Reality is not a no-brainer / A New Conquest of Reality. An interview with Jan Distelmeyer

Donald Trump with a dinosaur lizard in his arms, huge quantities of food being shoved into absurdly large, distorted mouths, Leonardo DiCaprio talking about climate change with the voices of Robert Downey Jr., Bill Gates and Kim Kardashian. Images in which bodies appear grotesquely distorted, text-to-video trash, but also works by conceptual artists who use AI software. In the sea of AI images currently flooding social media, one finds both rubbish and aesthetically successful works. Over this jumble, a voice – AI-generated, of course – explains what’s going on: “We live in a strange time.”

Silvia Dal Dosso’s highly amusing short film essay The Future Is Now Weird AF (Part 1) (1 May, 11 am Tickets), screened in the programme “Training the Image – How Images Learn”, takes stock of the situation using a striking montage of AI images that have been hyped in recent years. And it poses the questions that are central to the festival’s theme, “based on true events?”. Questions that take the AI complex as a starting point to examine how filmmakers have always distorted, stretched, shaped and deconstructed reality.

One starting point is the emerging breakdown of the relationship between image and reality in an interview (see below), media scholar Jan Distelmeyer no longer speaks of images, but of the visualisation of data sets. However, there is little that is artificial here, in the sense of ‘not made by humans’. Algorithms and probability theory, certainly. But for DALL·E and ChatGPT to be able to calculate the image the user is likely to want, vast amounts of human labour must be fed into the system. Several films in “Training the Image” vividly illustrate the production basis of AI and focus on the precarious workers, mostly living in the Global South, who feed the machines with data.

Their Eyes by Nicolas Gourault (29 April, 8.15 pm Tickets) documents how images are labelled. We see the clicking operations on screen, accompanied by the voices of workers who are well aware of the miserable conditions and the imbalance between labour and profit. This is enlightening in the classical sense: After all, the perception persists to this day that AI is primarily based on algorithms and is, so to speak, a self-sustaining system. Furthermore, Their Eyes is a tribute in solidarity with the people who make up the precarious global proletariat: every one of the exploited workers recorded by Gourault possesses more dignity in their little finger than Elon Musk has in his entire body, to put it with a touch of the appropriate pathos.

All That is Solid by Louis Henderson (30 April, 5 pm Tickets) combines digitally constructed images with documentary footage of illegal gold mines and mountains of electronic waste. Even in the latest technological revolutions, nothing is immaterial; every fleeting image made up of pixels is the result of human labour and material processing. The ‘cloud’ depicted in All That is Solid is, of course, a vast server farm. It is always about hardware and its production, even if key terms suggest cloud-like qualities and virtuality.

The majority of the films in “Training the Image”, however, take the AI aesthetic and spin it on its own axis before feeding it vigorously into a blender. We see isolated machines wandering through a monumental apocalyptic landscape and falling into philosophical musings (Third Impact, directed by S()fia Braga, 29 April, 8.15 pm Tickets), a cinematic meditation on the grotesque, viral image of the Pope in a bulky hip-hop puffer jacket, and a fundamental training session on the problem of knowledge “What is a chair?” (One & Infinite Chairs, directed by Egor Kraft, 1 May, 11.00 am Tickets).

Another part of “based on true events?” takes us back to the very beginnings of film history. In the mid-1990s, around 900 film reels produced by Mitchell and Kenyon were rediscovered – in three metal milk churns in the cellar of a shop in the northern English industrial town of Blackburn. The films had been commissioned around 1900 by travelling exhibitors. Billed as local films for local people, audiences paid to see their neighbours, children, families and themselves in these collective portraits on the screen. The titles of the films, most of which are between two and six minutes long, serve as their synopses: Workers Leaving a Factory in Droylsden, Cunard Mail Steamer Lucania Leaving for America or Whitehaven Street Scenes (2 May, 8.15 pm Tickets).

In essence, then, the promise of a direct, simple reflection of reality: person/object + camera + film material = image. However: the positioning of the camera, but also of the people, the workers that are filmed – when viewed as a whole, the films reveal the considerable construction work that goes into even the simplest images.

