The Art of AI: Exploring AICAN and the Future of Creativity

AICAN did not prove that a machine can become an artist. It proved how quickly the art world is prepared to invent an artist when the images, the narrative and the institutional setting are convincing enough.

When AICAN appeared, the most obvious question was also the least interesting one: can artificial intelligence make art?

It was the kind of question the media could easily turn into a headline. It offered a duel between human beings and machines, between the studio and the laboratory, between inspiration and calculation. It also allowed everyone to pretend that the definition of art had remained stable until a computer suddenly arrived to threaten it.

Yet, as the broader history of AI art and its crisis of creativity demonstrates, the production of a convincing image is only the beginning of the problem. Art had already survived photography, industrial production, the readymade, mechanical reproduction, conceptual art, appropriation and countless declarations of its own death. The production of an image by a non-human system was never going to be the real scandal.

The more disturbing question raised by AICAN was different: how little evidence of consciousness do we need before we begin to see intention?

A Machine Arrives, and We Immediately Give It a Biography

AICAN was developed from research conducted by Ahmed Elgammal and collaborators at the Art and Artificial Intelligence Laboratory at Rutgers University. Its name comes from Creative Adversarial Network, a system designed not only to produce images resembling those found in the history of art, but also to avoid becoming too easily associated with an existing style.

This technical distinction matters, but not for the reason usually suggested.

AICAN was not important because it had somehow discovered the mysterious source of creativity. It was important because it translated one of modern art’s most persistent myths into a mathematical objective: the artist must know tradition, but must not appear trapped inside it.

The system was trained on tens of thousands of digitised artworks. From this enormous visual archive, it learned patterns associated with recognised styles and periods. It then generated images that remained close enough to the category of art to be legible, while attempting to escape an obvious stylistic classification.

In other words, it was programmed to look original.

This places AICAN within a much longer history of generative art and algorithmic creativity. Long before artificial intelligence became a cultural obsession, artists had already been creating rules, systems and processes capable of producing forms they did not entirely control.

The real difference is rhetorical. Earlier generative artists usually remained visible as the authors of the system. With AICAN, the system itself was gradually presented as the artist.

AICAN did not experience the burden of influence, the fear of repetition or the embarrassment of discovering that an apparently new idea had already appeared fifty years earlier in a small exhibition catalogue. It did not wake up dissatisfied with its own work. It did not destroy a failed canvas. It did not change direction after a political event, a personal loss or a crisis of conscience.

It performed deviation because deviation had been assigned a value.

AICAN Did Not Learn Art History. It Learned Its Visible Remains

The distinction between learning art and learning images of art is frequently ignored in discussions of artificial intelligence.

An archive such as WikiArt can provide a system with colours, shapes, compositions, stylistic labels and recurring formal relationships. It cannot provide the experience of standing before a work, understanding the conditions under which it was produced or recognising what had to be excluded for a canon to appear coherent.

AICAN encountered art history as a flattened visual field. Paintings became data points. Historical conflicts became stylistic differences. Religious transformation, colonial violence, class struggle, gender, censorship and patronage were compressed into images and labels.

The machine could detect that two works looked different. It could not understand why that difference had once mattered.

This is not merely a limitation of technology. It is also a limitation of the archive we give it.

Every dataset presents itself as a collection of information, but it is also a history of previous decisions. Someone determined which artists entered the archive, which works were photographed, which images were correctly attributed and which categories were treated as significant.

This problem also runs through the work of Refik Anadol, whose installations transform enormous visual archives into flowing data environments. In Anadol’s case, however, the origin and cultural identity of the archive often remain visible as part of the work. AICAN presents a more universalising fiction: art history appears as a single visual reservoir from which novelty can be extracted.

A model trained on art history does not inherit art history itself. It inherits a particular organisation of visibility.

AICAN’s apparent freedom was already enclosed within that organisation.

The Seduction of Stylistic Ambiguity

Many AICAN images appear atmospheric, fragmented and strangely familiar. They suggest landscapes without describing a place, bodies without assuming anatomy, or painterly gestures without the physical event of painting.

The viewer moves through a field of recognition and uncertainty. Something seems to belong to the history of abstraction, but refuses to settle into a precise movement. A fragment may recall Surrealism, another Expressionism, another a decorative digital surface produced by contemporary design software.

This ambiguity is visually effective because it allows viewers to participate. Faced with an unstable image, we begin to complete it. We project intention onto it. We search for references, emotional states and symbolic structures. The less the image declares, the more freely interpretation can circulate.

A similar uncertainty appears in the work of Mario Klingemann, where neural networks generate faces and bodies that seem caught between memory and invention. But Klingemann does not disappear behind the machine. His practice is built around the decisions that shape the system, its failures and the cultural unease produced by its endless variations.

With AICAN, by contrast, the machine itself becomes the protagonist.

This is where the mythology of the artificial artist begins. We often mistake our own interpretative labour for evidence of the creator’s interior life. Because we can find meaning in an image, we assume that meaning must have been placed there. Because a composition appears deliberate, we imagine deliberation. Because the result surprises us, we attribute surprise to the system that produced it.

AICAN did not need to possess an interior world. The viewer supplied one.

The Art World Did the Rest

The machine did not enter art alone.

It arrived escorted by a university laboratory, a research paper, a name, a website, exhibitions, curatorial language, physical supports and a carefully constructed narrative of autonomy. Its images were printed, titled, selected, enlarged and placed on walls. They were presented as individual works rather than as anonymous outputs among thousands of possible variations.

This process is not a secondary detail. It is the process through which images become artworks.

As happens throughout the less visible machinery of the contemporary art world, value does not emerge from the object alone. It is produced through selection, access, institutional authority, critical framing, market positioning and the repeated circulation of a convincing story.

AICAN was described as an “Artificial Intelligence Artist,” but also as a collaborative creative partner. The contradiction is revealing. The first expression creates the spectacle. The second describes the reality more accurately.

Human beings selected the dataset. Human beings designed the system. Human beings chose which outputs deserved attention. Human beings gave the works titles, determined their dimensions, selected their materials, wrote their descriptions and negotiated their public presentation.

Without this chain of decisions, AICAN would not have been an artist. It would have been a generator producing files.

The art world did not simply accept the machine. It completed it.

From Output to Artwork

AICAN’s exhibition history includes presentations in galleries, festivals, technology events and art fairs. Its images were transferred onto materials such as aluminium Dibond and linen. The system was also involved in collaborations where its visual output reacted to musical performances.

These contexts moved AICAN away from the laboratory and toward the rituals through which contemporary art produces value.

A file becomes an edition. An image becomes a titled work. A generated variation becomes the selected variation. A technical experiment becomes part of a collection. A research project acquires the vocabulary of artistic identity.

The contrast with Sougwen Chung’s collaborations with drawing robots is useful. Chung does not present the machine as a mysterious replacement for the artist. The work is built around the visible negotiation between the artist’s body, accumulated gestures, robotic behaviour and computational memory.

In that relationship, authorship becomes unstable without disappearing. The machine changes the conditions of drawing, but the artistic position remains accountable and visible.

AICAN’s framing moved in the opposite direction. The human act of designation was often hidden behind the story of autonomous creation. The system was credited with producing the image, while the decisions that constructed its artistic identity retreated into the background.

The apparent disappearance of the author was itself authored.

Who Actually Made an AICAN Artwork?

The answer depends on what we mean by “made.”

The algorithm generated the visual configuration. The researchers created the conditions under which that configuration could emerge. The artists contained in the dataset contributed the visual histories the system recombined. Curators and collaborators selected particular outputs. Printers and technicians gave those outputs material form. Institutions granted them visibility and legitimacy.

Authorship is therefore not absent. It is distributed.

This distributed authorship is more interesting than the fiction of the machine working alone. It reflects the reality of much contemporary production, where artworks already depend on fabricators, assistants, engineers, programmers, curators and institutions.

The evolving simulations of Ian Cheng, for example, contain systems capable of producing actions the artist cannot fully predict. Yet Cheng remains responsible for constructing the world, defining its agents and deciding how the simulation enters the exhibition space.

AICAN did not invent collective or distributed authorship. It simply made it harder to continue pretending that authorship had ever been completely individual.

Yet the rhetoric surrounding artificial intelligence often performs the opposite operation. It concentrates authorship inside the machine because the idea of an autonomous artificial creator is easier to market than the description of a complex human and computational system.

The machine becomes the signature.

The Market Loves a First

AI-generated art entered the market through a familiar mechanism: the promise of historical novelty.

The appeal was not limited to the visual quality of the works. Collectors were invited to purchase evidence of a technological turning point. The object carried a story about the future, and the story often had more urgency than the image itself.

