AI, art, and bullshit : how far can we talk about “AI artists” ?

AI enters the studio

What is a work of art today ?

For a long time, we tended to define art by the mastery of a rare gesture : painting, sculpting, or composing meant knowing how to do what few people could do.

But as early as the 18th century, the philosopher Immanuel Kant showed that the essence of art lies not only in technique, but in genius: a capacity to produce original forms that cannot be derived from prior rules.

Genius, he explains, “gives the rule to art” : it is not someone who applies rules, but someone whose works themselves become examples from which general rules are then abstracted.

In other words, art is not simply what is difficult to do, but what inaugurates something, opening a path that others can follow without exhausting the source.

With modern art, this intuition became radicalized : when Marcel Duchamp exhibited a urinal as a work of art, he showed that the artistic act can be a gesture of thought and designation rather than sophisticated manual labor.

The artist is no longer just the one who executes with their hands, but the one who decides, frames, and interprets reality through a form, inventing problems and answers that no one had seen before.

From this perspective, AI does not immediately disqualify the possibility of art : if the essential elements are intention, vision, and the gesture of meaning, then an algorithmic medium can, in principle, be integrated into the artistic field, provided that the human using it does not simply let the machine “speak” in their place.

Solitude, collective, and the making of works

Art can be solitary : a writer facing their page, a painter in their studio, a musician working night after night on a composition.

But it is also, very often, a collective affair : art history is full of workshops, studios, and fashion houses where the work is made by many hands around a common vision.

During the Renaissance, masters like Raphael directed vast workshops: assistants prepared the supports, painted the drapery, the architecture, and sometimes the characters, based on the drawings and composition decided by the master.

Sources show that certain parts of paintings are likely entirely the work of students, trained to reproduce the master’s style so faithfully that the whole appears to have come from a single hand.

Parisian haute couture updates this structure: legally recognized houses must maintain workshops of “petites mains” (seamstresses) who materialize the silhouettes imagined by the artistic direction, sometimes at the cost of hundreds of hours on a single piece.

At Chanel or Dior, dozens of seamstresses, embroiderers, and pleaters work from sketches, canvases, and precise instructions, but it is the house name that signs the final work, because we recognize a global vision more than every individual stitch.

Cinema further reinforces this logic : a deeply collective art, it brings together screenwriters, actors, technicians, and producers around a work that will nevertheless bear the name of a director, to whom responsibility for the meaning and the overall form is attributed.

Sociologists of art thus speak of “art worlds,” highlighting that creation results from cooperation networks as much as from an individual’s inspiration.

In this framework, AI can be thought of as a new type of collaborator or a virtual workshop : it proposes variations, sketches, and intermediate states that the creator can accept, refuse, or subvert.

The risk is not that it makes art collective—it often already is—but that it makes possible a delegation without intention, where the artist no longer directs the process and settles for endorsing what the model produces.

What AI changes in the regime of images

Artistic bullshit : when form is emptied of meaning

To think about “artistic bullshit,” the distinction proposed by Harry Frankfurt between lying and bullshit is enlightening.

Frankfurt explains that the liar knows they are saying something false and remains defined by their relationship, however conflictual, to the truth ; the bullshitter, however, is indifferent to the truth, caring only about producing an effect, obtaining an impression, or gaining buy-in.

Transposed to art, this distinction designates forms that do not seek to express an inner necessity or a singular vision, but simply to produce a “look” consistent with the expectations of the market, the algorithm, or the public.

AI makes this type of production extremely easy : one can generate highly polished images in standardized styles—fantasy, cyberpunk, “Ghibli-like,” etc.—in seconds, solely to feed a feed or collect likes.

The result is works that are technically impressive but conceptually indifferent : forms that are not necessarily false, but where the author has no demanding relationship with what they show; what matters is that they “work.”

This is where AI can become a powerful amplifier of artistic bullshit : it reduces the entry cost for producing forms, while giving these forms the appearance of originality and virtuosity.

But this risk already existed : advertising, political communication, and certain media formats have long relied on images and narratives produced for effect rather than for truth or necessity.

