My early years at high school were marked by a protracted disagreement with my art teacher. Given the evidence of several years of atrocious drawings, I had concluded that I was no artist. Mr Swan, however, believed that I could be, but was being held back by my belief that I could not. The, at best, C-grade homework I produced did nothing to dissuade him, though it did anchor my conviction ever more firmly. No tears were shed by either of us when I dropped the subject at the earliest possible opportunity.
Now that I no longer receive a weekly reminder of my artistic inadequacies, my lack of talent barely intrudes on my consciousness. I have accepted that I am no artist. I can appreciate the skill involved without wishing to possess it myself—art is now something other people do.
And yet, when I heard about Dall∙E 2, I was intrigued. It’s a machine-learning model, designed by Open AI to produce images based on prompts entered by users in natural language. The claim is that the more detailed the description, the closer Dall∙E can get to the user’s vision. Here might be a way of using my long dormant creative faculties, despite my clumsiness with my hands. Perhaps technology could prove that my teacher had been right all along.
I have recently developed a fondness for the mid-century American artist Thomas Hart Benton, now best known for having taught Jackson Pollock. Benton’s work consists largely of murals located in mid-western cities. I asked Dall∙E to create a view of the City of London in his style. A few seconds later, it offered me four different views, of which this was the best:
This is certainly a view of a city, but not the City of London—unless the Gherkin has somehow acquired a twin. And, although it has some similarities with Benson’s work—such as its colour palette—and the lower part of the main building does resemble some of the images he created, the quality of the draughtsmanship is rougher and there is no trace of the interlaid images that feature in creations of his, such as America Today.
Benton is a bit obscure, though, so I decided to try a more mainstream artist. My own struggles at school had been based on my complete inability to render a physical object on paper. One of the artists who was most skilled at that task was the eighteenth-century English painter George Stubbs, who is renowned for his highly realistic paintings of horses such as Whistlejacket.
There was little point in asking Dall∙E for another equine portrait by Stubbs—there are so many of those in existence that there would be no way to ensure that the system did not simply copy an existing work. But, while Stubbs often painted dogs, he usually chose breeds like spaniels, whose fur obscures their musculature. A greyhound would allow him to display his mastery of anatomical detail, but in the only painting I could find that features one, the dog is a secondary figure and the focus is on one of his beloved horses. I entered “Greyhound by George Stubbs” into the system and let it work its magic.
That seems much better. It’s the sort of image I could imagine Stubbs producing, with his characteristic attention to the way muscle is visible beneath an animal’s coat. After all those fruitless art classes, I had finally produced an image that actually looked like the object it was meant to represent.
I made a few more iterations—a computer by Salvador Dalí, McDonalds as seen by Johannes Vermeer—but I realised that I was not creating anything original; I was simply producing pastiches. All the images were in the style of someone else. I was asking the system to “imagine” how a previous artist might have depicted a scene or object.
In an effort to be less derivative, I decided to create my own vanitas, a style of still-life painting popular in the fifteenth and sixteenth century Low Countries, which features symbols of death and decay that serve as visual memento mori. Since symbolism is important in the genre, I asked that my vanitas include a book (so much to learn), a compass (so many places to go), some grapes (earthly pleasures) and, of course, a skull. After a few seconds, Dall∙E produced this:
This seemed like a step forward. Together, Dall∙E and I had produced a recognisable vanitas. We had worked within the conventions of an existing genre to create a unique image, rather than a copy of someone else’s work. I had come up with the composition, and the system had generated the picture.
Perhaps for the first time I had produced something that appeared to be both original and a reasonable reproduction of the objects portrayed—but while I was no longer creating pastiches, I was still working within the boundaries of an existing genre. Would it be possible to create a new style?
In later cubist paintings, such as Jean Metzinger’s Soldier at a Game of Chess, artists created large images out of smaller straight-edged geometrical figures. Would it be possible, I wondered, to do something similar using circles—a kind of circle-ism?
