By: Andrew J. Costa, Associate
AI unsurprisingly continues to demand attention, and in this update, we will discuss two developments drawing a lot of recent focus – first, the US Copyright Office’s cancellation of Kristina Kashtanova’s copyright for the illustrations in her work Zarya of the Dawn because they were AI-derived (the entire work and story are still protected), and second, oral arguments before the UK Supreme Court on Dr. Stephen Thaler’s continuing mission to have his AI, DABUS, declared a patent inventor. Both exemplify the ongoing difficulty faced by inventors, creators and their lawyers to harmonize existing intellectual property law with growing use of AI technologies. While the Copyright Office’s decision may seem to provide yet another clear answer to the uncertainty surrounding AI, questions linger because Kashtanova may still respond or challenge the decision in the Courts. To combat such uncertainty, the Copyright Office launched an initiative on March 16, 2023 to examine AI. The initiative includes new guidance concerning “Works Containing Material Generated by Artificial Intelligence” and a series of public listening sessions to occur throughout April and May 2023 that specifically seek to better understand the “impact of generative AI in creative fields.” Likewise in the patent context, the UK Supreme Court has not rendered it decision on AI-inventorship yet, and Dr. Thaler’s quest in the United States continues as he seeks certiorari before the United States Supreme Court (the petition is due March 19, 2023). Suffice it to say, legal questions surrounding AI inventorship and authorship will continue to make headlines.
Zarya at the US Copyright Office
The Zarya story began in September 2022, when artist Kristina Kashtanova’s created the comic book Zarya of the Dawn and submitted her work to the Copyright Office for registration. The 18-page comic includes Kashtanova’s original story line in text, and several illustrations (see here for pictures of her comic). At that time however, the Office was unaware Kashtanova “created” the illustrations by supplying “prompts” to an AI tool called Midjourney that generated the illustrations through an AI algorithm. The Office originally granted the application registering the entire work, including the storyline text, compilation of illustrations and the illustrations themselves. However, shortly after registering the work, the Copyright Office learned through social media that Kashtanova’s illustrations were AI-derived, and the Office notified her it intended to cancel her registration citing inaccuracy and incompleteness. At issue was the authorship of the work – was the AI the author? Could the AI be the author for copyright purposes? Correspondence ensued between the Copyright Office and Kashtanova’s counsel, ultimately leading to the Copyright Office’s final determination on February 21, 2023 that AI-derived works are not copyrightable subject matter because they are not original works of human authorship.
While the Copyright Office agreed that the original story text and arrangement of the images are all protectable subject matter under 17 U.S.C. § 102(a) as original works of authorship, the Copyright Office determined the illustrations standing alone cannot be protectable because “works of authorship” are “uniformly limited  to the creations of human authors”. In support of this opinion, the Copyright Office relied on various decisions, both administrative and judicial, dating as far back as 1884, and the U.S. Copyright Office Compendium that explains that the Copyright Office “will refuse to register a claim if it determines that a human being did not create the work.” U.S. Copyright Office, Compendium of U.S. Copyright Office Practices § 313.2 (3d ed. 2021). One argument Kashtanova offered in support of her sole authorship in using generative AI, was that her use of the AI is analogous to a photographer using a camera as a tool of expression to capture her idea. However, the Copyright Office disagreed, discussing at length that AI use is fundamentally different because of the unique way a user “prompts” the Midjourney AI to, which actually creates the work.
The key element is predictability. The Copyright Office explains “Midjourney does not interpret prompts as specific instructions to create a particular expressive result . . . [t]he process is not controlled by the user because it is not possible to predict what Midjourney will create ahead of time”. Kashtanova argued however that she did control the outcomes because she directed the creative process through a lengthy (and specific) series of trial-and-error, through prompts she composed, that eventually yielded the final images – however the Copyright Office found this also unpersuasive. The Copyright Office explained a “prompt may ‘influence’ [the] generated image, but . . . does not dictate a specific result”. This dynamic results in a chasm between what a user directs the AI to create and what the AI actually creates. The key element in the Copyright Office’s assessment here is “a lack of sufficient control over generated images [for the author] to be treated as the ‘master mind’ behind them.” Thus the Copyright Office determined the Midjourney AI created the work, and cancelled the registration for failure to exclude non-human authorship under 37 CFR 201.7(c)(4), but reissued copyright registrations to Kashtanova for the story’s text and for the illustrations as a compilation.
