When Silicon Learned to Draw: The True Story of the First AI-Generated Image (1973)

Long before modern graphics processing units, neural networks, or billions of parameters scraped from the internet, a revolution in digital art quietly hummed to life in an academic laboratory.

While most computer scientists in the early 1970s were focused on teaching machines basic logic puzzles or simple chess moves, a British-born painter dared to ask a radically different question: Can a machine learn to be an artist?

That painter was Harold Cohen, and the result of his experiment in 1973 was AARON—widely recognized as the world’s first true artificial intelligence art system and the genesis of machine-generated imagery.

The Artist Who Traded His Brushes for Code

By the late 1960s, Harold Cohen was already an established, celebrated figure in the art world. He had represented Great Britain at the prestigious Venice Biennale and was at the peak of his career as a traditional abstract painter. Yet, he grew intensely curious about the cognitive mechanics of art-making. What actually happens when a human makes a mark on a canvas? How do we understand space, line, and composition?

Seeking answers, Cohen accepted a fellowship at the Stanford Artificial Intelligence Laboratory (SAIL) in 1968, later moving to the University of California, San Diego. Computers at the time were monolithic machines that filled entire rooms, running on punch cards with a fraction of the processing power found in a modern digital wristwatch.

Instead of treating the computer as a mere tool to draw lines mechanically, Cohen wanted the machine to make autonomous creative decisions.

Inside AARON: Rules Instead of Neural Networks

Unlike today’s deep learning models that ingest massive visual datasets to spot statistical patterns, AARON was built entirely on symbolic AI and hand-coded logic rules.

Cohen did not train the program on pictures. Instead, he spent years translating his own deep understanding of human draughtsmanship into code. He programmed AARON with foundational principles of visual art:

• How to distinguish figure from ground.
• How to draw closed, non-overlapping shapes.
• How to balance a composition and manage spatial depth.

In 1973, Cohen formally presented AARON at the International Joint Conference on Artificial Intelligence (IJCAI). The system generated abstract, intricate line drawings that resolved into coherent compositions.

Crucially, once Cohen hit “start,” he did not intervene. The software calculated where the lines should go in real-time, translating its decisions into code that drove a physical, turtle-like robotic drawing machine (a mechanical plotter) holding a pen across a sheet of paper. The machine was thinking, choosing, and executing entirely on its own.

A Forty-Year Partnership

Most computer experiments from the 1970s were short-lived proof-of-concepts that faded into obscurity. But AARON was different. Cohen dedicated the rest of his life to a continuous, four-decade collaboration with his software creation.

As computing power evolved, so did AARON. Over the decades, Cohen expanded the program’s rulebook, allowing it to transition from abstract geometric shapes into rendering organic botanical forms, plants, and eventually human figures. AARON’s artwork eventually found its way onto the walls of elite institutions like the Tate Modern and the Whitney Museum of American Art.

Why the 1973 Milestone Still Matters Today

When looking at modern AI image generators that produce hyper-realistic art in seconds, it is easy to view them as a sudden overnight phenomenon driven entirely by modern tech companies.

However, looking back at AARON’s humble, black-and-white plots from 1973 reveals the true roots of synthetic creativity. Harold Cohen’s pioneering work forced the art world and computer science to confront a profound philosophical dilemma—one that we are still navigating today:

Where does human intent end, and machine execution begin?

AARON proved that art does not exclusively belong to flesh and bone; it can emerge from logic, code, and silicon. Fifty years later, every prompt we type into an AI generator traces its lineage right back to Cohen sitting in a lab in 1973, watching a robotic arm hesitantly put its very first line to paper.