A Vital Critique of Fontcuberta’s Algorithmic Photography

This March, Spanish conceptual artist and photographer Joan Fontcuberta printed a brand new guide in Italy. Immagini Latenti concludes with a chapter on AI and pictures, referencing the debates surrounding Boris Eldagsen’s submission of an AI-generated picture to the Sony World Photography Awards in 2023 and Miles Astray’s submission of {a photograph} within the AI class of the 1839 Award in 2024.

When we obtained our copies, we had been struck by the context wherein our photographs appeared. Fontcuberta frames AI-generated photographs as “Second-Generation Photography” and proposes the time period “Algorithmic Photography”.

Over the previous 2.5 years, now we have repeatedly encountered comparable strains of reasoning. They usually are not solely logically inconsistent, however they’re additionally unhelpful, each for pictures and for democratic societies. For that cause, we discover it vital to reply collectively to what we contemplate a rudimentary concept. Below, we distinction excerpts from the guide (used with permission) with our personal perspective. The unique Italian textual content has been translated into English utilizing a number of AI instruments (ChatGPT, Gemini, DeepL).

I. The Naming Problem

Joan Fontcuberta:

When I acquired married, some pals gave me a lemon tree […] We planted it and it grew fortunately. […] after twenty-five years […] the lemon tree started to supply oranges. […] A buddy who’s an skilled in citrus fruits […] gave me a believable rationalization, […] our lemon tree had virtually actually been grafted onto a department of an orange tree, and over time it started to disclose its true hybrid nature—non-binary and ambivalent.

Personally, I most popular to maintain considering that the tree had discovered the braveness to return out of the closet. All the extra so as a result of it appeared to me a powerful metaphor for what is occurring to pictures immediately, which can be going by means of a part wherein it’s about to return out.

Let me clarify. For two centuries, now we have attributed to pictures a descriptive accuracy of actuality that assured absolute documentary constancy. Now, nevertheless, algorithmic pictures is mixing with optical pictures, and we now not know which option to flip.

Immediately, we encounter a semantic and terminological downside. There are photographic photographs produced by cameras and photo-optical recording programs. And there are others—apparently photographic—produced by means of generative AI visualization programs.

The former are youngsters of chemistry and lightweight; the latter of computing and darkness. We should subsequently start to determine whether or not each forms of picture must be thought of photographic.

If we give attention to the processes concerned, it’s apparent that they’re totally different sorts of photographs. Yet the issue of discovering a phrase able to classifying photorealistic representations of algorithmic origin weakens the decisiveness of that reply.

These are photographs and not using a actual referent—what we’d name nemotypes.

Some have proposed the time period promptography, as a result of such photographs originate from a immediate—that’s, natural-language directions given to a system with a purpose to receive the specified photographic outcome.

There have been different makes an attempt, similar to syntography, however none have prevailed.

When pictures was shaken by the arrival of digital expertise, it turned essential to specify that there had been a earlier kind to which a distinguishing adjective was now added: we had analog pictures—or photochemical pictures—versus digital pictures. At that point, there was no must invent or assign a particular new title, and nothing disastrous occurred. Therefore, we might most likely proceed in the identical manner now and nonetheless perceive each other completely.

Boris Eldagsen:

Fontcuberta clearly acknowledges the excellence between camera-made and AI-generated photographs on the degree of course of – however then argues that this distinction finally doesn’t matter.

The downside is: it issues. Considerably.

{A photograph} is made by gentle bouncing off an actual factor and hitting a sensor. An AI picture is made by a pc calculating what a believable picture would appear like, primarily based on patterns discovered from tens of millions of prior examples. The outputs could seem equivalent on display, however they emerge from essentially totally different processes. And it’s exactly this course of that grants pictures its authority as proof.

Calling AI photographs “Algorithmic Photography” treats this as a minor improve: a lemon tree merely producing oranges. But even in Fontcuberta’s personal metaphor, a lemon remains to be a lemon and an orange remains to be an orange. Grafting doesn’t change what a fruit is. Two fully totally different sorts of picture are being given the identical title, and that confusion has actual penalties.

By this logic, a photorealistic portray would grow to be “Acrylic Photography”. But we nonetheless name it portray, as a result of the method issues, and it has been created with canvas, brushes, and paint.

Arguing that the dearth of an satisfactory time period for “photorealistic representations of algorithmic origin” justifies subsuming AI photographs below pictures is weak. On the one hand, naming a brand new medium takes time. On the opposite hand, Fontcuberta stays confined inside photographic considering and fails to acknowledge what this new medium really is: LATENT SPACE.

It consists of the coaching information of an AI mannequin wherein all media is encoded as vectors. In Latent Space, totally different artwork types are now not separate supplies. They grow to be totally different projections of the identical underlying construction. A melody can morph into a picture. A textual content description can generate a video. A sketch can grow to be a sculpture. Latent area is a meta-medium.

