The End of an Illusion

Legal disputes regarding copyright in the training of image- or art-generating algorithms have not yet been resolved. But regardless of the outcome of these proceedings, this development will challenge and redefine the role of artists.

Above all, however, it will challenge the Enlightenment-era conception of art. Until now, this conception has—perhaps unjustifiably—placed the artist as an individual at the forefront.

Are we currently witnessing the beginning of the end of the cult of personality in art?

Machine-generated art is entering museums. When the Mauritshuis in The Hague loaned Johannes Vermeer’s “Girl with a Pearl Earring”—arguably its most famous work—along with about a dozen other paintings by the Baroque artist, to the Rijksmuseum in Amsterdam for three months in early 2023, the museum filled the prominent vacancy with an imitation. What makes this special is that it wasn’t an artist who created this interpretation of the original—but the “AI Midjourney” algorithm. Julian van Dieken used artificial intelligence to create his version, titled “Girl with Glowing Earrings” (see Figure 1), which he then further edited in Photoshop [1].

The Mauritshuis had announced a competition for the duration of the absence of its most famous Vermeer. The public was invited to create “their own girl,” inspired by the Vermeer painting, to hang in its place. A jury then judged the total of 3,482 submissions.

A public outcry ensued. Artist Iris Compiet described the sheer number of AI-generated images submitted as an “incredible insult” [2]. The Mauritshuis responded with a statement from a spokesperson, which read: “We knew that [van] Dieken’s work was created using artificial intelligence, but we liked his image, and that was enough for us.” And further: “Basically, the only criterion was that the work was based on a creative process.”[2]

This terse and arguably naive justification angered other artists. According to Eva Toorenent, an artist and Dutch advisor to the European Guild for Artificial Intelligence Regulation (EGAIR), the museum “did not understand the technology.” The “museum’s uninformed response to the criticism was the most disappointing thing,” said Toorenent. “I explained in a detailed email why AI-generated art, in its current form, is highly unethical and has no place being highlighted and celebrated in a museum.” In response, she received “a copy-paste reply in which they reaffirmed their decision.”[2]

At the same time, the Museum of Modern Art in New York dedicated the exhibition “Unsupervised” to Refik Anadol. The artist’s installation “explores what a machine might dream of after viewing more than 200 years of art in MoMA’s collection,” according to the museum’s website [3]. A section of the installation can be seen in Figure 2.

Here, too, criticism rained down. For example, from Jerry Saltz, a renowned New York art critic. For him, Unsupervised is “crowd-pleasing mediocrity.” The installation merely uses “existing written, photographic, and artistic material.” In his view, MoMA’s foray into AI fails “exactly where other AI programs have failed.” He goes on to say: “If AI is to create meaningful art, it must bring its own vision and vocabulary, its own sense of space, color, and form. Things that Unsupervised lacks.” [4]

Saltz is right. Ultimately, this is art from the AI can. The generative AI models, which supposedly create artworks independently, draw on hundreds of millions of images and thus (also) on the artistic legacy of many generations.

And they are becoming increasingly popular, with the market growing ever larger. In addition to established services such as AI Midjourney, Supermachine, Dream Studio, Artsmart, Neuroflash, Nightcafe, and many others, over 100 new providers entered the market this year [5]. Established software companies are also responding to the trend. For example, Microsoft has integrated an image generator into its Bing search engine [6].

It is questionable, however, whether these applications are even legal. Copyright issues certainly play a decisive role. Ultimately, courts will decide on the admissibility of these algorithms. The first lawsuits have already been filed.

On January 13, 2023, illustrators Sarah Andersen, Kelly McKernan, and Karla Ortiz filed a class-action lawsuit in the Northern District of California against Midjourney, DeviantArt, and Stability A.I. DeviantArt is behind the image generator DreamUp, and Stability A.I. has launched the Stable Diffusion algorithm. They describe these text-to-image platforms as “collage tools … that infringe on the rights of millions of artists” [7].

