Picture a portrait of the most generic late-eighteenth century gentleman you can imagine. Picture him through a nearsighted lens, the kind an optometrist uses to test your vision, so blurry he looks like a Dementor victim mid-soul-sucking. Picture the hand of the artist bringing this mediocrity into being. Now, replace that hand with the robotic arm of a machine. What you have in mind probably looks a great deal like “Portrait of Edmond Belamy,” the artificial intelligence-generated painting that recently sold for $432,500 at Christie’s.
It is incredible that AI-generated art has reached a point where it can be sold for a dumbfounding amount of money by one of the most prominent art auction houses in the world — a feat many human artists never accomplish in their lifetime. Unfortunately, this harbinger of a new era — if that is indeed what it is — is, in and of itself, disappointing. Quite frankly, “Portrait of Edmond Belamy” is stale, unoriginal and boring. While painted as something of a breakthrough, Obvious’ success story at Christie’s does not actually represent a new technique, nor is it the first of its field.
The sale, made on the morning of Oct. 25 to an anonymous phone bidder, was 45 times the $7,000 to $10,000 estimate predicted by Christie’s. “Portrait of Edmond Belamy” was created using a generative adversarial network (GAN) by the French art collective comprised of three 25-year-old artists known as Obvious. In order to train GAN, Obvious fed the machine 15,000 portraits painted between the 14th and 20th centuries. The program’s “Generator” then began developing its own images that fused or carried the traits from all those inputted paintings, while the program’s “Discriminator” detected differences between the machine-generated images and human-made images. After a period of trial and error “training,” the Generator is able to produce images that the Discriminator thinks are real portraits. Once the GAN Generator has created an image that seems real enough for the Discriminator, you have a machine that’s now an artist. The scale to which GAN is acknowledging and fusing traits and styles from 15,000 paintings over seven centuries to create a singular image is incredible in comparison to popular technologies like Photoshop, which simply transform a single image into one key style such as Pop Art or Comic Book.
The disillusionment with “Portrait of Edmond Belamy” is not only that it uses a machine to generate a combination of pre-existing images or that it uses borrowed code from 19 year old artist and programmer Robbie Barat, but that it used this nontraditional artistic medium to make art that is so traditionally Western. As AI gains recognition in the art world, I want to see new types of art created — not the same, if duller, version of European impressionist portraits. I want to see the harmony and interplay of “man” and machine, not the domination of one over the other.
Beginning in the late 1960s, abstract painter Harold Cohen created a computer software program called AARON which was inspired by Native American petroglyphs, children drawings and artists he admired. AARON generated unique art that mimicked freehand drawing and, over time, learned how to distinguish foreground and background, draw complete shapes and recognize when a work of art was “complete.” Cohen worked with the machine — at first painting AARON’s creations with color and, then, continuously developing AARON to progress from abstract, biomorphic forms to people and plants.
Modern AI technology that creates innovative art does exist. The Art and Artificial Intelligence Lab at New Jersey’s Rutgers University has developed CAN to detect the creativity of iconic paintings and, therefore, its influential value. While GAN trains its program to reproduce new images that fuse similar characteristics within human-made paintings, CAN focuses on embracing the differences between paintings to create images with characteristics of what made a painting unique and ground-breaking. On the lab’s website is a quote from former chief officer of Walt Disney Animation John Lasseter: “The art challenges technology and the technology inspires art.”
Modern technologies should not just be used to automate things. Can we develop machines with unique artistic styles just as the forms of painting differ from artist to artist? Why aren’t we promoting and auctioning art that, like Cohen’s AARON, aims for novelty, that has something unique to say? Could we utilize AI to further redefine artistic canons, art by artists like Kehinde Wiley or Yayoi Kusama or Shepard Fairey? These artists rendered class, race and sexuality through mediums like painting, performance art and street art that is remolding what society recognizes as “good art.” I’m hoping that AI-generated art continues to move in this direction, rather than reverting back to traditional Westernized canons that have dominated the art scene for far too long. Authoritative art structures such as Christie’s should be conscious of the type of AI art they put on a platform, because that is what becomes the canon and shapes the direction in which future art develops.