AI doesn’t create bullshit, it opens the tap wide ; the question is whether we want to use it to intensify this regime, or to saturate, subvert, or make it visible, which is already a possible artistic approach.

Energy and water footprint of algorithmic art

AI gives the impression of pure virtuality : images that only exist on screens, texts circulating online, invisible models in the cloud.

In reality, every request, every generation relies on servers, data centers, and networks whose energy footprint and water footprint are becoming very significant on a global scale.

Regarding electricity, several analyses indicate that data centers could consume 3 to 4% of global electricity by 2030, largely to power the storage and computing infrastructure required for digital services and AI.

Training large language models like GPT mobilizes thousands of GPUs for weeks, with an energy cost that translates into tens to hundreds of tons of CO₂ depending on the energy mix used.

Regarding water, work led by Shaolei Ren and colleagues has shown that a full training of this type of model can lead to the evaporation of hundreds of thousands of liters of fresh water in data center evaporative cooling systems—on the order of magnitude of manufacturing hundreds of cars.

The same team estimates that common use of a GPT-style chatbot can, depending on location and cooling technology, consume the equivalent of several hundred milliliters of drinking water for every few dozen questions and answers.

To this is added the indirect footprint : water used to produce electricity, to manufacture chips, servers, buildings, and the networks necessary for this algorithmic creativity.

In a context where some analyses anticipate a 56% global deficit in available fresh water by 2030 if nothing is done, AI thus becomes a discreet but real player in the water crisis.

Art installations based on giant projections, real-time computing, and immersive experiences powered by AI further amplify this footprint: they mobilize power, cooling, and assembly and transport infrastructure.

The ecological question of digital art is therefore not secondary : it concerns how we embed our works in an economy of limited resources, and the responsibility we accept—or not—as creators.

Paths toward de-alienation and the ecology of creative AI

Faced with this footprint, the challenge is not just to guilt-trip AI usage, but to think of forms of de-alienation: reducing dependence on large platforms, re-anchoring creation in infrastructures we understand, and experimenting with forms of digital sobriety.

On the technical front, several paths are emerging :

• Smaller and open models : research on small language models and open-source models shows that good performance can be achieved for many creative tasks with more compact models that are less hungry for energy and resources than a giant model.

• Edge AI and internal models : moving part of the inference to local devices (personal computers, studio servers, micro-data centers) can reduce data transport costs and allow for finer control of the energy mix, provided we don’t simply reproduce cloud inefficiencies on a small scale.

• On-premise and studio models : recent analyses show that, for some organizations, deploying large language models on internal infrastructure can become economically competitive, while offering leeway to optimize energy efficiency, compute localization, and cooling policy.

On the water front, research is exploring water circularity solutions : reusing cooling water, closed-loop systems, choosing sites where water is less scarce, and, more radically, using the waste heat from data centers to power water purification or carbon capture technologies.

A recent study suggests that by exploiting this heat for the thermal purification of salt or brackish water, some data centers could become “water-positive,” producing more drinking water than they consume, while contributing to CO₂ capture devices.

For an AI artist, these options are not technical details : choosing to work with sober local models, infrastructures powered by renewable energy, and platforms committed to water circularity is a way to re-embed one’s gesture in an ethic of creation.

De-alienating creation, here, means freeing oneself from both the cultural monopoly of digital giants and material ignorance : knowing where the energy and water supporting our works come from, and deciding what we are willing to consume for one more image or text.

Copyright, justice, and the legitimacy of AI artists

Legally, generative models often rely on corpora of works collected online without the authors being explicitly consulted or remunerated, which poses a major problem of justice.

In Europe, the text and data mining framework allows for the automated analysis of protected content, provided that rights holders have not explicitly reserved their rights, opening the possibility of using vast creative catalogs for model training.

Aware of the risks, the European Parliament recently called for stricter rules : the creation of a European register listing protected works used for training, a transparency obligation regarding datasets and scraped sites, an explicit right for artists to opt out, and the implementation of remuneration mechanisms proportional to when their works contribute to a model’s performance.

