In recent years, Artificial Intelligence (AI) has emerged as a useful—though controversial—tool for artists and illustrators alike. Using written prompts (such as “a cat painted in the style of Monet”), AI text-to-image generators are capable of creating images across a range of styles, from hyper-realistic to abstract.
In August 2022, an American video game designer won first prize at the Colorado State Fair Fine Arts Competition, for a work generated by AI. Predictably, his win went viral. “AI won an art contest, and artists are furious,” wrote CNN, on September 3. In an editorial published on September 4, a New York Magazine photo editor asked, “Will DALL·E the AI artist take my job?”
When I started using these systems, including Craiyon, DALL·E 2, and Midjourney, I avoided the tropes I’d seen online—the amusing stunt prompts shared on Twitter (“SpongeBob’s house after the apocalypse”; “Wizard of Oz passport photo”; “Lady Gaga in a Pingu episode”)—and the elaborate images of maidens and dystopias, like something ripped from the cover of an old pulp novel.
Instead, I wrote prompts designed to emulate botanical photographs by Anna Atkins (1799-1871) and William Henry Fox Talbot (1800-1877). I peppered my prompts with words like calotype (Talbot’s process), cyanotype (the medium Atkins made her own), photogram and lumen print (camera-less techniques that date back to the dawn of photography).
The resulting images challenge the photographic canon—which is now nearly 200 years old—while evoking the biological (and artistic) concepts of generation and reproduction.
DALL·E 2 can not only create a series of images, it can generate new variations of its own designs, and edit preexisting images into new forms. Which raises the question: if AI can create a variety of plant forms, could another AI system classify them?
After generating botanical imagery with DALL·E 2 and Craiyon, I uploaded the images to iNaturalist, a plant identification app which uses its own algorithm to identify wild species. (To be clear, I stopped at this step; I have not recorded these AI-generated images as real-world observations!)
Predictably, iNaturalist’s computer vision model didn’t know what to make of these fanciful images. “We’re not confident enough to make a recommendation,” it said, but offered a few suggestions for each. The app interpreted a wisp of foliage as a species of willow. A leafy stem: Culver’s root. A burst of petals: Gymnosiphon, a genus of flowering plants in the yam family.
Some of Still Life’s “found images” are displayed in the style of an old herbarium, in a collection playfully called, “Flora DALL·Eensis” (as in Flora Canadensis; a flora of Canada, for example). Labelled as real plants, this collection is accompanied by copies of matching herbarium specimens, downloaded from online biodiversity databases.
In this sense, Still Life becomes something like a process of divination. The project tests and questions the veracity of the images we encounter in real life and online, and represents a complex feedback loop between two systems of Artificial Intelligence; mediated by the artist.
Still Life is now on display at Modern Fuel Artist-run Centre in Kingston, Ontario, Canada.