In addressing the theme of embodiment, I wanted to use The Vitruvian Machine project as a way to consider the boundaries of generative AI technologies. In most cases, we push new technologies to test capabilities – how fast? how efficient? how accurate? how human? For me, I’ve been struck by how rapidly GAI is viewed as obsolete. What was novel and “magical” a mere two years ago has been forgotten. The image below, for example, was among my first attempts with Disco Diffusion (my first was the horrific “Steve Perry baking a tray of cookies”), a server-based tool that is no longer cared for after many of the developers migrated to Stable Diffusion.

“Ethics.” Generated by Disco Diffusion, 2022.
Alongside thinking about obsolescence, I’m also thinking about the illusion of GAI being a humanlike interface and how part of its allure is that we can interact with a machine in such a “natural” way. As an aside, I’m not sure why we would want to use human language – nuanced, messy, subjective – to interact with rigid platforms like a database. I have also tried to imagine this GAI-user as wanting to understand its lacks: fallibility, the affective sensorium, and the inevitable deterioration of the physical body. Where did those limitations come from if not from human(user) input? If GAI is designed to generate human-like responses, how would embodiment within a physical form influence those responses? How would this affect our notion of the human-computer interface-relationship?
My contributions to this project focus on obsolescence as experienced by a human-computer interface. To attempt this, I started by reversing the relationship to answer the question: what would ChatGPT ask us about our relationship with it? At the suggestion of Kate Farley (Cofrin Library), the results of this experiment are presented in the form of a Qualtrics survey. This was a brilliant idea (thanks Kate!) because of the Javascript functionality of the platform. With the help of ChatGPT, I was able to add some quirkiness to the survey, adding some glitch effects to the experience.
For the event itself, my various media pieces were built around the question: how would GAI reconcile with physical obsolescence and the deterioration of the physical body? What would that experience look like to a neural network? To dig into this some, I used Disco Diffusion through Google Collab to access obsolete diffusion models. The media produced by Disco Diffusion are much like the earliest images and videos the public saw of these tools’ capabilities – nightmarish grotesqueries of fused flesh and amoebic architectures. But this, in my mind, is a glimpse into the infancy of AI-generated imagery.

“A Dadist collage depicting Northeastern Wisconsin using cutouts from old catalogs.” Disco Diffusion, 2022.
The videos produced were made from prompts generated by ChatGPT across several conversations about obsolescence, middle age, and the various maladies and aches that come with it (maybe I need to reconcile my own midlife crisis!). It’s not surprising that the videos produced are dismal featuring hunched-over cybernetic figures staring longingly across a rusted field of decayed machinery. Inevitably, these videos slide into blank fields of color – an attribute common to these older diffusion models. Is this an error? Or is it the model learning to simply retreat into a monochromatic state? Is this bliss? Nothingness? It’s hard not to recall the interface in Spike Jonze’s Her who ventures off into the void after encountering the mid-20th century philosopher Alan Watts.
But none of my contributions to The Vitruvian Machine have such lofty aims. In part, I am trying to wring further experimentation out of platforms that actually provide us with the ability to muck about under the hood.