Three Rights of the Image

The habitual anxiety of the era of generative images is usually formulated as a technical task: how to distinguish a real photograph from a synthetic one? This formulation, however, is surprisingly shallow. It presupposes that the problem of the image is exhausted by its origin, whereas the genuine question is far deeper and essentially politico-philosophical in nature: on what grounds is an image granted the right to speak on behalf of reality at all? An image does not merely depict — it claims, testifies, persuades, is experienced. And each of these functions relies on its own, historically established mechanism of trust. Generative technologies are dangerous not because they have learned to fake pixels, but because they have learned to intercept these mechanisms of trust — and moreover, differently at each level.

It is convenient to break down this structure along three classical philosophical dimensions: the ontological (what exists), the epistemological (how we know that it exists), and the phenomenological (how the existent is experienced by the subject). At each level the image performs its own work — and at each level its own, specific substitution can be carried out.

The classical theory of photography — from Susan Sontag to Roland Barthes with his famous “that-has-been” from “Camera Lucida” — was built on indexicality: the photograph is a trace of reality, the physical result of the contact of light with an object. The synthetic image severs precisely this connection, but does so more insidiously than simple falsification. Consider a characteristic situation: a user asks a system what a specific existing building façade in Lausanne looks like, and receives a plausible image, generated without a single act of observing that building. A strange construction arises: the real object — the absent observation — the synthetic image — its presentation as the visual existence of the object.

Let us note a subtlety here: the image does not lie about reality in the ordinary sense. It occupies a place which, by the logic of the situation, ought to have been occupied by the trace of reality. This is not a false statement about the world, but a usurpation of an ontological position. As long as a synthetic explosion over Tehran is perceived as fantasy or artistic visualization, there is no ontological problem — just as there is none in historical painting, whose conventionality is stipulated by the genre. The problem arises at the moment the image is moved into the slot “this is what Tehran looks like right now.” The generic frame has been removed, and the question “what actually happened?” receives its answer from an entity that observed nothing.

Baudrillard described the simulacrum as a copy without an original; here we are dealing with a more bizarre case — a simulacrum of an object that does exist, but was never seen by the producer of the image. Reality remains, but its visual representation is appropriated by a procedure devoid of any connection to that reality.

The second level poses a different question: not “what exists,” but “why should I regard this image as testimony?” And here it turns out that the substitution does not require synthetics at all — it is possible with entirely authentic material as well.

A classic example from forensic practice: a surveillance camera did in fact capture a person running. The photograph is authentic, indexical, incontestable as a fact of exposure. But then follows a caption: “this is the criminal fleeing the scene of the crime” — and the authentic image begins to function as evidence for a claim that does not follow from it. All that follows from the frame is that some person was running at a particular place at a particular time. Identification of the person, intent, causal connection to the event — all of this is added by the interpreter, yet the epistemic credit of the photograph (“I see it — therefore it happened”) is unreflectively transferred to the entire construction as a whole.

Hence a formula which, in my view, carries independent theoretical significance: a real image can be an instrument of false knowledge. The authenticity of the material and the truth of the conclusion are orthogonal quantities. The history of forensics and propaganda furnishes abundant illustrations of this: authentic footage of unrest, attributed to the wrong country and the wrong year, has repeatedly been used in information conflicts — from the Syrian war to the Ukrainian one, where old videos are regularly returned to circulation with new captions. What is falsified is not the photograph but its epistemic frame. In doing so the photograph performs work that no text is capable of: it borrows trust from the very act of seeing.

The third level is the most uncomfortable for rationalist critique. The subject may be entirely indifferent as to whether the image constitutes proof: he looks — and receives the experience of “I see this.” The effect of presence, described already in the theory of virtual reality, does not require belief in authenticity; it operates at the level of perceptual evidentness, which outpaces reflection.

This is precisely why synthetic clips of missiles and burning cities act even upon a viewer who intellectually recognizes them as “AI slop.” He knows it is fake — and nonetheless receives the phenomenological effect: the missiles fly, the explosions occur, the event is present before his eyes. And if the content coincides with what the viewer wants to see, an additional mechanism switches on: “let this be a fake video — but this is exactly how everything must be.” The image ceases to be evidence of reality and becomes a machine for experiencing the desired reality. Here the substitution is subtlest of all, because it disputes neither ontology nor epistemology — it serves affect, and affect is not refuted by fact-checking. Psychology long ago described this phenomenon: repeated exposure to visual material increases the subjective plausibility of a statement (the illusory truth effect), and visual images possess a particular mnemonic power — the brain stores “what was seen” differently from “what was read.”

