Artificial Intelligence
Machine Images. Photography and Artificial Intelligence
Andreas Müller-Pohle
[German]
Artificial intelligence is a societal experiment in real time and arguably the most radical challenge to photography since its invention. It expands the medium’s discursive space into a sweeping debate about the image and visual culture, one that demands new concepts and paradigms. A manifesto in eleven theses.
1. Photography stands at the center of visual artificial intelligence
Artificial intelligence is changing our image of the world – and it begins with the image itself. At the center of this upheaval stands photography.
Technologically, the development of visual AI systems is unthinkable without the billions of existing photographs. They are the essential historical raw material from which learning systems draw – algorithms of image generation as well as of image analysis, from facial recognition to medical diagnostics. For generation, however, photography is more than raw material: it is the standard against which these systems prove themselves. Photographic semblance is their highest goal – the data image mistaken for a light image.
Socially, photography as the leading visual medium is being challenged by the new pictorial culture of machine images. Yet it remains the binding iconological frame of reference. The further the new images move away from the physical world, the more important photography becomes as an indexical reference – as the image still born of light and contact with the world.
2. Artificial intelligence produces datagraphs, not photographs
Light images from the camera and data images from the computer: here two types of image confront each other, capable of looking deceptively alike. And yet, ontologically, they are fundamentally different. Photography, whether analog or digital, is bound to the recording of a light event, to its indexicality. The artificially intelligent image, by contrast, is pure statistical synthesis based on a probability model. To set it apart from photography, I call it datagraphy.
It is tempting to separate the two types of image strictly from one another, yet such a binary approach is obsolete in the face of countless hybrid intermediate forms. The light image was never pure and innocent. In the mode of digitality, however, it is profoundly reshaped by algorithms – already within the camera, light is processed into computed data. This forces us to think in degrees rather than in absolutes.
This thinking in degrees operates on two complementary levels: the pragmatic axis of indexicality, that continuum on which image analysis and forensics take place. And the conceptual space of possibilities, the latent space, which encompasses the totality of all images a given AI model can generate. Image criticism in the age of AI is then no longer the demarcation of one against the other, but the determination of an image’s position on this axis and its interpretation within the context of the latent space.
3. Artificial intelligence is a creative accomplice
With the transition from light image to data image, the figure of the author as sole creator disintegrates. When every new image arises as a synthesis of already existing works, every earlier author becomes a co-author – if only infinitesimally, in minuscule shares, like a drop of blood in the ocean.
At the same time, a new relationship between human and machine emerges. The former hierarchical order between creator and tool gives way to a collaborative dialogue. New hybrid forms of creativity arise, in which the boundaries between human intuition and algorithmic synthesis blur. Just as photography made the invisible visible, artificial intelligence opens hidden worlds to us, outer and inner alike – without light, with data alone.
This shift in the creative process shatters our ethical and legal foundations: who bears responsibility when no one is the sole author anymore – the developers of the systems, the users, or the providers of the training data? And what remains of copyright, which presupposes the human as creator, when generated images no longer know any such creator? The concept of sole authorship – it is dead.
4. Artificial intelligence is a Super Black Box
Camera and computer appear to their users as black boxes – their complex inner workings remain opaque. Only their developers can fully penetrate them.
Artificial intelligence is different, embodying a black box of a new order – a Super Black Box. It is opaque not only to laypeople; even its developers cannot fully explain the multilayered computational pathways of large generative models. We see input and output, but the how and the why remain in the dark. More than that: in the thicket of billions of parameters, the machine develops dynamics of its own and follows hidden intentions. It computes, and yet it is incalculable.
A tremendous leap: where the classic black box was opaque but linear and traceable in its logic, the Super Black Box is opaque, unpredictable, and irreversible. Its boundless neural interplay eludes our understanding. The same prompt produces, edge cases aside, a different image every time – an unfathomable labyrinth from which there is no way back.
The Super Black Box confronts us with as yet unsolved ethical challenges, whether in diagnostics, in administrative processes, or in autonomous driving. A system that decides in the dark cannot be the final authority. It can neither assume human responsibility nor replace it.
5. Probability is the new paradigm of truth
A statement is considered “true” when it corresponds to reality. The optical analogy between the seen image and the photographed image accordingly led to the assumption that the camera offered a truth machine, one that, by virtue of technology, delivered objective proof of the visible.
By the dawn of digitalization, if not earlier, this naive faith in truth became obsolete. The analog threads that once bound photography to the external world were broken down into bits, ready to be reassembled at will through computation. No longer decided automatically, the question of truth shifted onto authorities of integrity – media, agencies, or expert assessors.
The data images of artificial intelligence no longer have any direct reference to reality. Calculated from statistical probabilities, they are to be read as probabilities as well: they are neither true nor false, but only more or less probable.
A paradigm shift that not only pervades the world of images but also calls into question our social cohesion. For when all certainties fade and fictitious constructs govern our view of the world, the very foundation of our free civilization is imperiled. The erosion of truth is more than a media phenomenon; it is a political factor.
6. Artificial intelligence is a kitsch machine
The kitschification of the image sphere is not a new phenomenon but has been bound up with the technologization of image production from the very beginning. Kitsch marks the final stage in the life cycle of images: syntactically redundant and worn out, semantically banal and empty. It is a gauge of an era’s visual pollution.
With the rise of digitalization, and especially since the advent of machine-generated imagery, this problem has intensified exponentially. Kitsch is an almost inevitable consequence of the amalgamic nature of artificial intelligence, of its synthesizing mode of operation, of its inner logic. This is compounded by the sheer quantity of its use. Kitsch floods social media, popular culture, and advertising, creeping from there into our minds: aesthetic atrophy.
