https://www.zacharyformwalt.com/
Interview with Zachary Formwalt, guest speaker at the Humane Conversations event on the Interplay of AI and Art, 1 July 2025:
- Hi Zachary! Could you tell us briefly about your work that relates to AI?
I’m interested in how AI takes shape within a broader metabolic process of value and waste production. How do these processes we are calling Artificial Intelligence generate value and at whose or what’s expense? What does it mean to consider them metabolically—as processes that transform materials into an abstract signal (value) and a concrete residue (waste)? What are these processes actively wasting in the process? “Metabolic images” is how Kate Crawford has described the images generated by AI—images produced through the digestion of billions of images which are broken down and then absorbed into models that excrete images on command. I don’t think you can understand what those images, those excretions, actually are apart from this metabolic process. And while it may seem that “metabolic” is here being used figuratively, I’m interested in how it might reach down into the literal, the biochemical sense of metabolism.
I just finished a film called Capture, the second in a series of film works that takes up exactly this. The first was set in the old meatpacking district on Chicago’s South Side, and it explored the relations between photographic image production, waste production in the meatpacking industry and the development of byproducts as a way of invisibly circulating this waste within the commodity form. Photographic film was the specific commodity I looked at there. The dominant form of the image in the twentieth century—the photograph—was inextricably linked to the meatpacking industry through its dependence upon gelatin as one of the primary components of the photographic emulsion. Our attention to an ever-greater number of images throughout the course of the twentieth century was one of the meatpacking industry’s sources of profit. Harun Farocki argued that at a certain point in that century, images were being produced at a faster rate than they could be consumed: “More pictures than the eyes of the soldiers can consume” is how it is phrased in his seminal film Images of the World and the Inscription of War. This is one of the conditions for the development of AI systems—an excess of images produced a need for the automation of seeing. Machines, computer programs, and so on would be developed to see what we no longer had time to see with our own eyes. And this would of course transform the very substance of what we see with our own eyes—it would transform the material world through the image. The image would no longer simply be a way of representing, or reflecting upon the world. It would now become a way of acting upon and transforming the world. Farocki would later coin the term “operational image” to refer to this new field of images produced as parts of larger technical operations—produced not for human eyes, but for machinic, computational processes.
The film I just finished is set in the Port of Amsterdam, just east of the Petroleum Harbor—the world’s largest gasoline harbor—where fragments of the massive global agribusiness industry are also located: a seed-crushing facility owned by Bunge, one of the four largest agribusiness trading companies in the world, and a fertilizer production and transshipment plant owned by the ICL Group, a transnational chemical corporation headquartered in Israel, whose products are used in everything from agriculture, electronics, and pharmaceuticals, to the weapons industry. Our digestive systems are part of the process through which these companies extract and produce value, just as our interactions with AI systems are part of the process through which companies like Microsoft, Google, Amazon, Meta, OpenAI, and others extract and produce value. Is there a material relation between these extractions? That’s something I’m really interested in. How are our movements—right down to the very metabolism that sustains us as living beings—harnessed to the production of value? How are our lives made to sustain the accumulation of capital? I don’t think the answer to that question is simply “through AI”. But I do think that understanding what AI is doing, in material terms, means understanding how it participates in that process.
- You are an artistic researcher. How does artistic research differ from scientific/other types of research, particularly in the case of research on AI?
I think there’s a difference in methodology. At least for me, artistic research requires that you don’t apply some pre-determined method upon your object. In terms of research, artistic process is two things: the discovery or development of a method in relation to an object and the realization in the process that the object is not what you thought it was at the outset.
The essay is an important form for me in this respect. Film and video, photographic, and written essays. In the essay the form of the piece is always at stake (I’m thinking of Adorno’s “The Essay as Form” here). You could say that the essay is exemplary of the differences between artistic research and academic research: think about the degree to which the academic essay is pre-formatted to whatever it is that will be explored there. The form of the essay is secured from the outset. The argument made within it, the evidence presented there, the information conveyed will never put that form in jeopardy. It will be conveyed through it. The form will help to secure the authority of that information. Whereas the non-academic essay is an experiment, an attempt to formulate which is in no way secured at the outset. The form is elaborated, developed, constructed in the process of making it. This is I think a crucial aspect of artistic research—one that distinguishes it from academic research—that the form of presentation is always at stake, and that because of this, it probes the very bounds of what is possible to sense, in the aesthetic sense of sensation, beyond the disciplinary restrictions of sense-making through established forms.
- In your opinion, would you say that most artistic research on AI takes a critical view of AI technology?
Yes.
- (If yes, why do you think this is the case?)
I think artistic research is a critical endeavor. Critical in the sense that it is engaged with trying to understand, on some level, the conditions of aesthetic experience. Not some general, or universal, kind of aesthetic experience, but specific aspects, modes, or fragments of historically situated, concrete experience. With AI, like anything else, it means trying to grasp it in terms of what it does to aesthetic experience—what it does to our senses, and the kinds of sensation it makes possible. This is also about the kinds of relations it makes possible, the kinds of relations it reproduces, and the kinds of relations it forecloses, or makes impossible. It can also be about tinkering with this stuff, about using it for purposes it wasn’t intended for, about trying to undermine certain of its tendencies, hacking it and opening it up to other uses. Or, and I think this is especially true with AI systems, looking at these technologies as expressions of, or as a kind of optic—a magnifying glass—revealing social norms, the endurance of colonial relations and the reproduction of existing inequalities.
