Long before the era of deep learning, text-to-image prompts, and neural networks spitting out photorealistic renderings in seconds, a very different kind of artificial intelligence was quietly drawing on sheets of paper.
To find the world’s first AI-generated image, we have to look back more than half a century—not to a server farm running a diffusion model, but to a pioneering software program called AARON.
Meet AARON: The Pioneer of Autonomous Art
In the late 1960s, British artist and programmer Harold Cohen embarked on a radical experiment at the University of California, San Diego. Cohen wasn’t interested in simply using a computer as a passive tool or a digital canvas; he wanted to know if a machine could actually be an artist. He wanted to encode the cognitive and physical act of drawing into code.
The result of his endeavor was AARON, a program that began producing its earliest recognizable artwork around 1973.
Unlike modern AI systems that scrape billions of images from the internet to learn statistical patterns, AARON operated on symbolic, rule-based AI (often associated with the “Good Old-Fashioned AI” or GOFAI era). Cohen painstakingly wrote rules about the physical world into the program’s logic:
- How objects overlap in space.
- How proportions balance.
- How light creates shadows.
- How abstract figures, foliage, and human forms are structured.
The program didn’t mimic a human photograph; instead, it made autonomous decisions. Given a set of boundaries, AARON would decide on its own composition, figuring out what to draw next without human intervention.
How the First Images Were Born
Because computer screens in the early 1970s were crude and printers were primitive, AARON’s creations didn’t exist as digital files on a desktop.
Instead, the program was connected to a custom-built, robotic drawing machine—a flatbed plotter. AARON would execute its code, sending coordinates to the machine, which gripped a physical pen and dragged it across a sheet of paper to ink out abstract shapes, minimalist human figures, and botanical forms. Later versions of the software even added color, painting the shapes it had autonomously conceptualized.
[ AARON Software Logic ] ---> [ Robotic Plotter / Pen ] ---> [ Physical Paper Artwork ]
By the mid-1970s and 1980s, AARON’s output was being showcased in prestigious institutions like the Tate Modern and the Whitney Museum of American Art, marking a historical milestone: the first time an autonomous system was widely recognized as a genuine artistic creator.
From Code-Driven Pens to Neural Networks
The evolution from AARON’s calculated lines to the hyper-realistic pixels of today’s technology unfolded in distinct eras:
- The Algorithmic Pioneers (1960s–1970s): Alongside Harold Cohen, artists and mathematicians like Frieder Nake and Georg Nees used mainframe computers to plot geometric, algorithmic patterns.
- The Deep Learning Revolution (2010s): The paradigm shifted entirely with the rise of deep neural networks. Instead of manually coding artistic rules like Cohen did, researchers trained algorithms on vast datasets, allowing networks to learn patterns implicitly.
- Generative Adversarial Networks (GANs) & Diffusion Models (Present): Systems like Midjourney, DALL-E, and Stable Diffusion revolutionized the space, transforming simple text prompts into complex, high-definition art in mere seconds.
A Legacy That Endures
While today’s AI models look nothing like a 1970s robotic pen plotter, the philosophical core remains identical. Harold Cohen spent decades refining AARON, proving that a machine could possess a form of mechanical creativity. When we look at the dazzling AI art of today, we are standing on the shoulders of a program that first picked up a mechanical pen fifty years ago.
Leave a Reply