AI image generation, pollution of our imagination

Analog Photography

Since its invention, many thinkers have tried to theorize photography.
What is photography? What are its unique characteristics? Is it different from other visual arts?

Roland Barthes (Camera Lucida, 1980) described himself as having an "ontological" desire regarding photography: a desire to understand what it was "in itself" and what essential trait distinguished it from the community of images.

Some, like Walter Benjamin (The Work of Art in the Age of Mechanical Reproduction, 1936), believed that the technical reproducibility of photography, while democratizing visual representations that were previously reserved for a social and cultural elite and now accessible to a broader audience, nonetheless destroyed the uniqueness and aura of traditional artworks, which, when exhibited in a specific place, create a particular relationship with the viewer.
Others, like André Bazin and Roland Barthes, thought that photography's particularity—its ability to mechanically reproduce reality—somehow allowed it to immortalize reality.
Roland Barthes introduced the concepts of Studium (the general or cultural interest a photograph evokes) and Punctum (the personal emotion triggered by an unexpected detail in a photograph) to distinguish photography.
Vilém Flusser (Towards a Philosophy of Photography, 1983) argued that photographic devices impose choices and creative limits on the photographer due to their technical constraints.

More recently, Philippe Dubois (From the Image-Trace to the Image-Fiction, 2016) applied the concept of possible worlds, developed notably by Thomas Pavel in relation to literary works (The Worlds of Fiction, 1988), to photography.
In literature, possible worlds refer to the infinite number of alternative universes that can be imagined from a narrative.
Photography is no longer merely a representation of reality but an autonomous space, either narrative or conceptual, that contains alternative realities, hypotheses, or fictions.

It is nevertheless possible to argue that the advent of photography before the digital age, beyond the genuinely distinct characteristics of earlier representational tools (engraving, mosaics, tapestries, painting, lithography...), did not fundamentally change the relationship between creators and spectators with images.

Is the intimate aesthetic experience of a viewer in front of the uniqueness of a work of art in an exhibition space fundamentally different from the aesthetic experience one might have when viewing a photograph reproduced in large quantities and accessible to the masses?
Is a photograph more of a certificate of authenticity of a "that-has-been" than a painting? Or is it above all a subjective construction of a "that-has-been," which, in reality, "was" only for the time it took to stage the photograph?
Are Roland Barthes' concepts of Studium and Punctum not applicable to all visual representations?
Do the technical constraints of photographic devices influence the creative process more than the technical constraints of engraving or painting?
Is the concept of possible worlds not applicable to sculpture? Does not all representation allow for the exploration of alternative realities?

Digital Photography

A new question arises alongside the ontological debates with digital photography.
Who is photographing?

While it could be argued that photography, for both creator and viewer, presented little difference compared to other modes of visual representation before the digital era, digital photography introduces a significant change: Taking a photograph becomes easy.
According to Walter Benjamin, the technical reproducibility of photography made visual representations accessible to a broader audience. Digital photography, however, democratizes and makes the very process of image creation accessible. Photography, like other modes of representation, was once reserved for professionals and experienced enthusiasts. With digital photography, taking a picture becomes accessible to everyone.

Viewers before the digital era of photography now become actual producers of images.
The learning and understanding of production techniques, the financial constraints, and the time required to prepare and produce an engraving, painting, or analog photograph are replaced by ease of use, invisible production processes, apparent gratuity, and instantaneity.

Photographs, produced and consumed massively through cameras and media, saturate our daily environment. They are no longer merely tools at our service but also, perhaps, mechanisms that unconsciously influence our actions and choices.

The nature of photographic intention, practices, and aesthetic experience also changes.
Initially used for documentary, artistic, scientific, advertising, or educational purposes, photography today is primarily used for personal and social purposes. Selfies, for example, far from the artistic, aesthetic, or symbolic intentions of traditional self-portraits, allow individuals to stage themselves in a place and share a moment, an emotion, or an experience with friends.

The need for chemical development in analog photography imposed a technical delay, a temporal gap between the moment of taking the photograph and the discovery of the photographed moment. However, digital technologies and built-in screens in cameras now enable simultaneity between the lived moment and the photographed moment, a capability that everyone can appropriate.
The experience of a place, an encounter, or a lived event becomes diverse. The instant verification of a lived moment through photographic recording encourages individuals to experience new things and pay attention to the present in ways that were different before the digital era.

