The Birth of Machine Imagination: Inside the Creation of the First AI-Generated Image in 1973

Long before text-to-image prompts, neural networks, and multi-billion-parameter models dominated headlines, a quiet revolution took place in a university computer lab.

In 1973, when most computers filled entire rooms and were primarily used to crunch numbers or solve chess puzzles, a British-born painter named Harold Cohen did something unprecedented. He taught a machine how to draw on its own.

The result was AARON—the world’s first artificial intelligence art system—and a series of abstract, meticulously composed pen-and-paper outputs that marked the true genesis of AI-generated imagery.

From the Canvas to the Code

To understand why the images produced in 1973 were so revolutionary, you have to look at who created them. Harold Cohen wasn’t a computer scientist; he was an established, elite artist who had represented Great Britain at the prestigious Venice Biennale.

By the late 1960s, Cohen had grown deeply curious about the cognitive mechanics of art. What is it that an artist actually does when they put a mark on a blank page? How do we understand boundaries, perspective, and composition?

Convinced that these weren’t mystical, unquantifiable human secrets, Cohen moved to the United States—spending time at the Stanford Artificial Intelligence Laboratory and later becoming a professor at the University of California, San Diego—to see if he could translate the act of drawing into code.

How 1973 AI Worked: Rules, Not Statistics

If you type a prompt into a modern image generator today, the system relies on deep learning, scanning billions of scraped internet photographs to guess statistical patterns.

AARON’s approach in 1973 was entirely different:

• Symbolic AI: Cohen did not feed AARON a dataset of existing images. Instead, he used symbolic artificial intelligence, writing explicit, rule-based logic into the program.
• The Grammar of Art: He taught the machine abstract rules of geometry, how lines intersect, how shapes close, and how to balance proportions within a canvas.
• Autonomous Decision-Making: Once Cohen hit “run,” he stepped away. The computer made independent aesthetic choices in real time—deciding what shapes to draw and where to place them without human intervention.

To bring the digital code into the physical world, AARON was connected to a mechanical drawing machine (a flatbed plotter) holding a pen or brush. As the computer calculated its next move, the mechanical arm physically executed the strokes on paper.

The resulting 1973 outputs were intricate, abstract compositions of lines and closed forms. They didn’t look like hyper-realistic modern photos, but they possessed a vital characteristic: they were conceived and executed entirely by a machine acting on its own programmed logic.

A Forty-Year Partnership

Most computer science experiments from the 1970s were temporary proof-of-concepts that were quickly dismantled. But Cohen’s work with AARON became a lifelong collaboration.

Over the next four decades, Cohen continuously updated the program. As computer processing power grew, so did AARON’s capabilities. The early abstract lines eventually evolved into recognizable figures, rocks, plants, and jungle-like foliage. AARON’s distinct evolution eventually earned it exhibitions at world-renowned venues, including the Tate Modern and the Whitney Museum of American Art.

Why the First AI Images Still Matter

Looking back at the modest black-and-white plots of 1973, it is easy to view them as primitive ancestors to today’s digital marvels. Yet, they carry immense historical and philosophical weight.