A second dive into film history reveals a highly idiosyncratic approach to capturing reality on camera. Werner Herzog, who has been part of the history of the Oberhausen Festival since the 1960s, refused to accept the distinction between fiction and documentary film from the start. All in the service of a cinematic quest for what Herzog calls ‘ecstatic truth’. The programme features several of his early short films. Some of them, Letzte Worte (Last Words, 1968), Maßnahmen gegen Fanatiker (Precautions Against Fanatics, 1969, both 30 April, 10 pm Tickets) or How Much Wood Would a Woodchuck Chuck… (1976, 2 May, 3 pm Tickets), present people whose language is no longer comprehensible. Either because their dialect is incomprehensible to outsiders or because they simply remain silent. Linguistic expression as a vehicle of meaning disappears, and another reality comes into focus: the reality of bodies, their gestures, their unintentional comedy.

Die große Ekstase des Bildschnitzers Steiner (The Great Ecstasy of Woodcarver Steiner, 30 April, 10 pm Tickets) spends over three quarters of an hour repeatedly showing ski jumper Walter Steiner in action. Risking his life, as Herzog – who plays a sort of sports reporter in this film – never tires of emphasising. And in Werner Herzog Eats His Shoe (dir. Les Blank, 1980, 2 May, 3 pm Tickets), exactly what the title promises happens, among other things (not without quoting Charlie Chaplin’s The Gold Rush). Here, too, reality is meant to emerge within the immaterial reality of the film (“it’s only a projection of light”, Herzog laments in his charming Bavarian English), which is intended to become as material as possible here: first the shoe is there, then it’s gone. An immediately physical process.

The programmes “Hide and Seek” and “Truth or Dare” in turn focus on recent, surprising cinematic interventions that challenge traditional ways of reconciling film and our view of the world. The works in “Truth or Dare” (1 May, 8.15 pm Tickets) reshape heteronormative reality through the lens of queer desire. These include the sailor fantasies in Kenneth Anger’s black-and-white film Fireworks, as well as the utopian cosmology of STARS (directed by Mojisola Adebayo, Candice Purwin, Debo Adebayo, Tyler Friedman) and the unsettling montage The Beginning of Identification, and its End (dir. Philipp Gufler), in which speeches by far-right gay politicians such as Alice Weidel and Pim Fortuyn are edited together with the beautiful image of a naked man being permanently hosed down with a jet of water.

“Hide and Seek” (3 May, 3.15 pm Tickets) is harder to pin down. Films such as Le prime volte (The First Times, directed by Giulia Cosentino and Perla Sardella), Para Carlos (For Carlos, directed by Carlos Cipriano) or Ese lugar partido (That Split-Up Place, directed by Inés Pintor Sierra and Pablo Santidrián) revolve around the shared pasts of people, and thus also around the cinematic reconstruction of the realities of relationships (and the unreliability of these reconstructions). In the deeply moving Le prime volte, a lifelong friendship is captured in photographs. In Para Carlos, a shared past is addressed in a letter. And Ese lugar partido tells the story of the separation of two childhood friends, whose shared reality is separated by a motorway.

JuJu vs The Possibilities of Love, Life & Death (directed by Htet Aung Lywn) and Aliens in Beirut (directed by Raghed Charabaty), meanwhile, also present decidedly queer world-building. Charabaty tells the story of a budding summer romance, set against an intensely colour-graded backdrop. It ends with the devastating explosion in the port of Beirut in August 2020. Htet Aung Lywn, for his part, deconstructs, in a sense, a tradition within queer cinema that tends to portray trans people primarily as victims. He does so by dreaming up, in collaboration with his protagonist, the artist Juju, alternative queer images and cinematic fantasies, so to speak.

Werner Herzog, queer realities, materialist critique of the production of so-called artificial intelligence: What unites all these films is something that, for the time being, still runs counter to the universe of AI imagery or visualisation: a radical and indomitable obstinacy. Successful and often promising attempts to break free from schematic visual worlds and the realm of the probable.

