Long before typing a prompt into a text-to-image generator became a daily habit, the quest to make machines create art was already underway. When people talk about the “first ever AI image,” they aren’t usually thinking of modern neural networks or server farms. Instead, the true genesis of AI-generated art dates back over half a century to a groundbreaking software named AARON.
Meet Harold Cohen and AARON (1970s)
In the late 1960s, British painter Harold Cohen was a successful, internationally exhibited artist representing Great Britain at prestigious global exhibitions. Yet, he became fascinated by a radical question: What makes an image an image? More specifically, could a machine be programmed not just to execute a command, but to make independent, creative decisions?
Leaving behind traditional painting for a time, Cohen began learning computer science at the University of California, San Diego. By 1973, he had developed AARON—widely recognized as the world’s first artificial intelligence program dedicated to fine art.
How AARON Created Images (Without Machine Learning)
Unlike today’s AI models, which are trained on billions of scraped photographs using deep learning, AARON didn’t look at pictures or use neural networks.
- Rule-Based Intelligence: Cohen manually translated his own understanding of human cognition, geometry, and drawing into code.
- Autonomous Decision-Making: AARON was given fundamental rules about physics, object occlusion, and composition. The computer decided where to place lines, how to construct shapes, and how to balance the composition on its own.
- Physical Output: In its early iterations, AARON drove custom robotic drawing machines holding pens or brushes to physically ink the creations onto paper or gallery walls.
As the decades progressed, AARON evolved from drawing simple black-and-white abstract lines into rendering complex botanical forms, figures, and rich jungle scenes. Cohen and AARON maintained a collaborative partnership for over 40 years until Cohen’s passing in 2016.
The Evolution of Machine-Made Imagery
AARON paved the way for a multi-era evolution in digital creativity:
| Era | Technology | Key Milestone |
| :— | :— | :— |
| 1960s–1970s | Algorithmic & Rule-Based Code | Pioneers like Georg Nees, Frieder Nake, and Harold Cohen (AARON) prove computers can generate aesthetic compositions. |
| 2014–2018 | Generative Adversarial Networks (GANs) | Neural networks pit two systems against each other (a generator and a discriminator) to produce eerie, dreamlike portraits and landscapes. |
| 2021–Present | Diffusion Models & Text-to-Image | Massive multimodal models (like DALL-E, Midjourney, and Stable Diffusion) instantly translate human text prompts into photorealistic art. |
A Legacy That Endures
The story of the first AI image reminds us that artificial intelligence didn’t emerge overnight out of a vacuum. It represents a decades-long human obsession with bridging logic and imagination. While AARON was built with explicit rules rather than statistical weights, it successfully crossed the threshold from a passive tool into an autonomous creative agent, proving that the dialogue between humans and machines began long before the modern AI boom.
Leave a Reply