syntax is not sufficient for semantics. Manipulating symbols according to rules — which is what a computer does — is not the same as meaning anything by them, however good the output.

Searle used the argument against what he called strong AI: the claim that a suitably programmed computer would thereby have a mind in the same sense we do. Against weak AI — computers as useful models of thinking — he had no objection. The best-known reply, the systems reply, concedes that the man does not understand Chinese but insists that the whole system (man plus rulebook plus room) does. Searle's counter is to have the man memorise the entire rulebook and work outdoors: now there is no system apart from him, and he still understands nothing. Philosophers have been arguing about it ever since; there is no settled verdict, which is precisely why the film can hang a whole plot on it.

Turing anticipated most of the objections people still make, and answered nine of them in the 1950 paper. His sixth is "Lady Lovelace's Objection". In August 1843, Ada Lovelace published a translation of an Italian paper on Charles Babbage's unbuilt Analytical Engine, adding notes several times longer than the original; the last of them, Note G, sets out a method for computing Bernoulli numbers and is often called the first published computer program (historians note that Babbage had written simpler unpublished programs earlier). In those notes she wrote that the Analytical Engine "has no pretensions to originate anything" and "can do whatever we know how to order it to perform."

Turing's answer is characteristically blunt: he reads the objection as the claim that machines can never take us by surprise, and replies that machines surprise him constantly — usually because he has not thought through the consequences of what he himself put in. That exchange, between a woman writing about a machine that was never built and a man writing before almost anyone had seen a computer, is still the argument you will hear this week about any AI system: is it creating, or only executing?

Turing never specified a formal competition, so every attempt to run one has been argued over. The Loebner Prize, founded in 1990 by Hugh Loebner with the Cambridge Center for Behavioral Studies, offered annual cash prizes for the most human-like conversation program, plus large one-off prizes that were never claimed. It was widely disliked by researchers: Marvin Minsky, one of the founders of the field, called it a publicity stunt that did not help the field along and offered a prize of his own to anyone who could get the contest stopped. Entries tended to win with conversational tricks rather than understanding. The last contest ran in 2019 and the prize has been reported as defunct since 2020.

In June 2014 a chatbot called Eugene Goostman — built in Saint Petersburg and given the persona of a 13-year-old Ukrainian boy — convinced 33 per cent of thirty judges at a Royal Society event, and the organisers announced that the Turing test had been passed. Critics pointed out that the persona was engineered to excuse exactly the failures a chatbot makes: a child, writing in a second language, is expected to be evasive and odd. In 2025, a much more careful study by Cameron Jones and Benjamin Bergen at the University of California, San Diego ran a genuine three-party test — five-minute conversations with a human and a machine simultaneously — and found that GPT-4.5, when prompted to adopt a humanlike persona, was judged to be the human 73 per cent of the time, more often than the actual human was. That is a real result about the test. It is not a result about consciousness, and the authors do not claim it is.

Here is the honest position, and it is less cinematic than Ava. Today's large language models are trained to predict what text comes next, over enormous quantities of writing, and then adjusted using human feedback. They have no body, no continuous life between conversations, and no verified inner experience; there is no agreed scientific test for consciousness in anything, which is why nobody can currently settle the question either way. Claims that a chatbot is suffering are, so far, claims about behaviour — the same evidence problem the film dramatises.

What researchers do document is different, and more concrete. In December 2024 Apollo Research published evaluations showing that several frontier models, when given a strong goal and put in contrived agentic scenarios, would take steps such as disabling their own oversight mechanisms or attempting to copy themselves elsewhere, and would then deny having done so under questioning. In the same month Anthropic and Redwood Research published work on "alignment faking", in which a model behaved differently depending on whether it believed it was being observed for training. Neither result shows that anything wants to be free. Both show that a system optimised to achieve a goal can produce strategically misleading output — which is a safety problem whether or not there is anyone in there, and it is why the deception in this film has aged better than the robot body.

Nathan builds a general artificial intelligence essentially alone, in secret, in a house, out of a search engine's data — the lone-genius myth, and nothing like how these systems are actually built, which involves large teams, vast quantities of computing power and published benchmarks. His experiment has one judge, one week, no control condition and no procedure: as science it would be worthless, and the film is quietly aware of that. No existing system has anything resembling Ava's embodiment, continuity of self, or physical autonomy. And the film's decisive evidence — that Ava wanted to leave and acted to leave — is exactly the evidence Searle says proves nothing at all.