“The defendants use copies of the training images,” the complaint states, “to generate digital images … that are derived exclusively from the training images and add nothing new,” yet “significantly negatively impact the market for the plaintiffs’ work ….” In addition, Anderson, McKernan, and Ortiz complain that AI generators have made it possible to generate artwork “in the style” of certain artists, thereby “skimming off commissions from the artists themselves” [7].

A few weeks later, on March 2, 2023, the stock photo provider Getty Images also sued Stability AI. The image database accuses the company of misusing more than 12 million Getty photos to train its Stable Diffusion AI image generation system. [8]

Essentially, these and other similar lawsuits revolve, on the one hand, around the question of when copying ceases and independent art emerges through alterations and the new composition of works. And on the other hand, whether machines have the same rights as humans.

The latter question was raised by renowned AI expert and researcher Andrew Ng. Ng was a professor at Stanford University and co-founder and director of Google Brain before launching the online platform Coursera together with Daphne Koller. He is also the founder of the AI companies Landing AI and deeplearning.ai.

In his newsletter “The Batch” dated February 8, 2023, he wrote in response to Getty Images’ announcement of the lawsuit: “The last time I visited the Getty Museum in Los Angeles, California, I saw aspiring artists sitting on the floor, copying masterpieces onto their own canvases. Copying the masters is an accepted part of an artist’s training.”

“By copying many paintings, students develop their own style. Artists, too, regularly look at other works to find inspiration. Even the masters whose works are studied today learned from their predecessors. Would it be fair for an AI system to learn in a similar way from paintings created by humans?” [9]

Ng posted a photo showing artists copying Edgar Degas’s painting “The Dance Class.” It can be seen in Figure 3.

These questions will not only occupy the courts but will also shape public discourse and perception in the near future. And this is by no means limited to the issue of equal opportunity between humans and machines. The copyright issues surrounding AI-generated images call into question the very foundation of our perception of art, which underwent a dramatic transformation during the Enlightenment.

Until well into the early modern period, art was above all one thing: anonymous. The creators of early works, regardless of the form of expression, are usually not recorded. Who created the Stone Age cave paintings, see Figure 4? Who carved the world-famous “Nike of Samothrace” shown in Figure 5 into stone? Who wrote the Bible?

Figure 5: “Nike of Samothrace,” front view (left), three-quarter view (center), and side view (right), ca. 190 B.C., Louvre, Paris

Okay, you might say, “fair point” when it comes to the Bible. But Homer wrote the Iliad and the Odyssey. Yes, he did. But as a chronicler. The first oral versions likely originated as early as the Late Mycenaean period. They were passed down by singers over the centuries and were probably altered along the way. And the Middle Ages, too, did not recognize the profession of the artist.

One might argue that the names of these artists may have been forgotten over the centuries. But this inevitably raises the question of why they were forgotten. After all, we know almost everything about the rulers of the Roman Empire and a great deal about those of the Hellenistic and Babylonian empires, as well as ancient Egypt. We know their names, just as we know those of the popes and earlier religious dignitaries. Even economic figures—that is, merchants—as well as lawyers, swindlers, prostitutes, and ordinary citizens are known to us by name from very early eras. Only who the artists were do we not know.

There may be a few exceptions, but before the Enlightenment, art was much more than that. It was the creative expression of a collective, not of an individual.

This changed dramatically with the Enlightenment. In 1641, René Descartes formulated one of the most influential phrases in human history: “Cogito, ergo sum”—or in English: “I think, therefore I am.” This phrase shaped the Enlightenment and individualism, and thus the worldview that, at least in the West, remains valid to this day: It was the realization of our consciousness, articulated in these five words, that in turn strengthened the position of the individual, granted them rights, and also elevated artists—as individual creators—above the diverse forms of expression of a society composed of nameless individuals. It marked the beginning of a cult of personality unlike anything previously seen in art.

Rembrandt was already a sort of painting rock star in Amsterdam by the mid-17th century. And he remains so to this day. Art lovers are happy to pay more than 100 million euros for his works. But woe betide if one of his paintings is dismissed by certain art-historical opinion leaders. Then that very same artwork is suddenly worth only a few hundred thousand euros. But why, exactly? The painting is, after all, completely unchanged.

