When most people think of AI-generated images, they picture the sleek, hyper-realistic masterpieces or surreal digital paintings conjured up in seconds by modern text-to-image engines like Midjourney, DALL-E, or Stable Diffusion.
However, the history of artificial intelligence creating visual art stretches back decades further than you might expect. Long before neural networks, diffusion models, or prompts written in plain English, a British painter and programmer named Harold Cohen set out to answer a radical question: Can a machine be taught to draw?
The result of his quest was AARON—the computer program responsible for creating the world’s first true AI-generated art.
Meet AARON: The Pioneer of Artificial Creativity
In the late 1960s, Harold Cohen was a successful, internationally acclaimed artist representing Great Britain at prestigious global showcases like the Venice Biennale. Yet, fascinated by the burgeoning intersection of computing and cognition, he walked away from traditional painting to move to the United States. In 1968, he began working at the University of California, San Diego, with zero prior programming experience.
His goal wasn’t just to use a computer as a passive tool or a brush, but to encode human artistic behavior into software. By the early 1970s—specifically taking shape through 1973 during a residency at the Stanford Artificial Intelligence Laboratory—Cohen had developed AARON.
How AARON Made the First AI Images
Unlike modern generative AI, which relies on machine learning algorithms trained by scraping billions of internet images, AARON was built entirely on symbolic, rule-based logic (popularly known as Good Old-Fashioned AI, or GOFAI).
Because computers at the time could not simply “look” at data or handle pixel-heavy graphics the way they do now, Cohen manually programmed rules of drawing into AARON:
- Structural Knowledge: The program understood basic geometry, human anatomy, and how objects occupy space.
- Autonomous Decision-Making: Cohen did not dictate every stroke. Instead, AARON made its own independent choices regarding composition, overlapping objects, and spatial relationships.
- Physical Execution: Early iterations of AARON didn’t output files to a screen. Instead, the program was connected to custom-built, turtle-like robotic drawing machines that held physical pens and physically plotted the lines onto large sheets of paper on the floor.
The black-and-white line drawings produced by AARON in the early 1970s marked a monumental milestone: artwork generated autonomously by a machine running a cognitive set of rules.
From Lines to Color, and the Evolution of AARON
As computer hardware evolved over the decades, so did AARON. Cohen continuously updated the program, shifting it from simple, abstract line drawings into complex organic shapes, human figures, and vibrant botanical scenes.
By the 1980s and 1990s, Cohen had even taught AARON how to handle color selection and paint mixing. The software would choose its own pigments, and a specialized painting machine would mix and apply them to canvas. AARON’s creations were so sophisticated that they were eventually exhibited in prestigious fine-art institutions worldwide, including the Tate Modern in London and the Whitney Museum of American Art in New York.
Cohen maintained and evolved AARON for over four decades until his passing in 2016, cementing his legacy as the godfather of AI-generated art.
The Evolutionary Leap: From Rules to Neural Networks
Comparing AARON’s 1973 geometric line drawings to today’s text-to-image generators highlights how radically the field has transformed:
| Era | Primary Technology | How Images Were Created |
| :— | :— | :— |
| 1970s–1990s (AARON) | Symbolic Rule-Based AI | Programmed human rules; code dictated composition, geometry, and robotic drawing execution. |
| 2010s (GANs) | Generative Adversarial Networks | Two neural networks (a generator and a discriminator) competed to create realistic synthetic images (
Leave a Reply