AI Art Cannabalism
Cal Berkeley Student Grace Zhou is an artist. When she signed up for her required college writing class under Benjamin Spanbock, she never imagined how AI (Artificial Intelligence) applications could infinitely generate and regenerate her art.
A Catalyst for Deep Inquiry
Grace Zhou’s Project AI Art Canabalism became for the class the “catalyst for deep inquiry, debate and genuine knowledge building concerning the impact of AI across a variety of disciplines and issues,” according to Spanbock.
The project goal was to create a loop in which AI can feed onto itself and infinitely generate new images and artwork from prompts she constructed. For her project, she used Microsoft Bing AI by GPT-4 and DALL-E-3.
Evolving Over 12 AI Generations
Zhou began with a painting she made and imported it into Bing AI. Her work was inspired by own interest in wildlife and Mark Tansey’s monochrome painting The Wake, which inspired her to use limited colors for the project. “The withered lotus flowers and dried pods depict the rapid loss of wetlands and their biodiversity due to climate change and pollution, while the heron in the marsh represents all the species impacted by this threat.”
Interpreting the Messaging in AI Images
In her first test, she asked Bing to write a description of the painting and compared it to one she wrote. Bing’s description was far more detailed, and the resulting image far more colorful and realistic and thus more positive than the intent of her image.
She tested these questions by again prompting Bing AI to feed on itself. Over 12 iterations, the painting changed significantly from the one Zhou created. The color palette, positions of the heron, and expressed mood differs between each image. One iteration even showed a futuristic city by a lake somewhere in space.
AI: Lacking Accuracy
Zhou observed that additions of elements like clouds, rocks, and the moon, etc, are all signs that “AI in its current form lacks accuracy and this game of Telestrations has helped amplify this concept.”
Zhou concluded that “While the generated images were all bound by similar prompts, each image possesses unique nominal authenticity.”
This project successfully proved for Zhou that AI image generators can create many, if not an infinite number of, unique art pieces, while the interpretation of any intended message within the image is stll left to the human being viewing the image.
Grace Zhou's project was part of the required work for her CWR4A beginning class in Reading and Composition.