Benjamin Moldenhauer

>>> back to the top

It used to be relatively simple: a photograph depicted something of the world. But did or does such a representation in this sense even exist, or is that a myth?

Yes, it is; an image is never simply a representation of the world, but always a construction of reality. And this construction is subject to certain conditions. The conditions of a film camera and a microphone are different from those of software with which I generate images and sounds. Film has very specific conditions of construction: for instance, the equipment, traditions of filmmaking, certain aesthetic choices, but also worldviews and ideologies. And there are strategies for not simply accepting this construction of images and the world, but for viewing it critically. bell hooks coined the lovely term ‘oppositional gaze’ for this. It starts from the realisation that what is presented as a document of reality is always already a construction that can also be read against the grain.

With a photographic image, at least, one still has the impression that reality is being imprinted onto a material. Then came the digital image, and now AI-generated images. Does a statement like ‘AI-generated images have less to do with reality than celluloid’ make sense?

I find it problematic to rashly declare AI a problem, only to then romanticise film as a noble machine of reality. Film cameras and microphones operate with the promise of a physical-causal relationship between the world and what can be captured on film and other recording media. The film theorist André Bazin once called this the ‘art of lying’. Film has something of the gesture of ‘that’s how it was’, and therein lies the very power of its lie. The cinematic oscillates between this promise and the selection, editing and montage of images and sounds. The conditions under which AI-generated images and sounds are constructed work in a fundamentally different way.

What, then, is unique about AI images – that is, images that refer solely to other images? Are we dealing with a circular visual world?

AI images are, first and foremost, data visualisations. They rely primarily on the generation of average values from vast data sets, from which images are then calculated that are likely to match what might be desired in a prompt. For text-generated images, hundreds of millions to billions of image-text pairs are required, from which the probability of the relationship between a word fragment and an image element is determined. We are dealing with a mixture of averages and probability in the relationship between image and text. Hito Steyerl has therefore called these images ‘mean images.’

What connection does this visuality still have to what we call reality?

If I may quote Bazin once more: film was about the ‘conquest of reality’, not merely its reproduction. This conquest can take many different forms, including, of course, non-realistic ones. In any case, it is based on photographic technology and on how people use and develop it. That is why it is very important to recognise that the connection to the world, to truth or to meaning remains the task of human agency, including with AI. Images must be labelled, terms sorted, meanings marked. For this stochastic image production is not, in fact, about meaning. That is why the connection between image and world in AI always depends on human validation. This fundamentally distinguishes this form of imagery from the photographic.

When you work with AI systems, you only see the surface, the interface. But behind it lie technical and economic processes, as well as work processes. What hidden forces are we dealing with here?

Let’s perhaps take a slight detour first. What slightly annoys me about the public discourse on AI is that people constantly speak in the future perfect tense, about what AI will have been. We’re constantly expected to keep our minds in the future: how AI will change the world, whether it will destroy humanity or take us to new heights. This makes it increasingly difficult to take a close look at the here and now of this technology. That’s not easy anyway, because, as you say, much of it remains hidden. What bothered me about the early debates on AI images, for example, was that the discussion almost always focused solely on the results – the Pope in a down jacket, Trump in a police headlock – but never on the interface configurations in which we encounter such images when we generate them. Yet that is precisely where there is much to learn. For example, about the probability inherent in these images. When I entered a prompt into DALL-E in the first few months after its launch, I received four images in response by default. The system does not present me with ‘the’ image, but with several possibilities. It performs its statistical uncertainty as well. It does not say: ‘This is the true relationship between text and image.’ Rather, it says: ‘These are probable relationships.’ If we take a closer look at these interfaces, we are led to ask questions other than simply: ‘Is the image genuine?’ We can then see that what we are being presented with here is a range of probabilities.

My impression is that, in this context; too, the distinction between truth and fake loses its relevance. At some point, it no longer matters whether it’s true or not.

