Cybernetic Serendipity: The Forgotten Origins of AI Art

In February of this year Christie's auction house in New York staged its first auction devoted entirely to AI-generated art.

Titled Augmented Intelligence, the sale featured over 20 lots spanning sculptures, prints, and interactive experiences, achieving a total of $728,784. The collectors were unusually young for the art market: 48% of the bidders were Millennials and Gen Z, and 37% were first-time Christie's buyers

Christie’s presented the event as evidence that a new artistic frontier had arrived. But the artists, technologists and theorists who laid the groundwork for today's machine-made imagery were grappling with many of the same questions decades ago. Long before prompts, datasets and generative models entered the cultural vocabulary, a landmark exhibition in London imagined a future in which computers made art that seemed to be created by the human hand.

The show deeply influenced Harold Cohen, the creator of AARON, a computer program that translated his creative concepts into digital imagery. 

It went on to flourish as an autonomous art generation machine, in what sounds like a familiar tale in today’s artwold – except that the British-born artist invented AARON in the early 1970s.

Franciszka Themerson, Cybernetic Serendipity, 1968, exhibition poster, ICA, London. Courtesy the artist and ICA, London

Cohen was a known figure at the time – he represented Great Britain at the Venice Biennial in 1966. But he began the AARON project, for which he is now best known, after visiting the exhibition Cybernetic Serendipity at London’s Institute of Contemporary Arts. 

Curated by the ICA’s then assistant director, Jasia Reichardt, the 1968 exhibition was revelatory. For the first time, here was an arts project assembled around artificial intelligence, a term coined 12 years earlier by US computer scientist John McCarthy to encompass his theory that ‘every aspect of learning, or any other feature of intelligence, can be so precisely described that a machine can be made to simulate it’. 

More crucially, the exhibition’s radical artistic and science-based explorations created the foundation for today’s embrace of AI-generated artworks, a section of the market which was worth $3.2 billion by 2024, according to business intelligence firm Grand View Research. The value of AI art is projected to rise to $40.4 billion by 2033. Meanwhile, the 2026 Art Basel & UBS Global Art Market Report notes that 51 percent of high-net-worth collectors have purchased digitally-assisted art, with the medium now constituting 13 percent of the collections it studied.

In addition to Christie’s Augmented Intelligence, the ongoing exhibition Machine Love, a compendium of AI art at Tokyo’s Mori Art Museum, is reporting record-breaking visitor numbers. In Europe, Art Basel has introduced Zero 10, the fair’s new global initiative that brings together art and technology through curated projects. It’s a platform ready made for AI art stars like Claire Silver, Holly Herndon and Refik Anadol, co-founder of Dataland in Los Angeles, ‘the world’s first Museum of AI Arts and digital ecosystem where human imagination meets the creative potential of machines’.

In 2021, Sotheby’s Hong Kong sold Anadol’s Space: Metaverse from his Machine Hallucinations series for $2.3 million. The Turkish-American media artist feeds enormous sets of existing data (weather maps, oceanographic radar scans, museum collection images) into algorithms that build large-scale animated walls of moving art. Meanwhile, Silver trains AI systems using data gleaned from traditional painting styles as part of her practice.

Refik Anadol, Machine Hallucinations – Space : Metaverse, 2021 (installation view, DAFA). Courtesy the artist and Sotheby’s

Yet for all the apparent novelty of today’s AI art boom, many of its defining ideas were already present in Cybernetic Serendipity. Look closely at the exhibition, which toured to Washington DC and San Francisco, and the parallels are striking. Artists and scientists were experimenting with machine-generated images, autonomous creative systems, responsive environments and algorithmic music decades before such practices became central to contemporary digital art. Back in 1964, academic turned artist A Michael Noll programmed a computer to convert a painting from Piet Mondrian’s abstract Composition In Line series of 1917 into digital data to generate new versions of the original. Cybernetic Serendipity exhibited prints of the results.

In 2019, gallerist Aidan Mellor collaborated with robotics company Engineered Arts to build Ai-Da. Driven by an algorithm and featuring cameras for eyes, Ai-Da is a startlingly human-looking robot that is capable of creating paintings and drawings. In 2024, Sotheby’s New York sold an Ai-Da painting, A.I. God, for $1.1 million. 

But Jean Tinguely got there first. Cybernetic Serendipity presented two of the Swiss artist’s Méta-Matic machines – electrically-driven scrap-metal sculptures that created random, scratchy paintings. A twin-pendulum harmonograph was also on show alongside the images it created. Constructed by scientist Ivan Moscovich, pencils attached to the harmonograph’s swinging cables created complicated Moiré patterns on paper, all generated without human intervention.

Herndon has created Holly+, an online AI version of herself that makes her singing voice freely available. Upload audio files to the AI system and Holly+ will reconfigure the sounds to mimic the artist’s voice. Likewise, Reichardt dedicated a section of Cybernetic Serendipity to music-making. Musician and scientist Peter Zinovieff exhibited a computer system that composed synthesised melodies based on tunes sung or whistled into microphones by visitors to the gallery.

Gustav Metzger, Five Screens With Computer, 1968, drawing (Cybernetic Serendipity catalogue). Courtesy the artist and ICA, London

Amongst images and essays, a special edition of Studio International magazine, published to accompany the exhibition, featured Gustav Metzger’s drawings for his unrealised Five Screens With Computer monument. In the late 1950s, driven by an interest in the Dadaist movement and a personal disgust at capitalist excess, Metzger invented Auto-Destructive Art – literally, artwork that destroys itself. He aimed to create five huge, rectangular steel screens filled with thousands of mechanical components. Over a ten-year period, a computer connected to the screens would select components to eject until none remained. The ejections were to be dictated by atmospheric conditions fed into the computer. 

Of all the pieces in the Cybernetic Serendipity show, Gordon Pask’s The Colloquy of Mobiles was perhaps the most prescient. Pask was developing Conversation Theory, his idea that machines could learn and develop via self-interaction. Nowadays, machine learning – a computer systems’ ability to learn from data and identify patterns independent of human oversight – is an essential element of AI and AI-augmented art. For Cybernetic Serendipity, Pask programmed five computers to communicate with each other, conclude decisions and make their outcomes known by influencing the movement of six mobiles in the gallery.

The use of AI systems in art is an evolving conversation. In the run-up to Christie’s Augmented Intelligence sale, 6,000 artists put their signatures to an open letter asking the auction house to cancel the event on the grounds that ‘...many of the artworks you plan to auction were created using AI models that are known to be trained on copyrighted work without a license’. In response, Christie’s issued a statement saying, ‘the artists represented in this sale have strong, existing multidisciplinary art practices, some recognised in leading museum collections. The works in this auction are using artificial intelligence to enhance their bodies of work.’ The auction went ahead, with sales exceeding the initial estimate of $600,000. And there, amongst the Anadol, Herndon and Silver lots, sat (i23-3758), an early artwork generated by AARON.