Long before modern neural networks could summon hyper-realistic portraits from a simple text prompt, a British-born artist named Harold Cohen sat in a room with a mainframe computer, trying to teach a machine how to see.
Today, we take for granted that typing a few words into a browser can conjure breathtaking digital paintings in seconds. But the lineage of artificial intelligence art didn’t begin in a Silicon Valley tech incubator with billions of parameters and vast libraries of scraped data. It began in the late 1960s with a piece of software named AARON—and a foundational question: Can a machine possess an internal understanding of what it means to draw?
The Birth of AARON
In 1968, Harold Cohen was an established painter representing Britain at prestigious global exhibitions like the Venice Biennale. Yet, he found himself gripped by an unorthodox curiosity about computers. Moving to the University of California, San Diego, he began learning programming, not as a tool to automate his chores, but as a medium to dissect the cognitive mechanics of human art.
By 1973, Cohen had developed the earliest iterations of AARON—a computer program designed to independently generate original artwork.
Unlike today’s AI models, which crunch billions of parameters to recognize patterns across massive datasets, AARON was built on symbolic, rule-based AI (often called “Good Old-Fashioned AI” or GOFAI). AARON didn’t look at millions of pictures of cats or landscapes to mimic them; it didn’t know what a rock or a human looked like photographically. Instead, Cohen painstakingly translated his own understanding of visual art into code.
He manually programmed rules regarding:
- Perspective and occlusion (how objects overlap one another in space).
- Proportions and structure (how a human figure or a plant is composed).
- Compositional balance (how to distribute weight across a canvas so the eye moves naturally).
Pen on Paper: The First Autonomous Strokes
The output of early AARON wasn’t a glowing screen filled with pixels. In the 1970s, the program fed its computed coordinates to custom-built, turtle-like mechanical drawing machines connected to the computer. These robotic arms gripped physical pens and ink, meticulously tracing lines onto large sheets of paper spread out on the floor.
When AARON plotted and executed its first autonomous abstract line drawings and primitive figures, it crossed a historic threshold. For the very first time, an artificial system was making autonomous compositional choices. Cohen didn’t pre-plan the final drawing; once he hit run, AARON decided where lines would intersect, how shapes would nest, and how forms would occupy the page based entirely on its internal rulebook.
As Cohen famously remarked years later, AARON was given a set of rules, but it exercised a form of cognitive autonomy within those boundaries. It was an independent actor rendering unique visual expressions.
From Lines to Living Ecosystems
Over the next four decades, Cohen continuously evolved AARON. What started as stark, black-and-white geometric and humanoid line drawings gradually grew more sophisticated.
- The 1980s and 90s: Cohen taught AARON how to handle color, adding complex painting routines that allowed the software to mix its own hues and fill in its shapes.
- The 2000s: AARON evolved to create lush, organic environments—featuring stylized human figures interwoven with rich botanical and jungle-like elements.
Institutions that once guarded the boundaries of traditional fine art took notice. AARON’s creations were exhibited globally, landing in legendary spaces like the Tate Modern and the Whitney Museum of American Art.
The Legacy of the First AI Artist
When Harold Cohen passed away in 2016, development on AARON ceased, leaving behind a monumental legacy.
The contrast between AARON and modern generative tools like Midjourney or Stable Diffusion is stark. Modern AI is statistical and data-hungry, predicting pixels based on vast internet-scale training pools. AARON was deliberate, philosophical, and algorithmic—a machine built from the inside out to understand the syntax of drawing rather than the mimicry of imagery.
Yet, every time we prompt an AI to paint a dreamscape today, we are walking down a path first cleared in 1973. AARON proved that art and code were not mutually exclusive, establishing that a machine could hold a pen, contemplate a blank page, and draw something entirely new.
Leave a Reply