I see your point. And I believe one of the major challenges in the near future will be to cultivate a new critical perspective, specifically in relation to AI-generated images. To do this, we need to engage with the technology and its various manifestations. After receiving responses from a system like ChatGPT, I am always asked to evaluate the answer. I am expected to provide constant feedback, evaluate, assess and remain engaged in dialogue. These systems are extremely reliant on this. In film, the connection to the world attempts to manifest itself within the image. In AI interfaces, by contrast, the connection to the world is queried.

So anyone who uses these tools inevitably ends up actively contributing to the way they construct the world or images?

Exactly. We’ve always been co-workers in this AI industry ‘mine’, without being paid for it. An incredible amount of human labour goes into these systems, particularly from precariously employed data workers. Estimates suggest there are between 150 and 430 million data workers worldwide. It is difficult to say exactly how many of them work specifically on AI, because this work is so often outsourced and organised in precarious conditions. Making this work visible is a central concern for many researchers and initiatives. But this is precisely what does not shape the public image of AI, because artificial intelligence is supposed to appear artificial.

Part of this oppositional perspective would therefore be, in a classic materialist sense, to reveal the conditions of production.

Yes – and the conditions under which I myself interact with AI. Because by engaging directly with the interfaces of, for example, ChatGPT, Gemini and so on, I can learn a great deal about what these systems actually want from me. They want input, they want evaluation, they want and give validation. Looking at the systems themselves, rather than just staring at their results, helps to demystify AI.

How can we engage with the results – the images – in a critical and informed way?

That is an incredibly difficult question, and I believe it is one of the central questions facing visual culture at the moment. I don’t have a simple answer to it. But we certainly need a new focus on the relationship between reality and the image. Scepticism cannot simply consist of saying: ‘None of this is true.’ As if, conversely, other images were simply true. A more interesting question is: What world view, what argument, what point is actually being put forward here? How does this relate to the conditions of production?

Perhaps the problem also lies in the fact that no distinct genre has yet emerged for AI-generated images. With action films, there are rules to guide you. With AI-generated imagery, everything still feels vague and tentative.

Yes, I would see it that way too. And perhaps the upheaval that AI-generated images caused in our relationship with imagery actually has an advantage: that we ask more strongly about the practical value of these images. In other words: What do I need this for? What do others need this for? If film is a conquest of reality, one might ask whether and how such a conquest still takes place in these images at all – or whether it needs to be recreated by us. Against this backdrop, AI images raise an interesting question: what is this image actually supposed to convince me of? That, too, is a reference to reality – but not one that relies on mere representation, rather one that seeks to make an argument.

An appropriation of AI?

Exactly. It’s a twofold appropriation: on the one hand, by those who wish to make a case using images, sound and moving pictures; on the other, by us, who must ask: What can we use this for? Do we buy into it? Do we read between the lines? Do we turn it into something else? With the small caveat that this alarmist ‘we must’ is always a bit tricky. AI-generated images are not a complete paradigm shift, but rather represent the culmination of a discussion that was long overdue anyway. What is new, however, is that anything that is not already exemplary in aesthetic or cultural terms is much harder to produce using what we call AI.

The interview was conducted by Benjamin Moldenhauer

>>> back to the top

Jan Distelmeyer is a professor of media history and media theory in the Department of Media Studies (EMW) at the Potsdam University of Applied Sciences and the University of Potsdam; his current research focuses on AI, interfaces, and data work, among other topics.

30 April, 10 am: Apocalypse AI? Based on true events.

What tools can we find in the history of film and art when it comes to distinguishing reality from illusion? Might these tools help us to move closer to the concept of a self-critical AI aesthetic? And in doing so, take into account the effects of the AI industry on society and the environment?

With Prof. Clemens von Wedemeyer, Professor for Expanded Cinema, Academy of Fine Arts Leipzig; Ariana Dongus, critical media scholar, Berlin; Prof. Dr. Jan Distelmeyer, Professor for Media History and Media Theory, Potsdam University of Applied Sciences and University of Potsdam; Dr. Inke Arns, Director of HMKV Hartware MedienKunstVerein, Dortmund.

Free entry

>>> back to the title page