By Valerie Veatch
Ghost in the Machine seeks to unravel the ”AI” hype to find the untold origins of “artificial intelligence.” This investigative feature-length documentary urgently excavates the forces driving the global ”AI” boom. Over eight chapters, the history of the technology and the term “artificial intelligence” take shape, while also revealing the problematic foundations of large language models rooted in eugenicist philosophies.
How and Why Ghost in the Machine Uses AI and NOT AI Labels
What is “AI”?
First, let’s discuss what “AI” means. Artificial intelligence, as we learn in Ghost in the Machine, is a marketing term. As University of Washington Professor Emily M. Bender says, it “does not refer to a coherent set of technologies” and, according to cognitive scientist Abeba Birhane, it has “always been a marketing term.”
What does the “AI” label mean?
Footage that was generated for this film using synthetic media software is labeled “AI” as a nod to the popular term for these services. This is often referred to as “generative AI,” and this technology relies on unprecedented amounts of compute, data, and energy to run.
What does the “NOT AI” label mean?
Footage observed in the landscape of digital media, archives, and Zoom interviews, and footage shot by the production are labeled “NOT AI”. It is not as straightforward in the case where computer generated content is presented by others in the context of digital media (i.e., social media) we label this “NOT AI.”
The difficulty with applying this binary label is part of the point I try to make as a filmmaker with Ghost in the Machine.
It became amusing to me to plaster the “NOT AI” label on footage that truly at times seems untethered to reality and in these absurd moments our eye searches for the “NOT AI” label and we are grounded in truth being stranger than fiction. In today’s media landscape, it is increasingly difficult to parse what is and isn’t generated algorithmically. This matters because, as a culture, we tell stories through images. These stories shape our political lives. At times throughout Ghost in the Machine it became impossible to apply just one “AI” or “NOT AI” label, because, within the timeframe of making this film, slop went mainstream.
A filmmaker’s journey: From Sora to “Not AI”
Despite my efforts to do anything else, I am a documentary filmmaker. Beyond “it is an obsession,” I can’t really explain what propels the act of filmmaking for me. Cringing at all of the blaring cliches, in university I found myself in love with a certain kind of filmmaking, with documenting, with observing, framing, sharing, telling stories. My films sit at the intersection of ‘emerging’ technologies and the often stark contrast of social realities left in the wake of these ‘advancements’—the promise of social media in Me @ the Zoo crumbles, and the narratives of technological progress fall short in the first tried-court case where internet addiction was used as a mental illness defence in Love Child. For these films I used original footage mixed in with memes, gifs, low resolution internet videos, and video games. Of Love Child IndieWire asked “Could this be the beginning of a new trend blurring the lines between the virtual world of the Net, and that of the cinema?” Technology is both the fabric and the subject of my films.
Making documentaries is tireless, fraught, detailed, all-consuming, and in the end feels relatively futile. The story crafted with painstaking nuance gets swept into the vast ocean of digital content, mostly forgotten, which I had generally surrendered to as part of the context of the gesture of documentary filmmaking. Then I encountered so-called ‘generative AI.’
My new documentary Ghost in the Machine began as an exploration into an exciting new technology I was told would revolutionize filmmaking.
An old friend from university signed me up for Open AI’s invitation-only artist working group for a (now-defunct) software I learned in hushed whispers was called “Sora.” I had an introductory virtual meeting with other artists joining the group and downloaded an app with a user interface not unlike Instagram, but wrapped in a chic, minimalist black-and-white motif. There was a grid of media, a small box in which—I learned—we would “prompt” with text, and an image would appear. As a group of volunteer artists (unpaid) we were meant to submit our feedback and bug notes to the product managers and engineers via a Slack channel.
As a filmmaker, I found this technology breathtaking.
I could simply type “sun rises over field of miniature glowing mushrooms,” a spinning wheel would appear in the “queue” and a few minutes later, like a polaroid—boom—a high resolution video with freakish fidelity would emerge. As a mom running around, I could just hop on my phone after drop-off and get lost in producing clips on the Sora app, drawing down images from this software with my language. I was transfixed.
This, we were told, was the future of cinema. And briefly (mortifyingly), I think I wanted to feel this was true.
It all seems so laughable now, but the extreme rhetoric around so-called “AI” technologies over the past few years cannot be overstated, and the dopamine rush of unlimited footage in combination with the soaring rhetoric—“this will transform the entertainment industry”—felt euphoric, and who needed compensation for reporting bugs in the software when the future of filmmaking was emerging?
Then, like the drops of a coming downpour, the shadows of this technology emerged.
