I’ve been a fan of Sol LeWitt’s work since not long after I first read this tremendous piece by Dan Cohen. In fact, I’ve made three different visits to the exhibit at MASS MoCA that Cohen discusses, and every time, I’ve been awed. If at all possible, I strongly recommend seeing it yourself in person. You have but a decade left to do so.

Anyhow, in the aforementioned piece, Cohen aptly explain’s LeWitt’s work thusly:

Sol LeWitt is an unusual artist in that he rarely painted, drew, or sculpted the art you see by him. Instead, he wrote out instructions for artwork, and then left it to “constructors”—often art students, museum curators, or others, to do the actual work of fabrication. LeWitt liked to be a recipe writer, not a chef.

Cohen’s example is a piece called “Wall Drawing 1180”, whose recipe is pictured here:

[Photo courtesy of P. Kafasis]

From those simple words, something incredible emerges. The result is star field like something captured by the Webb Telescope:

[Photo courtesy of P. Kafasis]

LeWitt’s recipes leave room for randomness, variations based on size and location, and perhaps some amount of interpretation. With that in mind, artist Amy Goodchild has been playing with using artificial intelligences as constructors for LeWitt’s works. She recently published a fascinating look at what GPT-4 can produce, and how it compares against GPT-3. Here’s “Wall Drawing 1180” as created by AI under her guidance:

Her article is a great demonstration of the rapid progression of these AIs, but it’s also a fun artistic journey to contemplate. Some of the most interesting results came about when Goodchild collaborated with GPT-4 to guide and adapt things.