Rapporteurs emphasize that the lack of transparency and compensation could be equated to copyright infringement and unfair competition practices.

Beyond legality, the question is one of recognition: can we speak of “AI artists” if their practice depends entirely on models trained on the works of other artists who have neither a voice nor a part in the process ?

A fairer approach would consist of designing licensed training models, co-curated corpora, revenue-sharing systems, or explicit collaborations between artists whose works serve as a basis and creators who use them as a medium—a kind of expanded workshop on a digital scale.

For an AI artist who wants to see themselves as such, responsibility includes the choice of tools : preferring transparent models, ethical datasets, and platforms that work toward redistribution is a refusal to base one’s practice on the silent extraction of others’ creative work.

Toward a demanding definition of the “AI artist”

After this detour through Kant, Benjamin, Frankfurt, ecology, and law, the initial question can be reformulated : not “is AI artistic ?” but “under what conditions can a human working with AI rightly claim the title of artist ?”

An “AI artist” in the strong sense would be :

• someone who assumes a clear intention, an inner necessity, and uses AI as a medium to express it rather than as a machine for producing effects ;
• someone who takes material constraints seriously—energy and water—and chooses their tools, models, and infrastructures knowingly, experimenting with forms of frugality and circularity ;
• someone who respects the creators whose work feeds the models, by favoring transparent, licensed solutions, or by building their own corpora in a spirit of a shared workshop rather than extraction ;
• someone who resists bullshit : who refuses to settle for images or texts that just “work,” but seeks to make AI a milieu for critique, play, and subversion, capable of exposing our visual and narrative habits.

Under these conditions, AI can indeed act as a creativity enhancer : it expands the field of possibilities, allows for combinations and rapid explorations, while forcing the artist to confront what, in them, cannot be automated—vision, courage, and responsibility.

But where intention dissolves, where all standards are abandoned, it becomes a machine for producing artistic bullshit, occupying symbolic space with brilliant but empty forms, while consuming energy and water that we will need for everything else.

AI does not kill art; it puts it to the test.

And it is perhaps there, paradoxically, that the work of AI artists begins : not in the naive celebration of the machine, but in the lucid exploration of what it reveals about ourselves.

FAQ – AI, Art, and AI Artists

Sources and further reading

• Immanuel Kant, Critique of Judgment, §§ 46-49 : genius, originality, definition of art by the production of rules rather than the simple application of techniques.

• Walter Benjamin, The Work of Art in the Age of Mechanical Reproduction : notion of aura, transformation of the status of works with photography and cinema, politicization of art.

• Harry G. Frankfurt, On Bullshit : distinction between lying and bullshit, definition of bullshit as indifference to truth, reflection on its proliferation in contemporary societies.

• Works on workshops and delegation in art : studies on Raphael’s workshops and the specialization of assistants, analyses of the collective dimension in artistic creation, reports on the “petites mains” of haute couture (Chanel, Dior), and the cooperative structure of “art worlds.”

• Analyses of the energy and carbon footprint of generative AI : studies on data center electricity consumption and the climate impact of training and inference for large language and image models.

• Studies on the “water footprint” of AI : work by Shaolei Ren’s team on water consumption and evaporation related to data center cooling for training and using GPT-style models; analyses of water volumes used by data centers and circularity issues.

• Paths for ecological and frugal solutions : reports on “water-positive” and “carbon-negative” data centers via the use of waste heat for water purification and carbon capture, practical guides for sustainable generative AI, reflection on model rightsizing.

• Research on edge AI and smaller models : surveys on green edge AI, studies on the energy efficiency of small language models on edge devices and the possible gains of local inference.

• European debates on AI and copyright : European Parliament report calling for transparent registers of works used for model training, the right to opt out, and remuneration mechanisms for creators.

Let’s explore your creative project with AI

Digitalization and AI are shaking up the graphic design profession, but they can also become a real creative springboard: discover how ROMAN3D analyzes these issues and supports graphic design agencies through the digital revolution