This work consisted of trying to use a language generation model (GPT-2, which was brand new at the time in 2019) to make sketches. The process is heavily inspired by Sol LeWitt's - I use GPT-2 to generate a set of rules describing a drawing; then based on to my interpretation of those rules; I make a processing sketch. For his "Wall Drawings" (concieved in 1968), Sol LeWitt would make cards of rules, which would be executed and drawn on the wall by someone else; the rules were written fairly straightforwardly; but still the executor would have to make decisions on their own regarding placement/etc of lines/shapes. Below are Sol LeWitt's instructions for a wall drawing at the Boston Museum:
"On a wall surface, any continuous stretch of wall, using a hard pencil, place fifty points at random. The points should be evenly distributed over the area of the wall. All of the points should be connected by straight lines."
This focus on the inevitable "misinterpretation" of rules perfectly lined up with what I was thinking about at the time in regards to traditional generative art (which has programmatically defined rules) and GAN-based art (where the rules are 'fuzzy' and learned by the system from seeing examples in the training set). This project was a nice way to explore misinterpretation in both traditional generative and AI based processes.
The bottom right image is an installation view from Gazelli Art House's 2024 show, "Awaken, Metamagical Hands". The GPT-2 work is shown on a screen (on the right side of the photo, different work; L-systems, are on the left), with a hidden raspberry pi running the generative sketch continually, producing a new "sketch" every ~7 seconds.
All images may be clicked to enlarge.