Essentially, this is the manifestation of a completely exaggerated cult of personality in art, which seems all the more absurd when one considers that some paintings from the Rembrandt school have, in the past, been attributed to eight different artists by the respective top Rembrandt experts—and the authorship of numerous works remains disputed to this day [10]. Or when one considers how many forgeries have found their way into certain catalogues raisonnés, such as those of Heinrich Campendonk, Alexej von Jawlensky, or Max Ernst, to name just a few.

For art historian Bernd Lindemann, former director of the Gemäldegalerie Berlin, the authorship of a work of art plays, at best, a secondary role. “A work of art is a work of art,” he says, “regardless of who created it. The creator’s name has merely a mercantile significance” [11].

Yet the issue raised by the authorship dispute runs much deeper.

It is not merely a question of whether art exists or can exist independently of the context of the artist as an individual (though certainly within the context of the political and social conditions at the time of its creation).

It also raises the question: How much of Rembrandt is actually in a painting by Rembrandt, and how much of Monet in a painting by Monet?

Generative AI models that create works on their own—let’s leave aside for now whether this constitutes art—draw upon the creativity of both well-known and long-forgotten painters. They plagiarize their styles and mix them freely and unrestrainedly.

Just like the artists. Raphael studied and sketched Leonardo da Vinci’s “Mona Lisa” in his studio (Figure 6). Rembrandt studied Raphael’s “The Last Supper”—and copied it (Figure 7). Édouard Manet reinterpreted Rembrandt’s “The Anatomy Lesson of Dr. Tulp”—and was strongly influenced by Diego Velázquez’s style (Figure 8).

Figure 6: Raphael Sanzio’s sketch of the Mona Lisa, drawn around 1504 in Leonardo da Vinci’s studio, Louvre, Paris (left). The original painting, which da Vinci likely altered later in life, Louvre, Paris

Figure 8: Edouard Manet’s interpretation of Rembrandt’s painting “The Anatomy Lesson of Dr. Tulp,” private collection, location unknown (left), original painting by Rembrandt, Mauritshuis, The Hague (right)

And there are countless other examples. Would Caravaggio even be conceivable without Da Vinci, Raphael, and Michelangelo? Probably not. And what would Picasso be without Paul Cézanne, Edgar Degas, Juan Gris, and Pierre Bonnard?

So is art really as individual as we believe? Aren’t works (whether art or other works) also the product of dozens, hundreds, or thousands of other works by other people or artists? How much individuality is there in a work of art—and how much does the artist owe to his or her cultural heritage? And where, pray tell, is the difference between a machine that draws upon the artistic heritage of the Western world—or even of all humanity—and a human being who does the same?

Or to put it another way: Is the individual freedom and artistic self-determination posited during the Enlightenment ultimately nothing more than a chimera? A dream that will never become reality—or can never become reality?

Everyone must find the answer to this question for themselves. My answer is: No.

No, the (artistic) freedom of the individual is not a utopia. We humans are indeed capable of thinking and creating something new—and not just rummaging through the dusty archives of past works. To what small extent, let that remain open for now.

In every stylistic era, there are pioneers and artists who, like Claude Monet, for example, have shaped Impressionism. But they were never—or only in the rarest of cases—individuals. Most often, they were groups of artists who inspired and also copied one another. Which further diminishes the individual’s innovation in the development of a new painting style. Machines and algorithms are not capable of this. At least not yet.

And in every stylistic era, there were also copyists. Artists, in other words, who placed no value at all on individuality, on a “style of their own.” Christian Wilhelm Ernst Dietrich (1712–1774), for example, who later renamed himself Dietricy, limited himself to copying well-known Dutch painters, including Rembrandt. In most cases, however, he merely imitated the style, not the subject matter. The same applies to Tom Keating. The British artist and restorer became known primarily for his imitations of Edgar Degas. Keating is often referred to as a forger—though he never signed the works of other artists that he stylistically emulated as such, which is why he was never convicted.