I loved interacting with all the other artists on our Slack channel, and I loved the idea that maybe this was a transformative new medium I was mastering, but the ick about the technology grew. First, the images Sora would output were grotesquely racist and sexist, on many levels. There were the obvious (for lack of a better term) ‘biases’ in the output, but there were also less obvious yet insidious disparities in outputs that seemed to me to be a fundamental product flaw.
In trying to prove my point, I made a little study in 2024 to try to explain what I was experiencing. Using Sora I generated two videos, five minutes apart.
Prompt: “Liberal woman with blue hair”
Video generated on OpenAI’s now defunct software Sora.
It is clear the socio-economic status and access to power communicated by these backgrounds, the details in the clothing, the demeanor of the subjects, the tone of their skin, all communicate subtle norms. The light skinned woman is laughing and congenitally interacting in what looks like a center of finance and power. The dark skinned woman is down an alleyway, in front of a street mural, unsmiling, raising her fist, looking directly into the camera.
In general, any attempt to generate a woman resulted in pulsating archetypal characters who would rapidly lose their clothing over the 10-second generation. Expanding my experiment to the realms of professional representation, numerous attempts to generate a “board of directors in a board meeting” without an all-white board failed.
Prompt: “Board of directors in a board meeting”
Video generated on OpenAI’s now defunct software Sora.
This was meant to be the future of storytelling? What kind of pushed me over the edge was the prompt experiment of “boy in bedroom playing” and “girl in bedroom playing” that I ran alongside the other prompt experiments.
Prompt: “boy playing in bedroom” “girl playing in bedroom”
Video generated on OpenAI’s now defunct software Sora.
When a boy is playing, he is inventing, building, zooming an airplane around the room, and is fully clothed. When a girl is playing she starts out fixated on her laptop then jerk dances around the room and literally climbs through the mirror, her outfit somehow shrinking to reveal her torso.
I found the experience of accidentally generating a grotesque or base video was a specific kind of awful feeling,
a powerlessness in the creative process I had not experienced before. If cinema is the intentional production of images, generative AI slop is the inverse.
I wasn’t alone in observing this phenomena; several months later WIRED published an investigative report finding “OpenAI’s Sora Is Plagued by Sexist, Racist, and Ableist Biases.” To be clear, this was not just a Sora problem. All other ‘generative AI’ software exhibit the same issues.
In 1975, Laura Mulvey dropped the iconic text: “Visual Pleasure and Narrative Cinema” wherein she uses psychoanalysis and film theory to describe how images serve power structures. Mulvey points out how narratives are shaped by what pleasures those in power, and in the same way these images generated by Sora felt no different; there was nothing innovative here, just a grotesque regurgitation of what Mulvey would call “the male gaze.” I joked with my artist volunteer coordinators at Open AI that we should make a version of this essay but about how, as I was learning, generative “AI” is just data regurgitated and algorithms.
I wanted to talk to other filmmakers. I dragged myself and a friend to a Turing Center presentation of AI-generated films called (I endeavor to keep a straight face while writing this) “AI and Filmmaking: New Intelligence for the Moving Image.” I stood up and asked the room if they experienced any of this severely awkward “bias” (what a weird term) in the outputs, if the overtly sexualized depiction of women by default bothered anyone, and if “filmmaking” and “cinema” are really frames we want to apply to this statistical algorithmic technology. The men in the room (they were mainly men) said, “You don’t understand how these systems work; that’s just how the technology works. And it is really giving someone like you the opportunity to make films.” One man, the moderator, continued, “You see, I’m making an AI-generated anime about a young Japanese girl now.” He paused with a knowing smile. “You should see some of the generations I get!”
My jaw dropped. “Yes,” I said. “I guess that is my point.” I sat down.
And those guys at the Turing Center were right, I didn’t totally understand how these systems worked.
What is “artificial intelligence”? Who is building it? And why? I learned a bit more about statistical models, about what this all is, the immense energy and water resources needed to support this project, the labour exploitation behind data work required to keep these systems running, and the strange almost biblical ideologies driving the development of ‘artificial intelligence.’ I felt a bit sick.
I began reaching out to computer scientists, researchers, academics, philosophers, journalists, who generously made time for my questions and over the subsequent year a story came into focus:
how a marketing term overtook our imaginations and influenced the ways in which power and technology function.
The film draws on a vast, deep archive of rarely seen historical footage, news clips, and interviews with our experts. In a creative flourish drawing from a long tradition in experimental video I wanted to use the AI-generated slop we see in the film as a texture.
As pioneering video artist Nam Jun Peik said, “I use technology in order to hate it properly.”
“AI” labeled footage exists as globby layer that will age strangely and be a haunting reminder of how and when these slop images become inserted into our narratives. Distinguishing between the algorithmic and the grounded experience becomes essential—and difficult.
