Before the Prompts: The Fascinating History of Early AI Art

We live in an era where generating a breathtaking, hyper-realistic image takes nothing more than a casual text prompt and a few seconds of waiting. But long before modern diffusion models and text-to-image engines revolutionized digital media, artists and computer scientists were already exploring the boundaries of machine creativity.

The story of early AI art is not one of instant gratification. Instead, it is a decades-long chronicle of logic gates, robotic arms, custom-built algorithms, and profound philosophical questions about what it means to create.

The Rule-Bound Genesis: Harold Cohen and AARON

The story of autonomous machine art truly begins in the late 1960s and early 1970s with Harold Cohen, a celebrated British painter who represented his country at the Venice Biennale. Troubled by the limits of his own studio practice, Cohen moved to the University of California, San Diego, and taught himself computer programming.

In 1973, he began developing AARON, a computer program designed not to copy existing art, but to generate original drawings autonomously.

• Symbolic Logic: Unlike today’s data-hungry neural networks, AARON operated on symbolic, rule-based AI. Cohen literally coded the rules of drawing—how lines intersect, how shapes enclose space, and how objects overlap.
• The Physical Output: AARON didn’t just live on a screen. Connected to custom-built robotic drawing machines (plotters), the program carefully guided physical pens across large sheets of paper to create intricate geometric and abstract compositions.
• Autonomous Evolution: Over four decades, Cohen continuously updated AARON, giving it the ability to render organic forms, plants, and human figures. Cohen famously remarked that he could never fully predict what AARON would draw, acting more as a collaborator or midwife to the machine’s creations than a traditional author.

The Neural Turning Point: DeepDream and GANs

As computing power grew and machine learning shifted from symbolic logic to deep neural networks, the visual style of computer-generated art underwent a radical transformation.

1. Google DeepDream (2015)

Originally designed to help researchers visualize how convolutional neural networks classify images, DeepDream turned the process inside out. By feeding an image into a network trained to recognize objects (like dogs or buildings) and instructing it to amplify those patterns, the algorithm produced hyper-psychedelic, hallucinatory landscapes swarming with algorithmic eyes and surreal organic shapes. DeepDream gave the world its first viral taste of what a machine “sees” when looking at art or nature.

2. Generative Adversarial Networks (GANs, 2014)

Invented by Ian Goodfellow, GANs introduced a brilliant framework: two neural networks pitted against each other.

• The Generator creates fake images from random noise.
• The Discriminator evaluates them against a dataset of real art or photographs, trying to spot the fakes.

Through this adversarial game, both networks improved exponentially. GANs bypassed the need for strict human-coded rules, allowing computers to learn the underlying statistical distribution of visual styles, paving the way for hyper-realistic portraits, novel textures, and entirely new digital aesthetics.

The Timeless Question of Authorship

Every major shift in art history—from the invention of photography to digital graphics—has triggered debates over authenticity and authorship. Early AI art forced philosophers and critics to confront these questions decades ahead of the current consumer boom.