The True History of the First AI-Generated Image in 1973

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

Long before text-to-image prompts became a daily staple of digital culture, and decades before neural networks dominated headlines, a quiet revolution took place in a university laboratory.

In 1973, while most computers occupied entire climate-controlled rooms and were used strictly for crunching numbers or playing basic text games, a British painter named Harold Cohen achieved something extraordinary. He gave a computer the ability to create original visual art autonomously.

The result was AARON—widely recognized by art historians and technologists as the creator of the world’s first artificial intelligence-generated images.

The Mind Behind the Machine: Harold Cohen

To understand the significance of 1973, you have to look at Harold Cohen’s background. Cohen wasn’t a computer scientist dabbling in hobbies; he was an internationally acclaimed painter who had represented Great Britain at the prestigious Venice Biennale.

In the late 1960s, Cohen grew fascinated by a fundamental question: What is it that makes a drawing recognizable as a drawing? He realized that humans don’t just copy the world pixel-by-pixel; we rely on a set of cognitive rules, symbols, and structural understandings to translate a three-dimensional world onto a flat canvas.

Intrigued by whether these cognitive rules could be codified, Cohen moved to the United States and eventually landed at the Stanford Artificial Intelligence Laboratory (SAIL). There, he set out to build a piece of software that didn’t just automate a mechanical process, but actually thought about composition.

Anatomy of 1973 AI: Rules, Not Neural Networks

Today’s image generators rely on massive deep learning datasets, massive GPUs, and probabilistic diffusion models trained on billions of scraped photographs. AARON was the exact opposite.

• Symbolic AI: AARON was built using symbolic AI (hand-coded logic). It possessed no dataset of historic paintings to mimic, nor did it look at photographs.
• The Rules of Engagement: Cohen manually programmed rules of visual cognition into the software. He taught AARON how to construct closed shapes, how to understand the relationship between a “figure” and “ground” (background), and how to overlap lines without crashing them into one another.
• Autonomous Execution: Once Cohen hit “run,” he did not intervene. The computer program made independent decisions about composition, line weight, and spatial arrangement.

Connected to a primitive physical drawing machine (a flatbed plotter controlled by the computer), the system would methodically drag a pen across paper, drafting intricate, abstract, and organic line art.

“The machine conceived and executed the artwork on its own. Once Cohen pressed start, he did not intervene in the creation of the drawing.”

A Paradigm Shift in Art History

The abstract line compositions produced by AARON in 1973 stunned the art and tech communities alike. For the first time, an artifact had been brought into existence through a machine’s internal logic rather than direct human muscle memory.

Crucially, Cohen didn’t view AARON as a tool—like a brush or a camera—or as a replacement for himself. Instead, he viewed AARON as an entity, a collaborative partner whose creative capabilities evolved over the next four decades.

As the decades progressed, Cohen continuously updated AARON. The software eventually graduated from simple black-and-white abstract forms to painting complex flora, human figures, and vibrant color palettes. Artworks by AARON eventually found permanent homes in premier institutions, including the Tate Modern and the Whitney Museum of American Art.

The Legacy of 1973 in Today’s AI Era

Looking back from the vantage point of modern generative media, the 1973 outputs of AARON look quaint—resembling minimalist blueprints or intricate geometric sketches. Yet, they laid the conceptual bedrock for everything we experience today.