Gerhard Richter recently won a lawsuit about unauthorized photos of his famous window at Cologne Cathedral.
In a TV documentary I saw that the colour pattern of the window was created as follows:
If this process was shown correctly on TV, does Gerhard Richter hold the copyright?
Randomness or computer generated randomness can form the basis for copyright, as long as the input of the computer is below a barrier.
Let me correct the statement of facts first:
The artist chose 74 out of 300 glass colors and ordered a computer to generate a single panel of numbers to be used as a base. Then he assigned his colors to those numbers on the panel, testing out the base feel. It was the artist who took the random sample and assigned color values till he liked this first iteration, and already the assiging of colors diminished the computer input's contribution.
Then, following more complex rules than the computer was tasked with, alterations were made to generate the whole set of windows. There was mirroring and flipping involved between the panels, done by the artist. Whenever a pattern or symbol emerged to his eye, he would deliberately alter the generated pattern to stop the mind from seeing such. That's all acts of human authorship, diminishing the computer input's contribution further.
Then the glass was made with a glue-on version to create seamless colors, unlike classic lead bar construction, which was part of the artist's vison and authorship, not something the computer told him.
The result is, that the computer had one single input on the work, and that input wasn't even followed in construction.
I said, the computer had a tiny input. It was de minimis, just a tool, not the generator of the final work. That means it has little to no bearing on the copyright. de minimis is the legal concept of something being a negligibly small part of the creation and is used in most Western copyrights, which are very similar due to the Berne Convention anyway.
In the US, the Copyright Office says, if you use AI or other generation methods that are non-human more than de-minimis, you get no copyright:
In early 2023, the Office announced the launch of a broad AI Initiative and issued a statement of policy providing guidance on the registration of works incorporating AI-generated material (the “Guidance” or “AI Registration Guidance”).12 The Guidance reiterated the Office’s longstanding position that human authorship is an essential requirement for copyright protection in the United States.13 It explained that if a work contains more than a de minimis amount of AI-generated material, the applicant should disclose that information and provide a brief statement describing the human author’s contribution.14
Théâtre D’opéra Spatial did not disclose, even after several requests to clarify, and had a larger amount than de minimis, thus copyright was denied:
After carefully examining the Work and considering the arguments made in the First and Second Requests, the Board finds that the Work contains more than a de minimis amount of AI-generated content, which must be disclaimed in an application for registration. Because Mr. Allen has refused to disclaim the material produced by AI, the Work cannot be registered as submitted.
But you can have copyright in those parts not made by AI, as Zarya of the Dawn showed.
Let's apply the modern AI standard to the Richter Window and look at the generation method. Assuming the number panel was AI made (it wasn't it was a dumb early 2000s random number generator), then still the work would be only having de minimis input by the computer, the choices were all human.
Generating one of 500 and choosing one of them, as long as the choice and re-work after it is human, we are at de minimis.
If the initial first iteration panel was ordered with an AI prompt specifying the available colorspace, then we still had a massive human oversight and redistribution of colors to match the wanted aesthetic, so it can get into de minimis easily by a lot of post processing - e.g. the alterations Richter did to the generated panel
If the panel was, however, designed in a fashion where the AI did the whole layout, with all the rules above just given to the AI and then taken unaltered, you'd look at a work without an Author. But on the other side, the complex prompt that generated the image, in its length and specificity, would have broken the creativity barrier to have a copyright. As in, the text made by the human, not the image that the tool spat out.
And still, taking the single panel and making it part of a multi-panel installation in which the patterning of the first panel was altered to generate the other panels might result in the AI generated image to be considered just about de minimis.
So far, there is no strict or clear creativity barrier how much alteration needs to be done to an AI made work to turn it from uncopyrightable slop into copyrightable art, but the standard is evolving slowly.
But for the Richter Window under the facts restated above? That's clearly de minimis computer usage and thus under copyright until some 7 decades after Richter's death - which means it's more than 70 years in the future, because the author is alive.
In the computer generated pattern? Maybe.
Consider this analogy. Nature randomly creates a large number of wildflowers that are randomly distributed. G chooses the most pleasing of these wildflowers and photographs it. Does G own copyright in the photo? Yes. Does G own copyright in the flower? No.
If G keeps the computer generated pattern or the location of the wildflower secret, then they have a strong case that any subsequent copying is of their copyright work, not the original work.
Of course, if the original computer generated image is random within parameters set by G, there’s an argument that they have copyright in the original, too.
