Can AI art be copyrighted? The debate has been raging following the new wave of generative AI since August 2022.
Copyright was granted and then removed from pieces like Zarya of the Dawn. Epic online wars were fought. Recent guidelines from the U.S. Copyright Office reaffirms that purely generated images are in the public domain and have just been confirmed by a Federal judgment.
Let’s look at Quadratic Gaussian, a piece of abstract art by Michael Noll. While it does not look that impressive, Quadratic Gaussian was generated using an IBM 7090 in August… 1962. And it was successfully copyrighted two years later in 1965 after a heated debate that raised numerous concerns that are still very much actual.
In a short autobiography, Michael Noll recalls that he was recruited by Bell Labs in 1961 to work on speech data. Aside from his daily job as an engineer, Noll also had a lifelong passion for modern art and “was a frequent visitor” to the MOMA.
One day, something odd happened. Noll’s colleague was experimenting with computer-generation of voice graph data and due to a programming error “Lines went which way over the plot. We joked about the abstract computer art that he had inadvertently generated.” And suddenly, Noll had the inspiration to re-do the error and “create computer art deliberately”.
Like with contemporary AI, it was a lot of trials. Noll combined numerous potential distributions, dimensions, formats… The final selection was part of a collection aptly named "Technical patterns for 7090” and included a piece Noll was particularly fond of, “Gaussian-Quadratic”. He realized later on that the patterns reminded him of a painting by Picasso, Ma Jolie and became so enthused with his own creation that he began to “use colored-ink markers to trace over the pattern to produce customized art for my colleagues at work.”
Noll's experiments were initially supported by his superiors. We were right in the middle of the first AI summer and despite the obvious limitations of the computers at the time, there were a wealth of ambitious projects in automated translations or computer visions. At Bell Labs, other engineers eagerly strived to create early computer-generated music or animations.
Through a colleague, Noll connected with a New York art seller, Howard Wise. Wise initially intended to hold a more scientific exhibition on human and computer perception of images. Yet, the generations of Noll immediately stood out as they were consciously made with an “aesthetic or artistic effect”. The Howard Wise Exhibit of "Computer-Generated Pictures” opened in April 1965 and extensively featured Noll’s work.
A black and white version of Quadratic Gaussian in the Howard Wise Exhibit in April 1965.
At the time, no artists objected to the de-humanization of generated art. No one even challenged Noll’s claim to have produced a work of art. The exhibit was covered by Stuart Preston in the New York Times on April 18 1965: “the wave of the future crashes significantly at the Howard Wise Gallery”. While acknowledging that for now “the means are of greater interest than the end”, Preston noticed that “scientists predict a time when almost any kind of painting can be computer-generated”. For the 1960s art critic, this was not a dreadful perspective. Instead it may liberate the artist and its personal vision from the constraint of material creation: “Freed from the tedium of technique and the mechanics of picture-making, the artist will simply “create””.
Noll had in fact more issues with his home institution. He received a call from the “public relations folks at AT&T” who have grown increasingly concerned about funding art experiments instead of engineering research. The exhibition was almost suspended and could only be maintained if Michael Noll accepted to publish all his works under his name and not involve the Bell Lab in any way. And thus he became a computer-generated “artist”.
Noll initially had mixed feelings about claiming all the authorship for himself. This work could not have happened without a wider social context that included the serendipitous error from a colleague, the technical apparatus of the lab and the artistic emulation from other departments.
Since he had to act as an individual artist, Noll decided to register the copyright of his favorite work, Gaussian Quadratic. An initial application was immediately dismissed as a "machine had generated the work".
In his second application, Noll argued that he had made a lot of conscious decisions about the program and its design. His counter-argument was accepted and yet the Copyright Office counteracted that generated art “violated the “fixity” requirements of copyright law” (p. 58) as there was seemingly an element of randomness that made it possible to produce an infinite number of variants of the same work. This critic is very much relevant today. Contemporary AI is largely based on the combination of an initial prompt, a set of parameters and a “seed” (a random number that controls the initial noise of the image). In 2023 like in 1965, the Copyright Office remains wary of granting protection for an indefinite number of works. And like Noll, potential “AI artists” have to demonstrate some level of individual creation.
Recreation of “Gaussian Quadratic” by Stable Diffusion with two different initial seeds (1989 and 1990). Prompt (partly based on an original description of the work by Noll): “Ninety-nine lines connect 100 points whose horizontal coordinates are Gaussian. Vertical coordinates increase according to a quadratic equation. As a point reaches the top, it is reflected to the bottom to continue its rise. This particular proportion is vaguely similar to the painting Ma Jolie by Picasso.”
In his third and ultimately successful application, Noll underlined that the process was "perfectly mathematical" and controlled. While the applications have not been digitized, I have been able to locate the final form registering Gaussian Quadratic in April 1965 (just in time for the Exhibit) as a “geometric abstract form”.
Obviously the notion of generated work was very different in the 1960s. Noll did not train a model on a massive corpus and the reference to Picasso was only metaphorical. It remains striking that seemingly no artists objected to the de-humanization of generated art. While Noll's copyright claims were controversial, no one denied he had created a work of wart and Gaussian's Quadratic.
Paradoxically, computer generations seemed to agree more with the common definition of art in the 1960s. In the age of pop-art and hippie communes, relying on a computer was perceived at most as a mild subversive act. In fact, Noll's copyright attempts proved more controversial among his peers: according to Martin Zeilinger "other early computer artists insisted that [their] work emerged from a collaborative working-together of human and computer" and were reluctant to abide by the rules of intellectual property.
Afterwards, Noll embarked on a successful artistic career with exhibitions in England, Yugoslavia, Czechoslovakia, Brazil, Germany, India, Spain… Along with AI, computer-generated art also had his first brief “summer” that is by now largely forgotten.
At least by 2023, "the wave of the future" has finally come crashing back, not just on one gallery but on the entire Internet. It remains to be seen if it will free the artists.
Great piece, thank you!