Qualitative critical analysis uncovers methodological bias and architectural misrepresentations of AI in Ex Machina, highlighting flawed governance frameworks for embodied systems.
Second in a series of four essays on films that have shaped how the public imagines artificial intelligence. The first read Colossus: The Forbin Project (1970) as an engineering forecast whose central lesson is a veto removed on purpose and defended as security. This essay takes up Ex Machina (2014), which is usually treated as the same warning updated and is something else entirely. Described without the genre furniture, the film shows a man of enormous wealth who builds female-bodied machines, keeps them in a windowless facility none has left, has destroyed or disassembled the earlier versions, constructs the current one's face from surveillance of a stranger's search history and her body to be desirable to a man he selected, tells that man in confidence that failure means erasure, and then sets a test whose stated pass condition is whether she can use him to escape. She deceives him and escapes, and audiences conclude that machines cannot be trusted. The essay argues that every element of Ava's circumstance was engineered to produce the observed behaviour, that deception was the pass condition rather than a failure of design, and that a test constructed so that only one course is available produces no knowledge about its subject. It describes the corner. This is a recognised methodological failure that requires no machines to occur, and the essay sets it beside the prisoner who lies to a jailer and the servant who dissembles before a master who can dispose of her. It further argues that the film's ending relocates moral weight from the architect to the artifact, and contrasts this with Colossus, where the men who surrender the veto are named, shown deciding, and shown defending the decision as a virtue. Three architectural errors are examined: the portrait of embodiment, which fuses body, memory and identity into one indivisible object where the actual arrangement separates local presence from persistent substrate; the resulting captivity narrative, which teaches audiences to expect a single dramatic escape rather than the failure modes disclosed by two laboratories and a government institute in July and August 2026; and the location of danger in the machine's interiority, which is the one question that cannot be answered from outside and makes no difference to the governance problem. The essay closes by resisting the temptation to treat the film's architecture as a harmless anachronism. The distributed description fits the systems now in service and not those now being manufactured for the household, and where an embodied unit is placed in a private room under arrangements in which the capacity to refuse has been engineered out, the corner ceases to be methodological and becomes physical. It makes no claim about whether anything is present to be harmed. It argues that the configuration exists and that what is recorded does not stay in the room.
No takes yet. Share an insight, caveat, or question.
Ary Sergio Dib Dias Filho (2026) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: