Just Fake! Why Generative AI Art is a Myth

The year 2018 was a turning point in the debate about AI art. This was when Christie’s offered the Portrait of Edmond de Belamy, the first such work of art for sale, claiming that it “is not the product of a human mind. It was created by an artificial intelligence.” This claim was underscored by the fact that, instead of an artist’s signature, the bottom edge of the painting indicated part of the algorithm that had created it.

The price for the picture skyrocketed at the time. It was sold for a fabulous $432,500, far above the estimated value of $7,000 to $10,000. Nowadays, the capacity of AI to generate images, music, poetry, and even movies, is far more advanced. The perfection of its products, which can no longer be distinguished from human creations, raises the question all the more vehemently of whether AI has surpassed humans in yet another domain that was once considered genuinely human: artistic creativity.

Generative AI art

The Portrait of Edmond de Belamy seems to be a paradigmatic example of generative AI art. Generative AI art has to be distinguished from AI-assisted art. The latter involves AI just as a tool that supports human art creation, comparable to a brush or a typewriter. In generative AI art, in contrast, the artistic achievement supposedly lies solely with the AI, while humans play no or only a minimal role in the creative process. That is, the AI is the artistic agent to whom the artwork should be attributed.

The Portrait was produced by a Generative Adversarial Network (GAN), a deep learning architecture in which two neural networks compete with each other to generate new data. Humans are only involved in providing the training data, programming the system, and curating the results. They play no role in the creation of the images themselves. In the case of Portrait, this specifically means that they did not determine whether a man or a woman is depicted, what the person looks like, what they are wearing, or what colors are used in the picture. For this reason, the Portrait is clearly generative, but is it art?

A Turing test for art?

One possible way of answering this question relies on a kind of Turing test. Applied to art, the Turing test is about determining whether an AI-generated object is recognized as a work of art. If human viewers are not able to distinguish an AI-generated object from an artwork created by humans, it meets the standard for being generative AI art.

Way back in the 1960s, the engineer Michael Noll performed a Turing test with computer-generated graphics in the style of Piet Mondrian, the results of which suggest that it had already been passed at that time. Only a minority of subjects were able to distinguish a computer-generated image from Mondrian’s art, and most even liked the computer image better than the original.

However, the fact that the artistic Turing test can be considered as passed when computers still filled entire rooms and were far less powerful than today’s AI raises doubts about whether the test is really adequate for art. A look at art history teaches us that the suggestion to apply the Turing test to art is naive, since it assumes that works of art can be distinguished from other things on the basis of their external appearance.

From ready-mades to fake art

It has been clear for over a hundred years now that works of art do not necessarily differ from everyday objects in their manifest appearance. In 1917, Marcel Duchamp took a standard urinal, signed it as “R. Mutt,” and presented it at the inaugural exhibition of the Society of Independent Artists in New York. The work, known as Fountain, is considered a milestone in art history. Andy Warhol’s Brillo Boxes (1964) has become no less famous. The sculpture consists of replicas of boxes for Brillo soap pads stacked on top of each other as in a warehouse.

The point of these works is to show that artworks cannot necessarily be distinguished visually from non-artworks. But while Fountain and Brillo Boxes are artworks that look like everyday objects, the AI-generated objects turn the tables. They teach us that there can be objects that look like artworks but are not. For this reason, I call them “fake art” to express that they are counterfeit objects that do not have the expressive and communicative force of artworks.

Authorship and aesthetic responsibility

At this point, “AI art” becomes interesting from an art theory perspective. Art theory is a lens that helps us better understand art and, in particular, what the essential contribution of human beings in producing art is. In my book, I claim that what AI lacks—and will lack for the foreseeable future—is the ability to be the author of an artwork because it cannot be aesthetically responsible. Authorship is required for the status, identity, and value of something as an artwork. That is, having an author is a necessary condition for something to be a work of art, for individuating which work of art it is, and for determining how valuable it is as such.

Time has moved on since the Portrait of Edmond de Belamy and, today, one might argue, AI art is produced through prompts. However, creating a pretty picture through prompting is not enough to create a work of art. It is comparable to commissioning a picture. But since AI itself cannot be the author of a work of art, the outcome is not an artwork either. There may be very ingenious forms of prompting that can count as artistic activity. Yet the most one can get in this way is AI-assisted art.

The myth of generative AI art

The claim that AI can create art is, therefore, a myth in the Enlightenment sense. It is a narrative that is not true but nevertheless influences how people understand themselves and the world. Myths typically involve supernatural forces or deities, and they often support social ideologies. This is the case here, as well. The ideology in question is that AI is better than humans in all areas, even the most human ones, and that it will culminate in a godlike superintelligence. However, behind the scenes are—as so often in the history of mankind—those all-too-human guys who just want power.