Let's continue the Artificial Intelligence (AI) thread started in yesterday's post analyzing the case of Kristina Kashtanova's AI-generated graphic novel that was refused copyright registration this week.
Kashtanova, a former software engineer, wrote the text of the book Zarya of the Dawn, but the artwork accompanying the story was AI-generated using the Midjourney platform, a detail originally not disclosed in the copyright application. The U.S. Copyright Office ("USCO") issued a registration for Zarya in September 2022, but started reconsidering it at the end of October.
The office alerted Kashtanova, highlighting that, according to the Copyright Office policy, an author must be human, as established by the Code of Federal Regulations 37 CFR § 201.7 (c) (1).
The latter states that a registration may be cancelled "Where the Copyright Office becomes aware after registration that a work is not copyrightable, either because the authorship is insufficiently creative or the work does not contain authorship subject to copyright" (a rule that was applied to the "monkey selfies" case that took place between 2011 and 2018, when there were disputes about the copyright status of selfies taken by Celebes crested macaques using equipment belonging to the British nature photographer David Slater. The photographer stated he held the copyright, but the United States Copyright Office stated that works created by a non-human, such as a photograph taken by a monkey, are not copyrightable). The Office gave Kashtanova 30 days to provide details and show there was significant human participation in the creation of Zarya.
In a letter dated Tuesday, the USCO decided the images should not have been granted copyright protection. According to the Copyright Office, the images generated using Midjourney technology "are not the product of human authorship" and the platform can't be compared to other tools used by artists and therefore must be treated differently when it comes to copyright issues as Midjourney's specific output cannot be predicted by users.
Midjourney is a text-to-image application that produces images in response to a prompt written by the user. The application produces four different images, all of them based on Midjourney's training data. Additional prompts can be added and the image can be upscaled or the user can be required variations to be done, but the user can't predict the results, even though they can tweak them.
Kashtanova explained the Copyright Office that she had fed the application thousands of prompts, tweaked the images, generated further intermediate pictures and, eventually, through a process of trial-and-error, obtained the final images.
Yet the USCO still highlighted that it was Midjourney that had created those images and therefore Midjourney's AI software users can't be considered as "the authors" of those images.
Kashtanova also edited some of the images: the AI-software produced over 2,000 images that were edited in Photoshop and she then used the software Comic Life 3 to lay out the pictures in comic book form, yet some changes were deemed as minor and therefore not offering the necessary copyright protection. However, the USCO highlighted that other substantive edits to intermediate images generated by Midjourney provided human authorship and would therefore not be excluded from the new registration certificate.
The Office in the end allowed copyright protection for the novel's text (that was written by Kashtanova without the help of any AI applications) and for the way the images were "selected, refined, cropped, positioned, framed, and arranged" to create the story. A new certificate covering only these materials was therefore issued.This is not the first case involving Artificial Intelligence and copyright: in 2019 a board reviewed a ruling against Steven Thaler, a software engineer and the CEO of Imagination Engines, Inc and the creator of the AI platform, DABUS (Device for the Autonomous Bootstrapping of Unified Sentience).
Thaler, who also applied for patent protection for a DABUS-created invention (but an inventor must be an individual, so the protection wasn't granted...), tried to copyright in 2018 an AI-created image entitled "A Recent Entrance to Paradise" and picturing a fictional narrative about the afterlife. On the copyright application, the author was listed as "Creativity Machine," an A.I. software. Thaler was listed as the owner of the copyright because he is the owner of the software.
Yet, as the image didn't include an element of "human authorship", Thaler was refused the copyright and, in June 2022, he filed a case in the District of Columbia asking to overturn USCO's decision. In that case Thaler argued that he could own the copyright as the computer was the artist and produced under circumstances similar to "work for hire" contracts with the hired artist listed as the author but the company still owning the copyright. In February this year Thaler's request was rejected, while the decision of the USCO was considered as "reasonably and consistently with the law".
These case highlight that, with the rise in popularity of AI generative software such as Midjourney, Dall-E and ChatGPT, we will see more copyright cases and more legal issues to solve (think about the growing number of lawsuits regarding the unauthorized use of various types of works to train AI generators and the practice of "scraping" , that is combing the Internet with a software collecting data from a variety of sources). Just like the Hermès Vs Metabirkin case became important for NFTs and trademark applications in real life and in the virtual world, these cases are intriguing as they may lead to the creation of new guidelines regarding AI-generated artworks that would allow to understand if the level of creativity that goes into an image can be deemed as human authorship.
You actually wonder if the decision of the USCO would have been the same if Kashtanova had used not a written prompt, but two or more images created by her, that the Midjourney Bot had combined with the "/blend" prompt. Because, while in this case the result can't still be predicted, the starting points for that final images would have been based on original artwork by the author herself.
Besides, in yesterday's post, we have seen a campaign by a big brand in which AI software was used to generate images: the pictures include the name and the logo of the brand, and the images were definitely Photoshopped to be able to add these details. Now, in these cases, would the USCO say that these images can be copyrighted as there has been enough human intervention or can we deem them as not subject to be copyrighted as they were made by an AI application?
And what to do now with images generated by AI-software and printed on designer clothes? If a fashion designer employs AI-generated images in their collection, does this mean that those images are not copyrightable and therefore not original enough and can therefore be used by anybody else? And who will own the copyright of the AI-generated images showcased in international events such as the first AI Fashion Week, nobody as those were generated by a machine? In that case any fashion designer could copy those clothes and produce them, after all they are not copyrightable.
In conclusion, while Kashtanova's case closes here, this is just the beginning of an entirely new chapter for copyright law. You can bet that additions and changes will have to be done as soon as possible to decide what is AI-generated artwork, how to register them and what constitutes authorship.
Comments