The Fever Dreams Before the Storm: A Short History of Early AI Pictures

Today, we live in a world saturated by hyper-detailed, instantaneous artificial intelligence imagery. With a simple text prompt, anyone can conjure photorealistic landscapes, intricate anime portraits, or cinematic oil paintings in seconds.

Yet, looking back at the infancy of AI-generated imagery reveals a completely different atmosphere—one defined less by sleek utility and more by digital surrealism, glitchy experimentation, and psychedelic computer “hallucinations.” Long before the era of mainstream text-to-image diffusion models, early AI pictures were born from code, trial, error, and a machine’s strange, alien interpretation of reality.

1. The Rule-Bound Pioneers (The Pre-Neural Era)

Long before deep learning neural networks existed, artists and computer scientists were experimenting with algorithms to create visual forms.

In the late 1960s, artist and programmer Harold Cohen created AARON, a robotic painting system built on symbolic, rule-based artificial intelligence. AARON didn’t “learn” from vast image datasets the way modern models do; instead, Cohen meticulously coded rules of geometry, perspective, and composition into the program. AARON would drive mechanical arms to physically draw botanical and abstract shapes on paper.

While primitive by today’s standards, these early systems introduced a philosophical shift that still underpins AI art today: the human artist designs the system, but the system executes autonomous variations beyond direct manual control.

2. 2015: Google’s DeepDream and the Era of Machine “Fever Dreams”

The true cultural explosion of early neural art arrived abruptly in 2015 with DeepDream (originally coded as “Inceptionism” by Google engineer Alexander Mordvintsev).

DeepDream was originally built to help researchers understand how convolutional neural networks classified images. By taking a trained network and running it in reverse, engineers asked the computer: “Instead of telling me what you see in this picture, take what you think you see and amplify it.”

The results were astonishingly bizarre. Fed a picture of ordinary clouds, a DeepDream algorithm would scan the pixels, recognize a faint correlation to a dog’s ear or an animal eye, and aggressively exaggerate it. The output wasn’t a clean photograph; it was a hallucinogenic landscape swarming with repeating canine snouts, bulging eyes, and strange, architectural pagan ornaments.

For the first time, millions of people got a glimpse into the internal logic of a machine. It wasn’t “thinking” like a human; it was a statistical pattern-matcher throwing a fever dream onto a canvas. DeepDream created a distinct, instantly recognizable aesthetic that dominated internet culture for years.

3. The GAN Revolution: Two Networks Enter, Art Leaves

Following DeepDream, the mid-to-late 2010s introduced GANs (Generative Adversarial Networks), pioneered by Ian Goodfellow and his colleagues. This architecture fundamentally changed how AI generated pictures from scratch.

A GAN relies on an internal game of cat-and-mouse between two neural networks:

• The Generator: Tries to create fake images from random noise.
• The Discriminator: Inspects the images and tries to determine whether they are real (from a training dataset) or fake.

As these two networks competed against each other thousands of times per second, the Generator grew remarkably skilled at fooling the Discriminator.

The early outputs of GANs were wonderfully uncanny. Faces generated by early GANs often featured melting features, asymmetric eyes, and eerie, dreamlike textures that hovered somewhere between a photograph and a ghost. Artists like Robbie Barrat and Mario Klingemann began training these networks on classical paintings and historical portraits, birthing a new wave of machine-generated fine art that forced galleries and museums to reckon with algorithmic creators.

What Early AI Art Taught Us

Looking back at these early epochs of AI imagery—from rule-based robotic arms to multi-layered psychedelic hallucinations—highlights a crucial truth about human creativity: technology has always been a mirror.

The early pictures produced by neural networks weren’t polished or commercially streamlined. They were clumsy, strange, and wildly unpredictable. Yet, they laid the psychological and technical groundwork for our modern era. They proved that humanity’s relationship with technology isn’t just about automation, but about collaboration with alien intelligences that help us see the world through a completely different lens.