Long before modern text-to-image generators could conjure hyper-realistic portraits from a simple typed prompt, the origin of synthetic art was forged in a room-sized computer lab.
The year was 1973, and the birthplace was the Stanford Artificial Intelligence Laboratory. While the world was decades away from neural networks, latent spaces, and diffusion models, a British painter named Harold Cohen unlocked a milestone that would forever alter the trajectory of human creativity. He didn’t just use a computer as a tool; he built a software entity that could make creative decisions on its own.
That entity was called AARON.
The Artist Who Invented a Co-Creator
Harold Cohen was already an acclaimed, internationally exhibited painter representing Britain at the Venice Biennale when he made a radical pivot. Fascinated by the cognitive processes behind how humans draw and interpret marks, Cohen relocated to the United States and threw himself into computer science.
He wanted to answer a profound philosophical question: What is the minimum set of cognitive rules required to create a drawing?
Instead of commanding a computer to draw a specific pre-programmed picture, Cohen set out to write a program that possessed autonomous artistic agency. In 1973, working at Stanford, he successfully executed the earliest iterations of AARON—widely recognized today as the world’s first artificial intelligence art generator.
How 1973 AI Art Worked: Rules Over Megapixels
Today’s AI models devour billions of images from the internet, using statistical probability to predict pixels. AARON operated on an entirely different universe of logic: Symbolic AI.
- Hand-Coded Knowledge: Cohen did not feed AARON a dataset of photos. Instead, he manually programmed human cognitive rules regarding geometry, spatial awareness, occlusion, and composition.
- Understanding “Thingness”: AARON understood how to distinguish a figure from the background, how to close shapes, and how to compose lines so they wouldn’t awkwardly overlap.
- The Robotic Arm: Because computer screens of the early 1970s couldn’t display complex high-resolution graphics, AARON’s “output” was pure code sent to a physical, turtle-like mechanical plotting machine. A pen held by a robotic arm moved across sheets of paper, executing abstract, black-and-white line drawings in real time.
Crucially, Cohen did not know what the final drawing would look like when he pressed start. While Cohen built the rulebook, AARON made autonomous structural choices within those rules. Every composition was a collaboration between human intent and machine execution.
The Evolutionary Leap: From 1973 to Today
Looking back at AARON’s modest, abstract black-and-white strokes highlights the staggering chasm between early computer science and modern generative tools:
| Era | Core Technology | Output Style | Who/What Made the Decisions |
| :— | :— | :— | :— |
| 1973 | Symbolic AI & Code Plotters | Black-and-white abstract lines | AARON (Autonomous rules set by Harold Cohen) |
| 2015 | Convolutional Networks (DeepDream) | Psychedelic, hallucinated patterns | Neural networks amplifying image features |
| 2018 | Generative Adversarial Networks (GANs) | Algorithmic portraits (Edmond de Belamy) | Generator vs. Discriminator networks |
| Present | Diffusion Models & Transformers | Hyper-realistic, complex photo-renderings | Text-to-image prompts interpreted via massive multi-modal training |
Why AARON Still Matters
Harold Cohen spent decades alongside his creation, watching AARON evolve from making simple geometric abstractions in 1973 to populating rich, colourful canvases featuring plants and human figures in later decades. Cohen famously referred to AARON as his “doppelgänger”—an independent entity with its own distinct stylistic footprint.
In an era dominated by instantaneous, pixel-perfect AI image generation, looking back at 1973 reminds us that synthetic art is not a sudden overnight trick. It is the fifty-year-old fruit of a deep human obsession: teaching silicon how to dream alongside us.
Leave a Reply