My first attempt at St Paul’s Cathedral Made out of Circles in Yellow was a disappointment. It looked more like an arrangement of buttons:
Instead of creating a new style, I’d produced a child’s craft project. Attempts to refine the criteria made little difference. St Paul’s Cathedral Made out of Yellow Loops looked like a macrame creation placed on top of a photo of Sir Christopher Wren’s masterpiece, while St Paul’s Cathedral Made out of Interlocking Yellow Circles was a mosaic. It was like being back in my high school art class: I could visualise what I wanted to create, but the image I produced looked nothing like it. Perhaps this was because of an inherent limitation in the system—it can only work with existing styles—or perhaps I’m limited by my inability to describe what I want to produce sufficiently clearly, but, either way, concept and reality just did not match.
Broadly speaking, the process of creation involves two stages: conception and execution. In all the works I had produced, I took care of the first, and Dall∙E the second. But the experiment with circle-ism demonstrated that, in some cases, even the power of the algorithm was not enough to allow me to produce something in keeping with my vision. Can an artist who cannot produce the art he wants really call himself an artist?
And, even in those cases in which I had produced something that matched my idea and seemed—to my eyes at least—reasonably satisfactory, was I actually doing anything artistic? I was issuing instructions and accepting or rejecting the results, but I was not doing anything to create the work: I was not choosing where to place the pixels that formed the image. My role was like that of a chef at the pass: I was calling out orders and checking the dishes before they were sent to the dining room, but I was not cooking the ingredients.
In one sense, this doesn’t matter. Like head chefs, artists have long taken credit for works they have not physically created. In the Renaissance, masters would have apprentices who would do much or all of the actual painting, having been trained in their employer’s style. He might only paint faces—or he might have nothing to do with the painting at all beyond checking it before it was dispatched to the client.
At another level, however, it matters deeply. Those artists were still able to produce the work painted in their studios—whether or not they actually did so in any particular case—just as, if something goes wrong in Gordon Ramsay’s kitchen, he can man the stove. When Dall∙E failed to produce what I wanted, I was powerless.
Furthermore, auction houses and galleries are careful to distinguish between works by an artist’s own hand and those by his studio because the former are far more valuable. For collectors, the question of who wields the paintbrush is of vital importance, and an artist is only given credit for a work if he was solely responsible for both conception and execution.
The art market, then, would deny me authorship of the works I had created through Dall∙E. In the unlikely event of them ever reaching the auction block, they would need to be marketed as made by my “studio” rather than by me. If I had produced no art, I could not, in all conscience, describe myself as an artist.
My role in the process was not negligible, however. I had caused a piece of art that otherwise would not exist to be created. I had used the system to produce something I wanted, which I could not make myself. Rather than making me an artist, then, Dall∙E had allowed me to become a patron.
And it is for the aspiring patron or owner of art that systems like Dall∙E could revolutionise things. The art market does not function like other retail sectors. There are relatively few galleries, and they are often small and carry limited stock. Whereas one has a reasonable chance of finding a shirt in a particular colour, size and style, the chance of finding the exact image one wants are far lower. As a result, art buyers choose from among the selection available rather than looking for exactly what they want. By expanding the range of available options to all the images a user can describe, Dall-E substantially increases one’s chances of owning the exact picture one desires and thereby reduces the role of serendipity in how we buy art.
Of course, artists have taken commissions from time immemorial. Like Dall∙E’s creations for me, some have been more successful than others. Whistlejacket has a blank background because Stubbs downed his paintbrushes halfway through, following a financial dispute with his client. However, commissions are expensive and require the customer to find an artist who can paint in the style he wants. With Dall∙E, all one needs is access to a decent printer. It may not democratise the process of creating art, but it will democratise patronage. In the future, we will be able to decorate our homes with the exact images we want, rather than the best images we can find.
Like the replicators in Star Trek—machines that can convert energy into any food the user desires—Dall∙E can create a painting to suit your exact tastes. It may not have made me an artist, but it has allowed me to bypass the need for artists altogether by providing me with whatever art I like.