The market has always understood the value of the “first”: the first artificial intelligence artwork sold at a major auction, the first algorithmic artist, the first machine collaboration, the first autonomous system to exhibit beside human artists.

Such claims generate attention because they transform ownership into participation in history.

A collector does not simply acquire an image. The collector acquires the possibility of saying: this was produced when machines supposedly began to create.

The same mechanism was visible during the rise of NFT art, where technological novelty, digital scarcity and market speculation frequently became inseparable. In both cases, the new infrastructure was sometimes treated as if it automatically produced new artistic meaning.

The risk is that technological novelty becomes a substitute for artistic necessity. The work is valued because it announces a new means of production, even when the image itself remains formally conservative. The machine appears radical while its output often returns to the safest territories of abstraction, beauty and ambiguity.

What changes is not always the artwork. Sometimes only the mythology of its producer changes.

What the Machine Cannot Risk

AICAN can generate an image that viewers consider moving, disturbing or beautiful. But it cannot be moved, disturbed or endangered by what it produces.

It cannot lose a commission because of a political position. It cannot be censored by a government and understand the consequences. It cannot place its body in a public space, expose its biography or recognise that an artwork has failed ethically even when it succeeds visually.

It cannot decide that the correct response to a situation is silence.

This is where the comparison between an artificial system and a human artist becomes most fragile. Art is not only the ability to produce visible difference. It is also the capacity to assume a position in relation to that difference.

The artist is not simply the source of the image. The artist is the person, or collective, that can be questioned about the image, contradicted by it and changed by its consequences.

This distinction is also central to my own experiment, The Painting with an Artificial Intelligence Soul. There, artificial intelligence does not generate the original painting or replace the artist. It gives a pre-existing physical work a voice through which viewers can enter into dialogue with its memory, ecological meaning and material history.

The machine becomes part of the artwork, but responsibility for what it says, how it has been constructed and why it exists remains human.

AICAN has outputs. It does not have consequences in the same sense. The consequences belong to the people and institutions that deploy it.

AICAN as a Mirror of Contemporary Art

Seen from this perspective, AICAN is less an artificial artist than a compressed model of how contemporary art manufactures authorship.

It demonstrates that originality can be described statistically, that style can be converted into data and that visual novelty can be generated without personal experience. It also demonstrates that none of these operations is sufficient to create an artist without a surrounding apparatus willing to name one.

The project reveals how much contemporary art depends on framing. A work requires a context, a discourse, a genealogy and a public capable of reading it. The machine does not abolish these structures. It depends on them more visibly than most human artists do.

AICAN therefore tells us less about the arrival of machine consciousness than about our desire to believe in it.

We want the machine artist because it allows us to stage a dramatic encounter with the future. We can admire it, fear it, buy it or declare ourselves obsolete beside it. What is harder is to admit that the system still depends on recognisably human forms of power: selection, ownership, institutional authority and the management of attention.

Why AICAN Still Matters

AICAN remains important because it appeared before prompt-based image generators made synthetic images ordinary. It belongs to a moment when the production of a convincing artificial image could still be presented as an event in itself.

That moment has passed. Images can now be generated in seconds, in countless styles and at a scale that makes the old question of whether AI can produce something aesthetically persuasive almost irrelevant.

As examined more broadly in From Canvas to Code, contemporary technology has not simply introduced new tools. It has altered the conditions under which images are produced, circulated, experienced and assigned value.

The question is no longer whether the machine can make an image.

The question is who controls the archive, who receives credit, whose labour disappears, who owns the result and who remains responsible for what enters the world.

In this changed context, AICAN can be seen more clearly. It was not the birth of a new artistic species. It was an early demonstration of how easily artistic identity could be detached from a body and attached to a system.

Conclusion: The Artist Was Never Inside the Machine

AICAN did not become an artist by generating images that resisted stylistic classification. It became an artist because researchers, institutions, curators, journalists, collectors and viewers agreed to treat it as one.

This does not diminish the project. It makes it more interesting.

The real achievement of AICAN was not the simulation of creativity. It was the exposure of the machinery through which creativity becomes visible, credible and valuable.

The machine generated the images. The art world generated the artist.

Related Articles

• AI Art Explained: Artists, Tools, Ethics and the Crisis of Creativity
  
• What Is Generative Art? History, Pioneers and Key Ideas
  
• Mario Klingemann: AI Art Between Code and Human Perception
  
• The Painting with an Artificial Intelligence Soul