Before Harold Cohen (1928–2016) was known as a pioneer of computer-generated art, he was simply an internationally recognized painter. Gazelli Art House’s New York location is showing exactly that period: “Painting About Representation (1962–71)”, running November 12 through January 9, 2027 — the gallery’s first solo exhibition of Cohen at that space, and the first time this particular body of work has been the whole point rather than a prelude.
Seventeen paintings, one decade before AARON
The show gathers seventeen paintings made before Cohen began experimenting with programming in the late 1960s. Per the gallery, the exhibition “reveals how his enduring fascination with representation laid the intellectual foundations for AARON, one of the earliest and most influential artificial intelligence art-making systems.” The framing is deliberate: rather than treating the paintings as a historical footnote to the computer work that made Cohen famous, the show argues they’re where the actual ideas came from.
A continuation, not a break
Cohen moved to San Diego in 1968 and began learning FORTRAN. In 1973 he was invited to Stanford’s Artificial Intelligence Laboratory for two years, where he kept developing the ideas that would become AARON — a program he continued refining for more than four decades. The gallery’s own framing is explicit that this wasn’t a rejection of painting: his “transition from canvas to computer was neither an abrupt departure nor a rejection of painting, but the continuation of a question that had occupied him for years.”
What made AARON different from today’s generative AI
AARON’s method is worth sitting with precisely because it’s the opposite of how current AI image tools work. The gallery describes AARON as operating “through an expanding body of rules and representational knowledge encoded by Cohen himself” — explicitly contrasting that with “contemporary generative AI systems trained on existing images.” AARON didn’t learn to draw from a dataset of other people’s drawings; it drew from rules about representation that Cohen wrote by hand and refined over decades, closer to a formalized theory of image-making than a statistical model of one.
Why this is worth a trip back before the AI
It’s easy to treat “AI art” as something that started with diffusion models a few years ago. Cohen’s paintings — made years before he ever touched a computer — are a reminder that the questions behind machine-made images (what does it mean to represent something, and can that process be formalized into rules another system can follow) predate the current generation of tools by half a century, and that the artist who built one of the earliest answers to them started as, simply, a painter trying to figure out representation.