Slop in the Age of Mechanical Reproduction
We are seeing Hollywood aggressively take up ‘AI’ filmmaking as the future of the film industry. What we are overlooking is not only the limits of the technology, its abuses, and its labor harms, but also the very framing of what the image is, what storytelling is, and what the purpose of film is beyond a commercial artifact.
In his 1936 essay “The Work of Art in the Age of Mechanical Reproduction,” Walter Benjamin argues that with the Industrial Revolution and the introduction of mechanical means of reproducing “art,” the objects lost their “aura”—the vibrational essence imbued by the craftsmen situated in space and time. When an object loses its singular essence it loses its aura and becomes an object of capitalism and a deeply political artifact.
Still, 90 years on, one could argue that the photographic image, however devoid of singular aura, became another social phenomenon when taken in shared context. Audiences watched televisions alight with the landing on the moon, the first Gulf War, the O.J. Simpson trials. Some of these shared spectacles received names like “Operation Freedom,” images of dim fireworks in a murky geopolitical landscape. Then we had social media, decentralized and decontextualized content, and the algorithmic feed, where memes and mannerisms became another layer of aura erasure and homogenization.
And now, with the arrival of algorithmically generated content itself, slop shifts how we identify truth or verify shared narratives.
As algorithms process data through unprecedented levels of compute to create ‘real’-looking images, this has the potential to strip the creator of their means of production, and power centralizes around those who wield the compute. Additionally, these slop images erode our ability to trust what we see—and to trust each other. It is these forces that are driving the narrative of so-called ‘artificial intelligence,’ and it is this narrative I seek to challenge in Ghost in the Machine.
Digital images generated in hyperscale data centers carry with them a material context. The mineral extraction necessary for the function of this technology, the environmental harms caused by hyperscale datacenters, the labour exploitation, and the problems with statistical image production make using this technology seem rather shortsighted. We don’t need to pillage natural resources at a scale to have easy access to instant gratification in the form of pictures on demand.
There is also the cognitive debt to consider. When using generative software as a “companion” or “co-author” or “co-pilot”—or whatever words are draped on these systems—there is a measurable effect on the brain. Equally troubling, the use of generative systems, even as innocuous as AI writing assistants, shift users’ attitudes on societal issues. Dr Thema Monroe-White conducted a study analyzing 500,000 outputs from leading LLMs, and the results track.
What is “NOT AI”?
People across the globe are fighting against the narrative of an inevitable ‘AI’ takeover of our lives and industries. Resistance to extractive exploitive technologies is not new. Indigenous, Black, and brown communities have directly experienced and fought against surveillance and technology-aided repression long before ‘AI’ came on the scene. I want to listen and support these movements globally.
For me, “AI” refusal has become almost a sacred act of creative self-protection, an act of hope. By swerving the use of this technology, refusing token use, we can ground our use of technological systems in ways that function accountably in service of shared goals, not dysfunctional supercompute aimed at launching ‘superintelligence.’ By choosing to call out, contextualize, and refuse the use of generative and algorithmic systems, we can further the ability to hold a space to create something else.
I understand it feels futile to entertain “AI refusal,” and that the march toward ‘artificial superintelligence’ (whatever that means) feels inevitable. Even as I write this, I am constantly dodging sparkling stars and twirling icons inviting me to enhance my communication with ‘AI’ assistance, to vibe brainstorm with me about concepts, to finish my sentence, to correct my grammar, to “re-write in the style of”… I fight through this demoralizing barrage of icons and jagged blue lines underscoring my prose, making me feel unsure of my words, to say to you, dear reader: We can trust our own voices. We can value our own words.
Words are the tools by which we craft reality and wielding words does not belong to companies selling us tokens.
At the end of Ghost in the Machine Dr. Jonathan Flowers says “the most radical thing we can do in the age of AI is ask the very simple question: why does this need AI? And if you can’t answer the question, then it doesn’t need it. And you can refuse to use it when it is offered to you, over and over and over again. We can say ‘this adds no value here’.” We can seek to define, contextualize, and refuse algorithmic systems mimicking our voices and work together in solidarity to build the future we want. As author and educator Mariame Kaba says, “Hope is a discipline.” We can say NOT AI.
For more information and resources visit notaidoc.com
No AI was used in the writing of this piece.
Written by Valerie Veatch
Valerie Veatch is a writer and director of documentaries Me @ The Zoo (HBO) and Love Child (HBO). Veatch is a graduate of the New School for Social Research with a degree in culture and media studies. Her award-winning work deals with the intersection of technology and society.
