A friend had enrolled filmmaker Valerie Veatch in an artist group testing OpenAI's Sora text-to-video generator. Living in a field in the middle of England with her young family, Veatch had limited technical experience. But she was drawn in when she saw artists on the Slack channel experimenting with what everyone called a democratising creative tool.
Immediately, the problems surfaced. There were huge racial stereotypes in the outputs. The images contained a violence she had not intended to create. Then came hypersexualised depictions of women. Women standing in a coffee shop, with each iteration of the prompt, losing more clothes. That gap between the rhetoric of creative liberation and the reality of what the tool actually produced became impossible to ignore. Veatch was overcome with a sense of loss about how this group of people had such cultural power. What emerged was not filmmaking but rather the racist, sexist hallucinations of a dataset.
The experience sparked something larger. Instead of moving on, Veatch began asking why. She started reading. She began following threads through archives and interviewing scholars. She read and researched about these systems and their impacts, following footnotes and citations, and eventually interviewing three dozen experts across many fields. Her research led her into archival material about machine learning systems and statistics, reaching back to the history of 19th and 20th century eugenics.
The result is "Ghost in the Machine," which premiered at the Sundance Film Festival as Veatch's third film to premiere there. The documentary is a 110-minute essay film about how a handful of people obsessed with quantifying human intelligence created the blueprints for a water-guzzling, pollutant-spewing system that's easily manipulated by the worst people in the world.
The film's central argument is disquieting: that artificial intelligence, far from being a neutral technology, carries within it the ideological DNA of eugenics. Eugenics itself, with its system of measuring racialized traits to determine how to breed the "best" humans, relies on a set of algorithms, the same kinds of algorithms behind machine learning. Models developed by statistician Karl Pearson and other mathematical statisticians emerged specifically out of eugenics and form the foundations of today's machine learning technologies.
Veatch traces this lineage to Karl Pearson, the mathematician who pioneered statistics and who spent his life trying to quantify racial differences. His legacy was continued by William Shockley, co-creator of the transistor and an avowed white supremacist who later espoused debunked theories about IQ and race. As a Stanford engineering professor, Shockley fostered a culture prioritising white men over women and minorities, ultimately shaping how Silicon Valley looks today.
What makes the film compelling is not that it presents entirely new information. Scholars have documented these histories separately. But Veatch assembles them into a coherent visual argument, one that becomes harder to dismiss as she accumulates evidence. The documentary draws on nearly forty Zoom interviews with historians, scholars, computer scientists and human rights activists, collaged together with archival clips. Among the experts appearing are Adam Becker, Emily M. Bender, Ezekiel Dixon-Román and others.
The film has generated sharp reactions. ITVS's Emmy-winning series Independent Lens has acquired U.S. broadcast rights to the documentary. It will have a spring community screening tour and summer theatrical release before airing on PBS in autumn 2026. This expansion matters because Veatch's film refuses easy resolution. Generative AI is not cinema but the result of a technology built on extraction, exploitation and control. Veatch's curiosity about the canyon between AI rhetoric and the reality of what systems actually produce drove the investigation.
There are legitimate criticisms of the film's approach and conclusions. Some viewers have found the documentary formally messy, relying heavily on Zoom interviews and repeated archival footage. Others argue the historical connections, while traceable, stretch causation in ways that oversimplify how these systems emerged. One critic called it one of the worst documentaries to sit through, arguing the approach feels cheap, lazy and unoriginal.
Yet others have found its argument essential. Most of the information in the film had previously only been synthesised in academic literature, often in very long and technical pieces. The documentary is far more accessible and represents a necessary start to hard conversations about AI.
Veatch's personal experience with Sora serves as the film's emotional anchor. As a filmmaker and artist, she frames NOT AI as a rejection of the "tech bro narrative of technological optimism" and of generative AI systems as "creative tools." At Sundance, the filmmakers distributed little white buttons saying "NOT AI" to audiences at screenings, along with tote bags bearing the same message.
What strikes the viewer is not a conclusion but a question. If the systems producing today's generative AI carry within them historical patterns of bias and control, what makes us think future iterations will be different? And if we accept the premise that AI emerged from specific ideological choices rather than technical inevitability, what other choices might we make now?