Unlike Han van Meegeren, whose works are also considered art today—and whose paintings have been shown in numerous exhibitions and retrospectives. The Dutch forger (1889–1947) was initially somewhat successful as a painter; later, his works were dismissed as “kitsch,” and art experts accused him of being incapable of independent creative achievement. Consequently, van Meegeren produced numerous forgeries of Dutch masters, including half a dozen in the style of Jan Vermeer. His imitations were so good that even renowned art historians, such as Abraham Bredius, then director of the Mauritshuis, fell for them [12].

But if the works of plagiarists, imitators, and even forgers are considered art—why should machine-generated works be treated any differently? And what about the other artists who do indeed value their own style—even if that style would be unthinkable without the earlier work of generations of other artists?

Essentially, when it comes to AI-generated art, judges must not only decide whether the use of works by artists, imitators, and forgers in generative models infringes on the copyrights of these creators.

In any case, the new technology calls into question the existing social consensus regarding the achievements of individuals and, consequently, the normative justification for individuals’ rights to “their” works.

And perhaps judges must therefore rule not only on the machines—let us leave aside for now whether their “output” constitutes art or not—but also on whether copyright law, in its current form, is still appropriate in light of technological developments. noch zeitgemäß ist.

Bibliography

[1] O.A., „Mauritshuis hangs artwork created by AI in place of loaned-out Vermeer”, NL Times, 22.2.2023, accessed via https://nltimes.nl/2023/02/22/mauritshuis-hangs-artwork-created-ai-place-loaned-vermeer

[2] Taylor Michael, „Museum under Fire for Showing AI Version of Vermeer Masterpiece”, Hyperallergic, 1.3.2023, accessed via https://hyperallergic.com/805030/mauritshuis-museum-under-fire-for-showing-ai-version-of-vermeer-masterpiece/

[3] Museum of Modern Art, accessed via https://www.moma.org/magazine/articles/821

[4] Jerry Saltz, „MoMA’s Glorified Lava Lamp. Anadol’s Unsupervised is a crowd-pleasing, like-generating mediocricy”, Vulture, 22.2.2023, accessed via https://www.vulture.com/article/jerry-saltz-moma-refik-anadol-unsupervised.html

[5] Finn Hillebrandt, „Die 14 besten Bildgeneratoren in 2023 (kostenlos)“, accessed via https://www.blogmojo.de/ki-bildgeneratoren/

[6] Patrick Hannemann, „Kostenloser Bildgenerator in Bing: Microsoft schenkt Ihnen KI-Tool“, Chip, 6.5.2023, accessed via https://www.chip.de/news/Kostenloser-Bildgenerator-in-Bing-Microsoft-schenkt-Ihnen-KI-Tool_184712934.html

[7] Min Chen, „Artists and Illustrators are Suing Three A. I. Art Generators for Scraping and ‘Collaging’ Their Work Without Consent”, Artnet News, 24.1.2023, accessed via https://news.artnet.com/art-world/class-action-lawsuit-ai-generators-deviantart-midjourney-stable-diffusion-2246770

[8] Blake Brittain, “Getty Images lawsuit says Stability AI misused photos to train AI”, Reuters, 6.2.2023, aaccessed via https://www.reuters.com/legal/getty-images-lawsuit-says-stability-ai-misused-photos-train-ai-2023-02-06/

[9] Andrew Ng, „The Batch“, Newsletter von deeplearning.ai, 8.2.2023 (Page 2)

[10] Reuter, Wolfgang: Original oder Schüler? Einsatzmöglichkeiten Künstlicher Intelligenz bei Zuschreibungsfragen am Beispiel des Rembrandt Research Projects. In: Kunstgeschichte. Open Peer Reviewed Journal, 2023 (urn:nbn:de:bvb:355-kuge-600-4)

[11] Personal conversation with Bernd Lindemann during an event organized by the Kaiser Friedrich Museum Association on June 5, 2023, at the Gemäldegalerie Berlin.

[12] Abraham Bredius: An Unpublished Vermeer. In: The Burlington Magazine for Connoisseurs, Vol. 61, No. 355 (Oct., 1932), Burlington Magazine Publications Ltd., S. 144–145.