Art by Machine

December 8, 2022 / Ellie Stephan

Artists are breathing life into code. And philosophy has a lot to say.

Inside an ornate gold frame, a painted man resides. His face is an undefined mass above his collar. The style of his portrait is classical, formal, and reminiscent of the 18th century, but the painting is fresh. Upon further examination, we find he isn’t painted at all — he’s composed of tricky pixels, coming together to form the out-of-focus man. The bottom right corner of the canvas explains why; there, signed with cursive Gallic script, is an algorithm.

In 2018, a painting called “Portrait of Edmond Belamy” sold for $432,000 at an auction house in New York. The work wasn’t born of brushes and paint; it was created via artificial intelligence, using a Generative Adversarial Network, or a GAN. The GAN, consisting of a generator and a discriminator, was trained on 15,000 paintings ranging from the 14th to the 20th century. The generator produced new images, and the discriminator compared them to the dataset of given paintings. Once the discriminator couldn’t tell the difference, “Portrait of Edmond Belamy” was on the market.

As the first AI-generated piece sold at an auction house, it rattled the boundaries of art. Its creators, an art collective called Obvious, seek to explore the limits of AI’s creativity. When you click on their website, Picasso’s words stamp the home page: “Computers are useless. They can only give answers.” Flippantly, Obvious responds, “Well, Picasso, we disagree.” Entering a prompt into a model isn’t artistic expression as we know it, and some artists feel it should stay that way. But without the label of art, it’s difficult to define what “Portrait of Edmond Belamy” is.

The theory of significant form attempts to explain what differentiates art from ordinary objects. Philosopher Clive Bell explains that all visual works of art share one common trait — significant form, or elements like lines and colors combined in a certain way that stirs our aesthetic emotions. We feel aesthetic emotions when we react to visual elements.

When my friend toured the Opéra Garnier, for example, and saw the ceiling painted by Marc Chagall, he bawled because of its beauty. But art need not be beautiful to have significant form; it simply must provoke emotion. Viewers find “Portrait of Edmond Belamy” outrageous, humorous, and thought-provoking; the portrait brims with aesthetic emotion. Sure, it’s no Chagall, but the two works share significant form.

Plato agrees: Art encourages emotion. He speaks about this with the term mimesis, with which he claims that all art is imitation. He believes that art reflects reality, stretching and dramatizing it in a way that clouds reason and stirs emotion.

In Monet’s famous “Water Lilies” series, he imitates his beloved pond. At Giverny, a willow tree slouches lazily in the summer heat, its branches grazing the surface of the pond. It’s quiet, but not suffocatingly so — birds chirp here and there, and insects buzz enthusiastically. Dark water gazes up at the surrounding canopy of trees, reflecting their greenery. Water lilies, dotted with cheery pink flowers, break up this mirror. It’s a lush marriage of blue, purple, and green, and it begs to be painted. Complying, Monet picks up his brush.

Monet’s water lilies, composed of brushstrokes and paint, are mimesis of the physical water lilies at Giverny. He’s stretching reality, portraying it in a storied way that speaks from museum walls. And he’s not alone in his imitation.

Rapid-fire, keyboard keys click click click, pulling up a photograph of water lilies. Another search query, and the laptop screen floods with color. Impressionist paintings fill Google Images, and one is saved to downloads. With bated breath, the photograph and painting marry through neural style transfer. And with the click of a button, we’re transported to Giverny.

Neural style transfer, or NST, is an AI technique of entering one content image and one style image into a model to produce the content image in the given style. Want your thirst trap to look like something out of the Met? Done! Welcome to the 21st century.

When Monet rendered his pond in the Impressionist technique, he chose an aspect of reality and reflected it in a given style. The process of AI art is similar: Artists choose content and reflect it via style inputs. What makes the final image art is not purely the content or the style, but their combination, guided by the creative direction of an artist. While digitized water lilies won’t grace the walls of the Musée de l'Orangerie anytime soon, Monet and artists behind NST works share a common goal: Plato’s mimesis, stretching and dramatizing reality in a way that clouds reason and stirs emotion. AI and paint are means to an end.