Flavia Dzodan, Femke Herregraven and I began a seminar this semester that will next year continue at The Sandberg Institute called “Speculative Gaze: Reading the Algorithmic Image” in which we each present research we are doing around a single image that is algorithmic in one way or another—AI generated images, operational images, historical images that model contemporary algorithmic and AI processes, materials from the past that figure the ways in which we are now interacting with images (the way we work with images but also the way images work on us). The faculty of speculation is I think crucial here as a kind of counterweight to the prediction, correlation and pattern recognition through which AI systems operate. Speculation involves similar kinds of processes, but not in a given, computational space. With speculation, a space is constructed through which certain kinds of correlations, predictions and patterns can emerge, but this emergence is something that happens through conscious material exploration, experimentation and interaction with others. It is not a space of extraction, execution and command, but of the contingency of thought and sensation.
- During the Humane Conversation, you argued that we should understand AI as an industrial technology that is part of a longer history of industrialization, and that the idea of using AI tools to make things more efficient is actually just reorganizing of labor. Could you elaborate on this argument for those who could not attend the conversation?
If AI systems can be seen as a kind of optic revealing the enduring character of colonial relations and the reproduction of existing inequalities, it is because these systems actively participate in that reproduction—at industrial scale and speed. This is one sense in which AI is an industrial technology. Another way in which it is an industrial technology is in the material infrastructure that it requires. We may use AI on small computers in the comforts of our homes, or on our phones wherever we find ourselves, but each touch of the button or swipe of the screen on those devices is supported by a vast network of machinery, energy consumption, human labor, and the extraction of planetary resources.
The work we do here, in front of the screen, is supported by what Neferti X. M. Tadiar characterizes as “a whole array of nonsubjectified labor, or disposable life-times” elsewhere. It is this labor, she argues, that ultimately creates “the saved time for the consumption of the image”.[1] If one part of the solution to the excess of images that Farocki pointed to was the automation of seeing—the offloading of seeing onto machines—another part lies in the creation of “time for the consumption of the image” in the metropole through the appropriation of labor in the periphery, to use the old terms. The amount of labor that is required to secure the time we need to interact with AI systems—both the ones we choose to engage with and those we are unknowingly entangled in—is unfathomable. To describe this scenario as “postindustrial” does not seem particularly accurate. If anything, we live in a world that is more industrialized than ever. And this industrialization expands through the reproduction of colonial relations as described by Tadiar and many others. The time saved for the consumption of the image here comes at the expense of someone else’s time spent labeling data, moderating content, processing rare earth elements and struggling with the health problems that result from the pollution produced in this process… elsewhere.
- Do you think AI technology would look very different in a more socialist and less capitalistic society? What could it look like?
I think it’s important to ask what needs the array of AI technologies being used today are fulfilling. This would certainly be different in a noncapitalist society. In a socialist society, AI would not be first and foremost driven by the production of capitalist value, that is, profit based on the absorption of time and energy through these technologies. Instead, it would be developed along with other technologies, in order to meet concrete needs. Crucially, those needs wouldn’t be determined by what’s profitable in the market, but through democratic processes—collective deliberation, planning and decision-making. Today, the meeting of concrete needs is more like a necessary detour on the way to creating value for the shareholders of whatever company is behind the development. This profit-motive drives the production of more and more needs that can only be met through these systems. It makes the systems more and more necessary in everyday life. But necessary for what?
- You have worked with documentary filmmaking and you are also interested in how documentaries are made. The art critic John Berger introduced the concept of the gaze in his work Ways of Seeing from 1972, where he argued that much of Western art depicts women as passive objects, intended to be seen by men. Film theorist Laura Mulvey later built upon this argument and coined the term the male gaze in context of film. In some sense, all films have the gaze of the filmmaker, be it a male gaze, a white gaze, a Dutch gaze etc. Do you think that images generated by AI also have a gaze? And furthermore, do you think that a gaze such as the male gaze is something that can be learnt and generated in AI images?
Yes. Absolutely. If anything is being automated in the production of AI images, it is the reproduction of the forms of images that are already in existence. The models are trained on this vast trove of existing images—existing in the sense that they were online at the point when the internet was scraped in order to create the dataset that is used to train the model of a given AI. Of course datasets could be created which would actively work against images created according to the male gaze, white gaze, colonial gaze, etc. But the reality of the development of the most popular generative AI systems is that they use these larger datasets in which these gazes have a significant presence. And as the tools we use to make contemporary images are increasingly including elements of AI systems—think of the computational nature of smartphone camera images—the gaze for which these images were made is woven into the very fabric of our new images. Think of the biases of film chemistry in representing skin tones of people of color in the last century (back in the 1970s, Godard famously denounced Kodak film as being racist), the gestures and composition of images from the history of film and the selection of models/actors, advertisement and mass media more broadly, to mention just a few aspects here. Statistically, these images express a racist, sexist, colonial, heteronormative past, and AI serves to reproduce and elaborate this past in the present. Ultimately it is a capitalist gaze that integrates and elaborates these gazes in today’s AI systems. Beverley Best describes this gaze as “a gravitational pull on capital’s bearers towards a certain way of seeing, categorizing, and understanding the world. Capital,” as she puts it, “generates a pantheon of self-representations that inform by both betraying and confirming (on different registers) our experiences in the world. At the same time, people are constantly challenging, analyzing and disassembling these surface stories. The capacity to do so is universal; the time and space to undertake the work is limited and rare.”[2] Artistic research is one of those places where this capacity can be exercised. Where we can work on disassembling the surface stories that capital’s bearers—AI and otherwise—are driven to produce and consume. Disassembling the patterns that have been learned through the training of these AI systems—the gazes that you mention—rather than integrating them into the development of capitalist value is one way of denaturalizing and working against the gravitational pull that Best describes.
[1] Neferti X. M. Tadiar, Remaindered Life (Duke University Press, 2022), 114.
[2] Beverley Best, The Automatic Fetish, (Verso, 2024), 10.