This is why prompts have grow to be multimodal. The immediate is a management interface to latent area, navigating likelihood.

And that’s exactly why I steered the time period “promptography”. It encompasses every thing produced with a immediate: textual content, sound, video – not simply photographs resembling pictures, but additionally these resembling drawing or portray.

Because Fontcuberta limits his evaluation to “photorealistic representations”, he reduces the dialogue to a slim subset of outputs—and consequently struggles with the arguments that observe.

Miles Astray:

The lemon is a tough fruit—linguistically talking—past Fontcuberta’s allegory. In his residence nation, Spain, a lemon is named “limón”, whereas throughout Latin America, “limón” means lime. Lemons, limes, oranges – they’re all citrus fruits, however likening them will not be in contrast to evaluating apples and oranges.

Here’s the factor, plain and easy, all fruits apart: this isn’t about wishy-washy linguistic interpretations of images and artwork; that is about stable scientific reality. Photography is written with gentle; AI imagery is written with code. The former captures the true world, the latter conjures imaginary worlds.

A linguistic disagreement on terminology doesn’t translate right into a scientific dispute across the factual distinction between the processes concerned in creating photographs, from work to pictures to AI footage. There is a science to artwork, and it’s within the course of.

The distinction between analogue and digital pictures might simply be summed up by a prefix as a result of the underlying photographic course of (capturing gentle) hadn‘t modified, solely the technique of the way it was captured and saved (chemically vs. electronically, movie vs. sensor). However, to reach at an AI picture, it’s a must to take a totally totally different procedural route, which deserves a totally totally different title.

To disregard a large procedural distinction between two mediums in lieu of developing with one little phrase to explain the brand new property is disproportionate and misdirected. That can be like calling each fruit that got here after the banana—which got here lengthy earlier than oranges and lemons—additionally banana, and that might be bananas.

II. The DNA Problem

Joan Fontcuberta:

[…] But the talk goes deeper: are we coping with photographs belonging to totally different lessons, or just pictures of various rank?

[…] It is simple to think about that everybody dreamed of inventing a method able to producing devoted representations impartial of human talent—as if nature might characterize itself with out the mediation of pencil or brush.

The digicam finally fulfilled that function, producing rigorous and detailed visible data. Since then, billions of pictures have been produced, and these photographs now represent the very materials used to coach generative neural networks.

In reality, AI features like an ogre pressured to devour huge portions of photographs with a purpose to produce believable outcomes.

Thus, algorithmic photographic photographs, though derived from the visible heritage of the whole historical past of pictures, carry an plain photographic DNA. For this cause, they might fairly be thought of second-generation pictures.

Roland Barthes as soon as wrote that each {photograph} awaits a textual content. Now the state of affairs is reversed: it’s the textual content that generates the {photograph}.

Boris Eldagsen:

Fontcuberta’s “Barthes reversal” is rhetorically interesting however conceptually shallow. In Camera Lucida, Roland Barthes argues that pictures are unstable with out language. The caption stabilizes the {photograph}. The identical photograph will change its that means with totally different captions.

But Fontcuberta overlooks a vital improvement: prompts usually are not captions. They are directions to a probabilistic system. Moreover, it’s now not merely “text” producing photographs. Multimodal prompting has been commonplace for years. Any enter modality can generate any output modality inside latent area. What collapses right here is media classes.

The “Second-Generation Photography” argument is elegant, nevertheless it rests on a logical error. AI fashions are educated on tens of millions of pictures: that’s true. But that doesn’t make their outputs pictures. What the mannequin inherits is visible fashion, a set of statistical patterns. It doesn’t inherit what defines pictures: a direct bodily relationship between gentle, an actual occasion, and a sensor.

Miles Astray:

“Reverse engineering Fontcuberta’s example and following his argument that favors rank over class, photographs of paintings “could reasonably be considered” second-generation work. But if we began calling that $10 Van Gogh print from the reward store a portray, we “could reasonably be considered” madder than the Dutch grasp himself.

When Microsoft had an AI hallucinate “The next Rembrandt,” and a 3D printer imitate the feel of oil on canvas, we couldn’t name the outcome a “painting” with out placing the phrase in citation marks. It’s not the true deal. In the identical vein, a photorealistic AI picture doesn’t grow to be {a photograph} (identical to a photorealistic portray doesn’t grow to be {a photograph}).

All it takes to cease this purely dialectical carousel round rank and sophistication is frequent sense—we all know intuitively what’s what: work are work, pictures are pictures, and AI photographs are AI photographs, as a result of they’re derived from vastly totally different processes and intentions.”

