I find myself constantly talking to ChatGPT, prompting or tweaking codes; I read and then tweak and iterate. This feedback loop feels less like using a paintbrush and more like conversing with a system, an active engagement instead of observation. Like the programming process of traditional generative art, interaction with AI can also be serendipitous. A generative AI output is by no means indeterminate; modern AI functions give the appearance that the predicted response is chosen by random (though true randomness could never be executed by a computer). However, many unexpected results are from human interaction, such as the way we prompt such systems.
This creative process has an iterative quality reminiscent of how the diffusion model itself works (iterative refinement of images through noise and feedback). There is always an illusion when working with generative AI; one types in a description, and an image/text appears that often matches or even exceeds our expectations. It can feel as if the output was somehow already there inside the computer, just waiting to be retrieved. Phrases like "the AI dreamed up an image" anthropomorphize the process, suggesting a kind of imaginative interiority to the machine. Yet, the more I look inside the system, the more I realize how much of a lack of imagination it has, and the more I appreciate human qualities that could not be achieved in such manners.