Each of these works starts with a text prompt. The prompts are often an amalgamation of headlines (climate change, floods, immigration) and art history (color-field, historical painting). Then a conversation happens—a back and forth—the text is tweaked, images are selected, variations generated, sections refined, and then the piece is further digitally altered and scaled until the composition is set. This often takes hundreds (sometimes thousands) of back-and-forth discussions. The works are then executed as an inkjet with acrylic on canvas (each unique).
In this way, the works are a dialogue with an artificial intelligence (in most cases, DALL•E). Although “artificial intelligence” might be a bit strong; the program is really a neural network (technically, a large-language model, or LLM, and a denoising diffusion probabilistic model, or DDPM, paired together), or maybe think of it as a consensus engine. It (they?) “reads” the Internet, models that information, builds correlations, creates weightings and associations, and then adds a bit of randomness. In this sense, it builds a consensus—not a Platonic ideal but more of an Internet ideal—the sum-thinking of images available on the Internet. It thus produces an average (really a random average, if there is such a thing), and for many things, the output is just that—average.
But the corner cases are interesting—the odd combination of concepts, objects, and styles. There are times when things don’t add up. The program has no sense of space or what physical reality requires and doesn’t. It only knows weightings and correlations, not our rules. And because of this, you can coax out interesting scenes, situations, and even aesthetics. It is where things diverge that the images become interesting—twisted faces, illogical rooms, and the odd combination or insertion. There are different textures and tones, splatters and scrapes, zips and color fields.
The genesis is human (the prompt) while the initial composition is artificial (created by a program trained on images, presumably made by humans, from the Internet), and herein lies the dichotomy: the output at times is deeply human, expressive, interesting, curious, haunting, and maybe even true, despite the artificialness of how the images were created—a form of artificial expressionism, if you will. That’s the interesting thing for me, the tension—the not-quite-right and yet the humanness of it all.
Maybe now computers can ask us questions and not just give us answers.
Lives and works in Chicago
Graduate Student at Large, University of Chicago
BS & MS, Electrical and Computer Engineering, University of Illinois Urbana-Champaign