III. The Validation Problem

Joan Fontcuberta:

This terminological concern—behind which lies a deeper ontological query—got here to the eye of the media when the work The Electrician, belonging to the sequence Pseudomnesia by the German photographer Boris Eldagsen, received the Sony World Photography Award 2023 within the “Creative” class. […]

The Canadian photographer Miles Astray, specializing in nature and journey pictures, reversed the logic of Eldagsen’s motion: he submitted an actual {photograph} to the newly created AI-image class of one other vital competitors, the Color Photography Awards.[…]

Indeed, each instances spotlight an uncomfortable however unavoidable actuality: the dividing line between human creation and that generated by synthetic intelligence is quickly fading, if it has not already disappeared fully. […] Their intention was to disclose the unreliability of validation programs in competitions of this sort.

These could have been minor infractions, however they pointed towards a way more essential concern: figuring out the standing and labeling of photographs, their lineage, their pedigree.

Both initiatives may seem as provocations, however in actuality, they supplied a vital critique: if {a photograph} taken with a digicam may be mistaken for a picture generated by a machine – or vice versa – then we should rethink how we outline the boundaries between photographs, and in addition ideas of authorship, creativity, and visible fact. Rather than making us victims of deception, these gestures present a helpful conceptual shock.

Boris Eldagsen:

What these two incidents really uncovered is that the establishments evaluating the pictures had no coherent framework for telling them aside.

If these instances train us something, it’s this: the credibility of a picture can now not reside within the picture itself. It should reside within the course of—who made it, how, and below what situations of accountability. Documentary authority doesn’t disappear; it migrates. It turns into procedural.

This is exactly why Fontcuberta’s dismissal of course of is problematic.

Miles Astray:

To right all of the false data on this passage—from my photographic focus and the intentions behind my stunt to the title of the competitors I participated in—would transcend the scope of this rebuttal. But it is very important level out that it’s affected by false data. Facts nonetheless matter, whether or not they’re captured in imagery or phrases. In reality, they matter greater than ever on this post-truth epoch. If a textual content on the very subject of “documentary fidelity,” written by an mental with the very best intentions, is riddled with errors, fact is placed on its deathbed.

Admittedly, the idea of fact may be obscure to start with. Universal truths are arduous to seek out, and private truths—tethered to opinions—are ample. Fontcuberta’s hybrid tree is each a lemon and an orange, relying on the way you take a look at it. Opposing views can coexist. The idea of actuality is a bit of firmer than fact if you squeeze it; nonetheless, it stays foremost an idea as effectively. Oranges usually are not inherently orange—their coloration will not be a bodily property however the interplay of sunshine with their floor, which is able to replicate some wavelengths and take up others. Moreover, totally different animal species observe totally different wavelengths of sunshine, perceiving diverging realities whereas cohabiting the identical planet. And if that wasn’t sufficient confusion, actuality collapses right into a mere likelihood perform on the quantum degree.

However, as soon as we return from these meta realms to our human dimension, pragmaticism is of the essence. Society frays if we can not agree on a common material holding it collectively. If we can not agree on sure info, actuality turns into optionally available, with actual penalties. Powered by social media and supercharged by AI, the exponential unfold of disinformation and misinformation is already beginning to erode democracies and societal cohesion all over the world.“

IV. The Doubt Problem

Joan Fontcuberta:

Despite every thing, the elemental concern that troubles each specialists and the general public considerations the credibility of photographs.

Some wonder if a prompt-generated {photograph} will at some point win the World Press Photo award. But maybe the query is wrongly framed.

What ought to actually be questioned is whether or not competitions just like the World Press Photo nonetheless make sense.

We now stay in a visible regime wherein photographs more and more assemble the world slightly than merely characterize it.

[…] Perhaps we should always even be glad about their proliferation, as a result of they remind us of the need of doubt.

Algorithmic pictures reinforces the concept each picture is, inevitably, an phantasm and forces us to rethink the belief we place in photographs.

[…] Photography, subsequently, has by no means really been goal; we merely selected to imagine that it was.

Today, with AI performing as a brand new demiurge, documentary pictures quietly slips between historic narrative and fabricated illustration.

Deepfake applied sciences have opened Pandora’s field of iconography: 1000’s of hyperreal scenes and faces created from nothing flood our screens.

We now not look with a purpose to perceive—we glance with a purpose to doubt.

[…] Every expertise of imaginative and prescient has reshaped how we understand the world.

What we’re witnessing immediately is the transition from optical realism to informational realism—an artificial realism summoned by instructions, texts, and strings of code.

From Greek realism, to Renaissance perspective, to Enlightenment aspirations for accuracy, now we have all of the sudden arrived at a condensed synthesis of all these visible regimes.

And now a single immediate can generate a picture that may as soon as have required centuries of technological evolution.