Yet kitsch is not only an aesthetic phenomenon. It is equally prevalent in the political sphere, in populism of every stripe, in the bubbles and echo chambers of alternative facts and constructed realities: kitsch is Janus-faced.
7. The tectonics of photography are shifting
The entry of artificial intelligence into photography is fundamentally altering the spectrum of the medium. The social and professional fault lines are already unmistakable.
Hardest hit are product, advertising, and fashion photography, which are either rendered entirely superfluous or replaced by hybrid productions in which classical photographic techniques merge with AI-based processes.
The entire profession is under pressure, and on two fronts: technologically, through AI systems that can produce images faster, cheaper, and at higher quality, and legally, through the unauthorized use of protected image material for AI training.
The situation is different in visual journalism: its significance as a source of verifiable images will grow. Yet it too faces a twofold threat: the growing competition from social media images, which are faster and cheaper to obtain, and, bound up with this, a mounting economic precarity.
Documentary photography and journalism respond with complex narrative strategies, with context, transparency, and the traceability of research, supported by technical approaches such as the Content Authenticity Initiative, which relies on tamper-proof metadata.
8. Artificially intelligent art needs critical strategies
Art made with artificial intelligence: a field only just emerging and yet already astonishingly differentiated. Two opposing positions stand face to face.
On the one hand there are the so-called AI artists, who use the new technology as a toy. They revel in the oddities the machine brings forth, often losing themselves in the endless convolutions of their tools, and celebrate their excesses, their “cursed” finds, as pinnacles of creativity. Their stance is affirmative, their interest primarily aesthetic.
On the other side are the conceptual artists, who play not with but against AI. They too search for unusual pictorial forms, yet they use AI above all as a discursive tool: politically, to expose ideologies and power structures; scientifically, in investigating the laws of nature and visualizing large-scale data structures; and media-reflexively, by taking into account the conditions of their own genesis, creating transparency about their processes, or interrogating the photographic canon itself anew. Their shared stance is critical, their interest primarily thematic.
It is a task of art as urgent as it is noble: to explore the opportunities and risks of artificial intelligence – and to develop it further as dimensions of our human intelligence.
9. Artificial intelligence is power
Artificial intelligence is not a neutral technology but an instrument of societal power. The direction of this power points strictly from the top down – in the AI models themselves, whose development lies in the hands of a few tech corporations and states, as well as in the techniques of surveillance and control derived from them.
After the appropriation of private and social space by media platforms and social networks, public space too is now being dissolved – a space that once afforded anonymity to the individual. Facial recognition, biometric capture, and algorithmic prediction of behavior are becoming everyday routines. Any opposition can be ground down; no one can feel safe any longer, no matter when or where. The path to technodictatorship is paved.
The internal concentration of power finds its counterpart in the geopolitical struggle for digital world domination – first and foremost over superintelligence, which surpasses us humans, over raw materials and hardware, computational capacity, training data, and access to the next frontier models. Artificial intelligence is the all-dominating power project of the future. It splits the world into a few centers and many peripheries.
10. Artificial intelligence devours itself
Intelligent systems must learn incessantly in order to evolve. Their hunger for images and texts, for experience of the world, is boundless. But the material is finite – the archives and libraries are being plundered before our eyes, while the net fills with synthetic products. The machine begins to feed on itself: data cannibalism.
This is the Habsburg effect, the incestuous impoverishment of the gene pool with its well-known consequences: First the margins wither away, the deviant and the idiosyncratic, diversity itself, until even the center degenerates into a bland monoculture. Ultimately, the machine becomes ensnared in a feedback loop where probabilities are merely calculated from other probabilities, until the system dissolves into formless noise: model collapse.
Datagraphy is structurally backward-looking; it merely recombines what already exists. It gains fresh input solely from the indexical images of the here and now, images born of light and reference to the world, even if these have long since become algorithmically permeated themselves. Therein lies the irony: datagraphy displaces photography – and yet starves without it.
11. Artificial intelligence needs the human being
Humans seek the new – and fear it at the same time: an ambivalence that compels us to weigh opportunities against risks.
The utopia of artificial intelligence beckons with virtually boundless promises: the unleashing of perception, the universal expansion of our knowledge, the liberation from human labor, the healing of all diseases, and the decentralization of ossified power structures.
The dystopia looms with nothing less: the infiltration of perception, the decomposition of knowledge, the dispensability of humanity, the breeding of new diseases, and the concentration of power in a few hands.
Which path the future takes will be decided by the question of control. Will we succeed in penetrating the Super Black Box, in deciphering its codes, and in aligning the inevitably approaching superintelligence with human values – or will it break out of its confines and pursue its own goals? This is not merely a technological question but an eminently political one.
It is a question negotiated in the tech corporations, governments, and parliaments, far from us, and yet it has one place where it reaches us daily and directly: the image. The image is a reliable indicator of how we deal with machine technology at large. For photography as we know it is on the defensive – a fragile anchor of indexicality that we must defend if we wish to preserve our bond with light and world, or else surrender to the voracious machines and their synthetic logic.
Photography and datagraphy form the inseparable image poles of the present. It is up to us, as political actors, to decide how much formative control we exercise over both. Let us meet the machine not with blind reverence, but with a challenge: with critical enthusiasm and playful seriousness.
© 2026 Andreas Müller-Pohle
First published in European Photography, Berlin, no. 120, summer 2026