AI Image Generation

As with photography, beyond the ontology of AI-generated images, it is important to question the mechanisms of production and the new practices.
Who generates these images, in what context, and for what purpose?

The focus should not be on questioning the quality or originality of AI-generated images, nor the aesthetic appreciation of a visual representation, as this is subjective and belongs to each viewer. Instead, we must question the impact of these images on our imagination.
Moreover, as Serge Tisseron states in the Manifesto for an Inferior School of Photography, there are no "rich" or "poor" images.

AI image generation tools allow anyone to produce images more massively than ever, without any learning or skill. Our native language is enough; simple textual instructions (prompts) suffice for anyone to create images. Moreover, it is no longer necessary to leave one's home, experience something real, or press a button in contact with reality.

Text Generation VS Image Generation

AI in creative domains seems to be becoming widespread, but there is a significant difference between text and images.

Language models like ChatGPT allow users to simplify, accelerate, or automate an activity that already existed before these models, such as finding information or drafting a text.
Novelists, journalists, administrators, lawyers, and software developers continue their usual activities, but differently: they outsource and delegate part of their expertise to a language model in the name of "optimization."

The case of image generation systems appears different.
Artists, painters, illustrators, graphic designers, and photographers are not the users but rather, unbeknownst to them, the upstream creators of the content that feeds the databases necessary for these systems to operate.

Users of generative image systems, like users of digital cameras, are no longer just enthusiasts or image professionals. For various reasons—lack of interest, skills, or time—many had never considered creating visual universes, designs, objects, architecture, or landscapes before the emergence of image generation tools.

Examples abound.
A company manager looking for a logo no longer relies on the expertise of a graphic designer but selects their logo from the graphic proposals generated by an image generation system.
An art director at an advertising agency no longer hires an artist to create a visual for an advertising campaign but generates the visual themselves.
Furthermore, more and more individuals reinvent themselves as artists or designers and sell artistic works such as Porter Art Guild or furniture like Futuristic Decors, entirely generated by AI.
Aspiring art directors like Jonas Peterson or Polina Kostanda have never been so numerous.
The proliferation of web pages entirely written and illustrated by AI to enhance visibility on the Internet is staggering. Cleaning companies like Écolavage create pages comparing fictitious vacuum cleaners. Travel agencies like X Holidays create Instagram accounts featuring fake idyllic houses.
Individuals flood the internet with AI-generated images hoping to sell T-shirts (Nordikido) or mugs (IllustNation).

It should be noted that users of text-generative systems outsource their cognitive skills, risking a loss of this know-how—a loss all the more alarming in the case of young learners who have not yet fully acquired these skills. However, this is not the case for users of image-generative systems: They do not outsource their creative skills because they lack such skills and therefore do not risk losing them.

Users of image-generative systems also often lack a critical perspective on the images these systems generate based on their textual requests. Is the generated image truly original? What meaning or impact might it have? Does it maintain a particular relationship with existing works? Does it reference a specific period in art history? The apparent graphic quality of generated images often obscures these questions.

Creators vs. Consumers

Since the public release of AI-assisted image generation systems in 2022, the production of images has grown exponentially.
It is worth noting that neither AI image generation tools nor social media provide any data on the volume of generated and shared images, which is nevertheless estimated at several billion.

But why all these images?

The ease of producing AI-generated images and their effortless dissemination on the Internet, without the need for institutional recognition such as that of a museum or a journal, is undoubtedly a key factor.
As with digital photography, we produce images simply because it is possible to do so.

But beyond the ease of use for personal or professional purposes, or the fascination with the apparent artistic quality of these images, the mode of production itself seems to be the main driver of their proliferation.
We no longer create images in the traditional sense, as was still the case with digital photography; instead, we commission images in the form of textual requests to an algorithm.
The mode of production for AI-generated images seems to operate as a form of consumption.

We provide an algorithm with keywords and visual references without having a precise idea of the outcome, then rephrase these keywords until a satisfactory result is achieved.
Our consumer society has accustomed us to addiction and the pleasure of achieving satisfaction.
Could the pleasure of the consumer be one of the motivations behind the production of AI-generated images?