Boris Eldagsen:

The declare that “every image has always been a fiction” is barely half true—and half-truths are harmful in public discourse.

Every {photograph} is framed, chosen, edited – that’s plain. But a digicam {photograph} nonetheless begins with one thing actual: gentle from an precise occasion, recorded by a sensor. A generated picture begins with statistical inference throughout a database of prior photographs. These usually are not the identical act.

Treating them as equal doesn’t sharpen our crucial considering. Eliminating establishments like World Press Photo doesn’t remedy the issue both. The actual process is to defend accountability: the place a picture comes from, who produced it, and below what situations.

Trust is shifting—from the picture to the method. Provenance, metadata, editorial chains of custody, and clear sourcing grow to be central. The picture is now not proof. The course of is.

What is putting is that Fontcuberta doesn’t tackle the democratic implications of this shift on this chapter. Public discourse is dependent upon visible proof. When all photographs grow to be equally suspect, societies lose a vital epistemic software.

Doubt, sparsely, is productive. In extra, it turns into disorienting – and disorientation is well exploited.

If any picture can simulate proof of occasions that by no means occurred, those that profit most are these least deserving of belief. Blurring the excellence between photographic seize and artificial era doesn’t liberate us from naivety. It supplies cowl for manipulation.

When visible proof turns into a class of common suspicion, the burden of proof shifts in ways in which favour these in energy and drawback these making an attempt to carry them accountable.

The reply is to not have fun doubt as an finish in itself. The reply is to assemble new distinctions: between seize and synthesis, between enhancement and invention, between proof and illustration—and to construct establishments able to sustaining these distinctions.

Miles Astray:

AI as a brand new visible undercurrent received’t wash away bedrock establishments like World Press Photo. It’s within the title: world. press. photograph. Three pillars AI might by no means shake. It can not produce actual photographs of an actual world for actual press articles.

Of course, it’s true that pictures “has never truly been objective.” A photographer‘s selections—like what’s not noted of a body and subsequently not noted of the visible narrative—have at all times rendered accuracy as an approximation, which is why captions should give context to documentary photographs exhibited by World Press Photo.

These are pure limitations that truly improve a photographer’s ambition of documentary accuracy. Doubting the continued relevance of press photographs, Fontcuberta diminishes these efforts by shrugging off vital distinctions of picture creation and equating photographic proof with illustrative exemplification.

As a lot as pictures is perhaps restricted in accuracy, AI is technologically absolutely incapable of recording precise occasions. It has no bearing on such photograph awards apart from to contribute to the notion that they’re extra related than ever.

The assertion “we no longer look in order to understand—we look in order to doubt” is catchy. Unfortunately, sober info can look fairly boring subsequent to such sensational one-liners, which is strictly why the press is struggling to compete for consideration with viral social media accounts. The boring fact is that we nonetheless look with a purpose to perceive—hardcoded considering associated to our survival didn’t change in a single day when LLM algorithms hijacked our brainwork in 2022. What modified is that we have to doubt extra now. And possibly Fontcuberta is considerably proper when he muses whether or not that’s an excellent factor—actually, we might use extra crucial thinkers.

But we’re already midway down a slippery slope right here. Historically, the veracity of photographs was pretty simple to determine. The manipulation of pictures was a cumbersome darkroom course of that took time and talent. There had been few who mastered it and lots of who might debunk it. That steadiness shifted with digital postproduction software program, and absolutely flipped with AI. No matter what number of crucial thinkers we will increase, irrespective of how well-trained they’re, you don’t cease an unchecked flood of AI slop with crucial ideas alone. Institutional guardrails and entrepreneurial ethics should serve civil society to the identical diploma we maintain governments and the personal sector accountable with our voting and buying selections.

If these actors act collectively, Fontcuberta’s “synthetic realism” stays however a catchy phrase that tries to shrink eons of visible historical past—from cave work to pictorial messages flying by means of the cosmos aboard our area probes—by squeezing them into one binary modern age of catchall imagery.

To depict humanity’s numerous instruments and strategies of visible creation as culminating in a synthetic smoothie is a misrepresentation of their evolution: pictures will not be an evolutionary development of portray that changed its predecessor, and AI doesn’t change cameras; these mediums, instruments, and processes coexist, and can proceed to coexist as advanced types of expression, the identical manner lemons and oranges coexist as they succeed their frequent citric ancestor.

---

About the authors: Boris Eldagsen is a Berlin-based photograph & video artist, investigating the unconscious thoughts. In search of the timeless, his visible poetry unites the chic and the uncanny. You can discover extra of his work on his website, Facebook, YouTube, and Instagram.

Miles Astray is an activist artist mixing writing and pictures impressed by sluggish journey. You can discover extra of his work on his website and Facebook.