Because AI is nothing without style images, some artists claim AI-generated images aren’t art; they’re reproductions of the style images or even plagiarism of the artists behind the style images. The core issue, critics argue, is in the origin of artistic style. While artists seem to have innately unique styles, AI does not.

Diverging from Plato, Aristotle believes that art is mimetic of the character of nature, not its product. All human actions are mimetic or imitative, and imitation is how humans learn. In school, we follow along with math problems. At a Pilates class, we echo the instructor. And at the Louvre, Degas copies paintings.

Skirts flutter, slightly lifting the pages of the sketchbook. Faint hushes of a pencil making eager lines is the only sound to be heard. The man on the stool is an island; streams of passerby part around him, sneaking glances at the charcoal figures taking their first, shaky breaths. A 20-year-old Edgar Degas is planted at the Louvre, sketching clones.

Every artist has sources of inspiration. Before Degas was famous for his ballerinas, he honed his skills by copying paintings at the Louvre and Old Masters in Italy. Like inputs in an AI model, Degas took inspiration from Japanese prints to 16th-century Italian Mannerists, trying on styles for size. He didn’t wake up one day as an artist with an innately unique style; he copied others to learn how to create.

Similarly, AI is hollow, not capable of spontaneous creation. Through machine learning, it mixes up, learns from, and copies human artists, but artists using AI eventually produce original pieces. To become an artist, one must first be an imitator.

To some people, AI art conjures up soulless robots, stripping art of its humanity. Van Gogh was ahead of his time, saying, “Painted portraits have a life of their own that comes from deep in the soul of the painter and where the machine can't go.” Art by machine sounds like an oxymoron; without emotion, there is no art. The theory of Expressionism claims that art is a product of emotion and that creating art helps the artist to understand their emotions. And no one exemplifies this better than the quintessential tortured artist.

The sun has long set. Van Gogh paces around his studio at the asylum at Saint-Rémy-de-Provence, alone. His mind is an unrelenting tempest, and splatters of blue coat his arm erratically. He works on the canvas until he’s staring back at himself.

Van Gogh returned to self portraits time and time again in his artistic career. Call it soul-searching or being too cheap to hire a model, his work is an intimate look into his mental state. One of his most famous works, “Self Portrait, 1889,” captures a swirling background of blue, a pattern physicians liken to the work of mental patients. His paintings were a probe of his own making.

Expressionism makes sense of how his ostensibly simple arrangements of color were so deeply intimate. Art is an individual expression of emotions, and Van Gogh’s portraits come from deep within his soul. But what Van Gogh failed to consider in his condemnation of the machine is the human being behind every algorithm. Through curating dataset images and writing prompts, human creativity is at work. When painted portraits have a life of their own, it is not the paint that makes the portrait art. The artist, their emotions, and how they express them do. Who’s to say they can’t express them through AI?

When photography was first created, artists viewed it as cheapening art. Because photography involves mechanical features, traditionalists didn’t view it as an art form. Two centuries later, we accept photography and painting as distinct, but equally esteemed, mediums to create art. History repeats itself. Critics of “Portrait of Edmond Belamy,” GANS, and NST fume about the mechanical nature of creating art via code, but humans have always breathed art into the developing digital landscape.

As a marketing ploy, the bottom right corner of “Portrait of Edmond Belamy” was signed with an algorithm. But it was just that — a ploy. The algorithm did not replace the artist. Behind the piece, there was a team of artists, emoting and creating and expressing.

Like paint, AI is a tool. With the creative touch of a human being, art is born. ■

By: Ellie Stephan

Photographer: Leah Blom

Models: Emily Gift, Shareefa Gyami & Aaron Boehmer

Stylists: Fernanda Lopez & Saturn Eclair

HMUAs: Claire Philpot & Jessie Delfino

View the full spread as it appeared in Issue No. 19 here.