Artificial Intelligence Agents are not Artists (2024)

Paint Brushes

I am working on an oil painting that I started months ago; or perhaps it has been a year. When I began at that fuzzy beginning, I had tried a new approach. I had skipped the dead coat that I tend to use to establish the form and values of the composition, and leaped straight to laying in strokes of color. The composition was not right. The values and form were wanting. I struggled to get the desired optical effects that oil paints allow. I now recall why I had decided to wrap the large canvas in a bed sheet and store the unfinished piece those many months ago. I have labored the past weeks to cover the mishap with a white and umber dead coat. I should had done this in the first place.

Artificial Intelligence

Oil painting is my hobby. Applied artificial intelligence (AI) is what I do for a living. And what a time it is to be an artificial intelligence engineer! It is an uncommon day that passes when one does not encounter discourse on artificial intelligence research and practice: A deep learning agent defeats an expert Go champion; Elon Musk, Stephen Hawking and Max Tegmark warn of the perils of unregulated AI; A deep learning agent learns to play Go from scratch, then defeats the deep learning agent that learned to defeat an expert Go champion … Oh, DeepMind, you do not disappoint.

Many graybeards frown at such buzzword bingo. This is understandable: they survived an AI winter that gave Game of Thrones a run for its money. And there is a lot of hype surrounding AI. Still, the facts are that such buzz words are more than hype. Artificial intelligence — and especially data-driven machine learning — is profoundly influencing a diverse range of disciplines, including: agriculture, conservation, and construction; finance, healthcare, and manufacturing; and retail, travel, and utilities. For the most part, we humans benefit.

But are there areas that should stiff arm artificial intelligence?

Artificial Art

Recent developments have me deep thinking: Artificial intelligence that creates art. Artificial intelligence that creates art? Whoa! That is crossing an imaginary line I just drew in the sand.

There are many claims of artificial intelligence creating art. For example, researchers from Rutgers University, Facebook AI Research, and College of Charleston, led by Ahmed Elgammal, have developed a Generative Adversarial Network (GAN) — a type of artificial intelligence algorithm in which two neural nets play off against each other to improve results — to generate (dare one say “create”) original works of art. The preprint is found here.

In the GAN, which the authors call a Creative Adversarial Network (CAN), a generator network creates images and a discriminator network, which is trained on 81,500 paintings, critiques the generated images based on aesthetics. Interestingly, when the CAN images were placed beside contemporary human artworks, human evaluators could not tell which images were artificially-generated. In many cases, the CAN images were rated aesthetically higher than the human artwork.

I will concede that upon inspection, many generated images are artistic and aesthetic. And the research is fascinating.

But are CANs creating art?

Leonardo Di Vinci

Of all artists, Leonardo da Vinci is persistently toward the top of my most admired list. He is the archetypal renaissance man: artist, scientist, engineer. We are all likely familiar with his Vitruvian Man (c. 1490), The Last Supper (c. 1498) and Mona Lisa (c. 1503). Da Vinci’s work is universal.

In his book Art and the Creative Unconscious, Erich Neumann, a Jungian psychologist, had this to ask about Leonardo da Vinci:

In any attempt to come closer to the personality of Leonardo da Vinci, it will be well to bear in mind the words of Jakob Burckhardt: ‘The colossal outlines of Leonardo’s nature can never be more than dimly and distantly conceived.’ And yet this towering figure, great artist and great scientist in one, will always represent a challenge: What was the mysterious force that made such a phenomenon possible?

His answer is that the mysterious force was da Vinci’s psyche. The archetypes of the collective unconscious are intrinsically formless psychic components that assume form in his art, like images and motifs such as the mother and the child. The work of art, and art itself, were for da Vinci not ends in themselves but rather instruments and expressions of his inner situation.

