A short thread on the long and complicated history of AI art and copyright. This is Quadratic Gaussian, a piece of digital art generated by Michael Noll exactly 60 years ago (August 28th 1962) and successfully copyrighted two years later in 1965 after a lengthy debate.
Conversation
Michael Noll was an intern at Bell Labs working on the digital recording of speech data. Although he had no formal artistic education he has been immediately fascinated by the artistry of computed patterns.
1
4
Gaussian-Quadratic was part of a series of generations published under the title "Technical patterns for 7090”. It immediately stood out due to its relative similarity to a painting by Picasso, Ma Jolie. collections.vam.ac.uk/item/O1193785/
1
1
2
As he felt a personal connection to this work, Noll attempted to exhibit it and copyright it. As he recalled in a short autobiographical essay (jstor.org/stable/1575947) this was not an easy process and the copyright application has to be resubmitted several times:
1
2
“At first they refused since a machine had generated the work. I explained that a human being had written the program that in- corporated randomness and order. They again refused to register the work, stating that randomness was not acceptable. (…)
1
2
3
I finally explained that although the numbers generated by the program appeared "random" to humans, the algorithm generating them was perfectly mathematical and not random at all.” (Michael Noll, The Beginnings of Computer Art in the United States: A Memoir, 1994)
1
3
3
This last line of argument must have been convincing enough. On April 6th 1965, Gaussian Quadratic was finally registered by the copyright office of the US as a “Geometric Abstract Forms” (I have been able to track the original computer card in copyright.gov)
Replying to
What is really striking is that the debate over #AIArt and copyright hasn’t changed all that much since 1962-1965. The two main issues are still personal originality (has the art been made by someone?) and unicity (is the work unique and not a large universe of “possible” works?)
2
4
6
While the material infrastructure and the algorithms used by Noll are radically different, the process is similar: Noll has defined the parameters of the generation and then he has chosen one particular work, in the same way as AI artist can associate a prompt and a seed.
Quote Tweet
5. Regardless of preexisting IP can the model creation be protected? It's not settled. I long thought this was not the case but I'm now increasingly verging on yes.
Show this thread
1
5
Another aspect that remain very much actual is that Bell Labs had contemplated applying some form of corporate ownership to generative art (and Noll seemed OK with it as he did not initially considered himself a professional artist).
1
1
2
Yet, generative art was not considered sufficiently “serious” to the point that early experiments were scrapped: “public relations folks at AT&T became worried that the Bell Telephone companies that supported Bell Labs would not view computer art as serious scientific research.”
1
2
Concretely for Noll it meant that an early exhibit of generative art at Howard Wise Gallery in April 1965 (photo) was nearly canceled Conversely he got back his full ownership to the generated work, as Bell Labs even supported his own copyright application to distance themselves.
1
2
The leading creators of AI visual model seems to go through the same road and acknowledge the generator as the "author" although not because AI art is "not serious", but rather because it is *too* serious and create potentially huge issues of public accountability.
1
1
3
In a strange twist of history, AIArt brings back digital art to its original roots and its connection to modern art conception of a work as a “communicative” act.
1
3
Noll has not hesitated to put Gaussian Quadratic on display in a New York Gallery, simply because it was not wholly different from what Warhol was doing in the same time (that’s an alt version of Gaussian Quadratic in a picture of the original 1965 exhibit)
1
1
4
If a working version of #midjourney or #stablediffusion had magically appeared in the 1960s, the small emerging community of digital artists would have completely shocked and amazed. But that would not have radically challenged their own views of authorship and art making
1
2
In fact the only significant change is immediately visible when I try to run the elaborated description of Gaussian Quadratic as a prompt in StableDiffusion: Picasso is not just a very subdued analogy but the massively visible stylistic influence.
2
4
On top of the generation process, there is now this immense cumbersome collective memory of all major artworks made that do complicate further the assignment of authorship (at least if the prompt are not careful enough to dilute the mention of influential authors and artworks.)
1
1
3
For additional details, I really recommend this extensive study of on the long history of AI Art and copyright (and which brought me to Noll in the first place) library.oapen.org/bitstream/hand
1
6