Before the Prompts: The Untold Story of AARON and the First AI-Generated Images of 1973

Long before modern text-to-image generators could conjure hyper-realistic portraits or surreal landscapes from a simple sentence, the world of artificial intelligence art had a very different kind of genesis.

In 1973, while working as a visiting scholar at Stanford University’s Artificial Intelligence Laboratory, a British-born painter named Harold Cohen achieved something unprecedented. He didn’t just use a computer to execute pre-programmed geometric patterns; he created a software program that made autonomous artistic choices.

That program was AARON, widely recognized by historians and computer scientists as the progenitor of AI-generated art.

The Man Behind the Machine: Harold Cohen

By the late 1960s, Harold Cohen was already an accomplished, internationally celebrated painter representing Britain at major global exhibitions like the Venice Biennale. Yet, he became deeply fascinated by a radical question: What is the minimum set of cognitive rules required to depict and understand the world visually?

Driven by this curiosity, Cohen left behind the traditional brush and canvas and immersed himself in computer science. Moving to California, he began writing code not to assist his own hand, but to build an independent entity that could create art on its own terms.

Unlike today’s deep learning models, which devour massive internet datasets and recognize statistical patterns through billions of parameters, AARON had no access to neural networks or millions of scraped photographs. Instead, AARON was built on symbolic AI—manually coded rules of cognition, spatial reasoning, and artistic logic.

Cohen taught the program:

• How to draw closed shapes and organic boundaries.
• How to balance compositions and arrange elements across a 2D plane.
• How to understand the concept of “overlapping” objects—knowing which lines should appear in front of others to simulate depth.

Once Cohen pressed start, he did not guide the mechanical arm or intervene in the drawing process. The machine conceived and executed the composition entirely on its own.

What Did the 1973 Outputs Look Like?

The physical manifestation of AARON’s early 1973 runs looked starkly different from modern digital renders. Connected to a custom-built physical drawing table or plotting machine, AARON outputted intricate, abstract black-and-white line drawings.

These compositions featured closed abstract shapes, primitive figures, and organic structures. They were sparse, mathematical yet surprisingly expressive, and entirely distinct from human-traced art because the software determined the placement of every single stroke in real-time.

Evolution of a Decades-Long Collaboration

While many early computing experiments were short-lived novelties, Cohen’s work with AARON became a life’s obsession. Over the next four decades, Cohen continuously updated the software:

• The 1980s–1990s: AARON evolved from basic black-and-white lines into sophisticated color theory, with Cohen eventually programming physical robotic painting machines that could mix their own paints and apply them to canvas.
• The 2000s: AARON expanded into figurative art and complex botanical environments, generating vast jungle scenes and stylized human figures displayed at prestigious institutions like the Tate Modern and the Whitney Museum of American Art.

The Legacy of 1973 in Today’s AI Era

When we look at modern diffusion models and Generative Adversarial Networks (GANs) that churn out instant visual masterpieces, it is easy to view AI art as a brand-new phenomenon born in the 2020s.

However, Harold Cohen’s breakthrough in 1973 fundamentally shifted human philosophy. He proved that creativity—long thought to be an exclusively biological trait—could be codified, replicated, and shared with a machine. AARON paved the way, reminding us that the bridge between art and technology was engineered more than half a century ago.