This view applies to most celebrated artists. Like Joseph Campbell’s Hero with One Thousand Faces, successful artists create art that captures the symbols and archetypes that arouse the relationship of the viewer’s ego consciousness to the archetypes in his or her collective unconsciousness. Indeed, Neumann views an artist as a hero in isolation, whose mission is often to oppose the cultural canon of his age. From cave painters to Courbet, it is the human psyche traversing the stages of the hero’s journey that is behind the creative power of the artist.

AI Agents are not Artists!

Artificial art lacks its own intrinsic psychic meaning to the agent. AI agents are not creating art; rather, they are replicating art. For example, the CAN agents were trained on tens of thousands of original artworks created by humans. When a CAN agent generates a new image, it is not drawing upon its personal or collective experiences, neither conscious nor unconscious. It’s generated images are predicated on human experiences, as manifest in the symbols and archetypes captured in our human artwork on which the CAN agent is conditioned and trained.

This explains why humans resonate with the CAN’s artificial art: after all, it is capturing our human experiences, our human condition, our human existence. The CAN agent is not creating art because its generated images are not manifestations of the symbols and archetypes swimming in its own unconscious. If fact, the CAN’s do not have psychic structure.

Well, it’s no Delacroix

Should we expect an AI agent to paint The 28th July: Liberty leading the People? A painting by the same father of modern art, Eugene Delacroix, who stated:

If one means by my Romanticism the free manifestation of my personal impressions, my estrangement from the standards practiced in the schools and my repugnance for academic recipes, I must avow that not only am I Romantic, but that I was already so at the age of 15.

Upon his passing, Delacroix’s biographer and confidant wrote a eulogy about the famed painter that read:

Thus expired, almost smiling, on the 13th of August 1863, Ferdinand-Victor-Eugene Delacroix, a painter of high breading, who had a sun in his head and a thunderstorm in his heart; who over forty years touched every chord of human passion, and whose grandiose, suave and terrible brush went from saints to warriors, from warriors to lovers, from lovers to tigers and from tigers to flowers.

Would we expect such words to be written about an AI agent? The bottom line is that AI agents are not artists because these agents do not have their own psychological experiences. They do not have a collective psychology that developed over millennium. Collective archetypes do not swim in their unconscious pool. Nor do they have an individual psychology that developed through encounters with collective archetypes. Since AI agents are not artists, it therefore follows that their generative products are not art.

Tossing a Bone

Allow me to toss a bone. If anything, current AI art is analogous to the art of the toddler, or the prehistoric cave painter. Much like these, the AI agent is generating symbols and forms whose intrinsic meanings it does not understand. It cannot understand them because it has no fully demarcated ego consciousness.

The source of the symbols and motifs in AI art is hundreds of thousands of years of human evolution that lurks encoded in the parameters of the AI’s art algorithm. Erich Neumann might state the developmental stage of the artificial artist’s psyche is that where the conscious and unconscious are undifferentiated, and in uroboric (the snake that bites its tail) unification. Its ego does not yet exist as a separate entity, so it does not understand the world, nor its relationship with the world. It does not understand its own artistic expressions.

In my opinion, this is giving too much credit to an algorithm, albeit a very complex and non-linear algorithm.

Dr. Ian Malcom

CANs are impressive. And I write a bit keyboard-in-cheek. Still, I can’t help to hear echoes of a conversation from Jurassic Park with which I have often resonated:

John Hammond [Richard Attenborough]: I don’t think you’re giving us our due credit. Our scientists have done things which nobody’s ever done before …

Dr. Ian Malcolm [Jeff Goldblum]: Yeah, yeah, but your scientists were so preoccupied with whether or not they could that they didn’t stop to think if they should.

Maybe this exchange is not as hyperbolic than at first glance. Very few endeavors tap into the human condition more than artistic expression.

Paint Brushes: Reprise

The dead coat of my painting is almost completed. I will finish the piece, eventually; perhaps a year from now. I am not willing to allow an artificial artist to beat me to it. Artificial intelligence might take my factory job but it will keep its hands off my paint brushes, thank you very much.