A Sense of Creation

AI image producers tend to interpret this consumption as an act of creation.
The language used explicitly reflects the appropriation of these images as creations.
For instance: “Look at what I created,” “This is my work,” “I conducted many tests and combinations of keywords before creating my image...”
“With this tool, we draw our illustrations three times faster!”
There is also a sense of having learned and acquired a skill: “I know how to prompt to get what I want.”
The language of viewers is equally revealing: “I love your work,” “You’re my favorite artist...”

However, it can be argued that, due to their mode of production, AI-generated representations differ from traditional visual representations, including digital photography, in that all participants—viewers and producers—merely become consumers. In exchange for a financial contribution, such as a subscription to an AI service, we obtain, through image consumption, the satisfaction of our desires.

It is worth noting that the claim to creator status is problematic, as copyright legislation for AI-generated images remains inadequately defined. Who are the authors? The users who draft the textual prompts? The developers of the systems and algorithms? The artists whose works were used to train AI systems and generate the images?
Many countries, such as the United States, specify that works created by artificial intelligence cannot be granted copyright protection without significant human contribution.

For example, the United States Copyright Office stipulated in March 2023, in its Guidance for Registration of Works Containing Material Generated by Artificial Intelligence, that it will not register “works produced by a machine or mere mechanical process that operates randomly or automatically,” but only “works that are fundamentally human (with the computer serving merely as an assistant), where the traditional elements of authorship (composition, selection of colors and shapes, etc.) were actually conceived and executed by a human, not a machine.”
The quality of textual prompts used to automatically generate images therefore becomes essential in defining the notion of authorship. Image generation systems make users’ textual prompts public, thus making it easier to evaluate their “creative contribution.”
Examples of textual prompts from MidJourney include: “a swarm of magical insects with luminous wings,” “a small tree by the riverside with a blurry background,” and “a disco samurai robot, official royal portrait.”
It appears that today, many generated images do not result from a sufficiently significant creative contribution to consider users as authors; consequently, these images fall under the public domain.

Toward a 7th Digital Continent?

Just as excessive consumption of physical goods causes pollution, the consumption of images leads to pollution as well.

A material pollution that degrades our physical world, as the production and operation of the infrastructure required for AI generation systems rely on significant resources, including rare metals such as cobalt and lithium, and consume vast amounts of electricity and water for server cooling.

A visual pollution that alters our imagination, as the fictitious images generated, seemingly original and astonishing, overwhelm us and saturate our minds.
Will generative AI systems create a growing disinterest in unique images?
The choice of themes and subjects requested from an AI algorithm plays an important role.
The extraordinary and poetic image of a solitary dwelling in a spectacular natural setting, such as a massive iceberg, once a source of wonder, becomes ordinary and mundane when faced with thousands of similar images generated daily by AI. The striking and emotional expression of a wild animal caught off guard by a camera no longer surprises or moves us.

Furthermore, the probabilistic functioning of image generation systems avoids atypical combinations and favors the most common ones. The results obtained, though seemingly original, converge toward an average.
Originality becomes predictable, homogeneous, and mundane.

The overconsumption of AI-generated images trivializes the unusual and makes us insensitive to mystery and uniqueness.
Will we never be surprised again?

Images We Didn't Want

It should be noted that, unlike other visual representations, the random mode of AI image generation implies that the results never exactly match our expectations. We generate images that surprise us because they differ from what we had in mind, but their apparent originality and graphic quality entice us to use and share them.
“It's not what I wanted, but hey, it's not bad, right?”

Can We Depollute?

Cleaning up our planet seems difficult, but hope remains: energy transitions, waste management, recycling, reforestation, ocean and soil rehabilitation, environmental awareness, and responsible consumption.
Is it possible to clean up our imagination?
The "false real photographs" are invading our imagination. The pervasive doubt about the truthfulness of an image gives way to habit. Our minds disengage, and we accept this new reality generated by algorithms, depriving us of the poetry of visual representations. Are the consequences already irreversible?
Will images ever make us dream again?

A Small Note of Optimism

Does AI image generation pollute our imagination to the point of depriving us of our capacity for wonder?
Or, on the contrary, from an optimistic perspective, will the proliferation of extraordinary fictitious worlds and the growing pollution of our imagination—just like the proliferation of real goods and the pollution of our planet—encourage us to look at the beauty of the ordinary with more sensitivity, care, and attention?

Will the saturation of AI-generated images lead us to focus, not on the result, but on the real lived experience, to escape the virtual and once again live physically and poetically in our world?