On the Expressive Limitations of Prompt-Directed Image Generation

Abstract

Though images produced with generative AI today are often met with negative reception, Kathryn Wojtkiewicz has convincingly argued that these images may still be considered works of art. To do so, she relies on Claire Anscomb’s notion of creative agency pertaining to automatic processes and leverages an analogy between generative AI and photography. Wojtkiewicz’s additionally argues that the causes for distaste towards generative AI are social, and not aesthetic, ones. Her analogy between photography and generative AI relies on the fact that the artist can predict the outcomes of the process to realize creative agency. However, in her analogy, she hand-waves the differences between these processes, differences which affect how well a generative AI user can predict the resulting features of the image. In particular, photography begins with something seen before the camera, whereas generative AI begins with a linguistic prompt, introducing ambiguity about the user’s intentions. Furthermore, the user iteratively adjusts the prompts and selects the image’s features from what was generated, and thereby expressing fewer of the user’s beleifs, attitudes, and emotions than other mediums. Consequently, we find that these limit the potential for representational expression in a work created with generative AI as compared to other artistic processes.