The critical situation arises when the three levels are superimposed upon one another. Then a full-fledged chain is assembled: a generative system creates something that does not exist (ontological substitution) → the result is presented as observed, and the observed as proven (epistemic substitution) → the proven is experienced as self-evident, emotionally present (phenomenological substitution). Each link reinforces the preceding one, and it becomes difficult to work out precisely where the failure occurred — because the failure is distributed across the entire chain, rather than localized in a single pixel or a single caption.

It is precisely this that renders futile any hope placed in purely technical detectors of synthetics: a detector operates only at the first link, whereas the second and third substitutions are available to entirely authentic material as well.

Up to this point the discussion has concerned the properties of an individual image. But there is also a second, perhaps more decisive factor — a quantitative one. The transition of quantity into quality, so dear to the dialectical tradition, acquires here a quite concrete mechanism.

Consider a thought experiment that has entirely real prototypes. Suppose that an array of authentic compromising documents surfaces in open access. Classical censorship would respond with prohibition: “you must not see this document.” But there exists a more sophisticated strategy — the strategy of flooding: authentic documents are submerged in a gigantic mass of plausible garbage, in which real intelligence data is intermixed with invented bribes, golden toilets, fabricated “leaks,” stylistically indistinguishable from the genuine ones. The key principle is formulated sharply: in order to discredit reliable information, it is not necessary to prove its falsity — it suffices to make it indistinguishable from an enormous mass of false information and sharply raise the cost of verification. What is attacked is not the content of the document but the informational environment in which the document exists.

RAND researchers as early as 2016 described a proposed model of universal propaganda as the “firehose of falsehood”: high intensity, multichannel delivery, rejection of consistency, rapid switching of versions. Generative technologies have industrialized this model: what previously required state apparatuses and budgets is now produced almost free of charge and in unlimited volume.

The mechanism of the transition of quantity into quality can be described through the economics of verification. While there are ten images, each one can be checked. There are ten thousand of them — sorting and prioritization appear. There are ten million of them — a fundamentally different question arises: why should we be paying attention to this particular image at all? Scale begins to change the very epistemic status of information. Particularly destructive is not random noise but noise structurally similar to signal: genuine and forged documents, real and generated photographs, actual and invented episodes of corruption, mixed together within a single environment.

At the same time, the cost of establishing the truth grows far faster than the volume of information. In a normal environment the chain is short: document → verification → conclusion. After the mass production of synthetic material it expands: verification of the source, of provenance, comparison of versions, analysis of metadata, checking of related materials, photographs, context, cross-verification of every claim. Hence a formula which I consider central to understanding contemporary disinformation: falsehood becomes effective not when there comes to be more of it than truth, but when the cost of separating truth from falsehood exceeds the resources available for verification.

A fundamental asymmetry arises: producing a hundred thousand plausible units of informational garbage is cheap, verifying them is expensive. This is an asymmetry of the same type as in cybersecurity or in the terrorist threat: the defense must win every time, it is enough for the attack to win once. The practical consequences are already visible: journalistic investigations that have encountered massive leaks require months and years of verification (the experience of consortiums such as ICIJ working with the “Panama” and “Pandora” papers shows that verifying even authentic arrays is years of labor by hundreds of specialists); an environment in which the authentic and the synthetic are mixed multiplies these timeframes toward infinity. The expected outcome is not belief in the lie, but fatigue with the very category of truth — something that can be called epistemic exhaustion: the subject renounces not a specific claim, but the procedure of verification as such.

Reducing the problem of generative images to the question “real or not” belongs to yesterday. The precise formulation sounds different: at what level is the image granted the right to speak on behalf of reality — at the level of being, of knowledge, or of experience? Substitution is possible at each of the three, and in its mature forms — at all of them at once, forming a chain from the generated non-existent to the experienced-as-self-evident. And at the level of the environment a second, quantitative mechanism operates: a critical mass of plausible material raises the cost of truth above any available resources for establishing it.

From this follows also the direction of possible responses — not technocratic, but institutional. If trust can no longer rest upon the indexicality of an individual image, it is compelled to migrate toward the reputation of the chains of production and transmission of information: toward provenance, signed sources, institutions bearing responsibility for their publ