Regardless of whether Kashtanova appeals the decision to the courts or accepts this more limited copyright protection for her work, for now, it seems that in the United States, AI-generated artwork cannot, on its own, be copyrightable subject matter and must be disclosed or excluded in an application for registration. One thing that is unclear however, is whether the Copyright Office’s reasoning will apply to other generative-AI solutions that might work differently than Midjourney. Is the lack of predictability and control a consistent element of all AI generated works? Or could another AI-tool operate with enough predictability and control to support human authorship? The answer to such questions remains unknown and subject to potential future developments (legal and technical).
In the meantime, the Copyright Office’s new guidance published on March 16, 2023 as part of the initiative for works containing material generated by artificial intelligence seeks to reinforce the principals of control and predictability that supported the Office’s Opinion in Kashtanova. The guidance reiterates that human authorship is the foundation of copyrightability, and works wherein the AI directs the creative process, rather than the human author, are not copyrightable. Importantly, the guidance specifically clarifies that use of AI tools, like Adobe’s Photoshop, will not prevent copyrightability, explaining “what matters is the extent to which the human had creative control over the work’s expression and ‘actually formed’ the traditional elements of authorship.” See Burrow-Giles Lithographic Co. v. Sarony, 111 U.S. 53, 60 (1884). Furthermore, the guidance explains applicants have a duty to disclose AI-generated elements, so that the Office may exclude these elements from an application. Interestingly, the guidance cautions that previously registered works will need to be updated in order to avoid cancellation. So while authors will need to consider the nature and extent of their AI use more carefully, it remains to be seen where, in practice, the Office will define that line between protectable use and not.
To that end, the Copyright Office has scheduled a series of public listening sessions throughout April and May 2023 specifically designed to better understand creative use of AI. Each session will address AI as it relates to the type of underlying copyrightable work, including literary works, visual works, audiovisual works, and music and sound recordings. Artists, creative industries, AI developers, researchers and lawyers are all invited to attend the remote sessions on April 19, May 2, May 17, and May 31. The Office plans to use these sessions to inform a notice of inquiry to be published on the Federal Register later this year. (For the full Copyright Office determination including the letters between Kashtanova’s attorney and the Office, see here; for the published guidance, see here; and to attend the public listening sessions, see here
(For the full Copyright Office determination including the letters between Kashtanova’s attorney and the Office, see here; for the published guidance, see here; and to attend the public listening sessions, see here.)
Thaler at the UK Supreme Court
While Dr. Thaler prepares to petition the US Supreme Court for a writ of certiorari in his quest to name his AI invention machine, DABUS, as the inventor of two devices in the US, his arguments have reached the highest court across the pond in England. On March 2, 2023, the Supreme Court of the United Kingdom heard oral arguments regarding whether, under English law, an AI can be the named inventor on a patent application. Unlike its American counterparts, however, the British Supreme Court will need to contend with more nuanced questions of inventorship. First some brief background.
In the United Kingdom, Dr. Thaler filed two patent applications naming DABUS as the inventor. Similarly to the US rejections, the UK Intellectual Property Office (UKIPO) withdrew Dr. Thaler’s patent applications for failure to name a human inventor under Section 13(2) of the Patents Act of 1977. Thereafter, both the High Court and UK Court of Appeal upheld that determination limiting inventorship to human inventors. On appeal to the Supreme Court of the UK, three key issues are before the Court: (1) Does Section 13(2)(a) require a human be named the inventor in all patent applications, including where an applicant believes the invention was created by an AI machine? (2) Does the Act provide for a patent without a human inventor? (3) Is the owner, creator, or user of an AI machine entitled to the grant of a patent for an invention made by that machine? (Notably, the United States Court of Appeals for the Federal Circuit did not opine on (2) or (3), nor did it need to, but instead directly addressed the statutory interpretation of the comparable US statute under the Patent Act.)
In England, Dr. Thaler similarly argues that the British Act’s language, statutory construction and interpretation do not preclude AI inventorship, and further argue that naming himself would be a false statement because of his good faith belief that DABUS is the inventor rather than himself. Dr. Thaler also argues the term “person” in §13 of the Act is a procedural matter that does not impose a substantive bar to patentability. Conversely, the UK Comptroller General (representing the UKIPO) argues that the definition of “person” strikes at the core of inventorship that is distinctly human. The Comptroller argues an inventor under the Act must be a natural person, because of the precedent set by the House of Lords in the Yeda decision of 2007 wherein they held an inventor must be “the natural person who came up with the inventive concept”. The UK Supreme Court has yet to issue a ruling in this matter.
However, broader questions concerning how the UK patent system will accommodate growing AI adoption by inventors remain without clear answers. During oral argument, two such questions were: (1) whether an invention without a human inventor can be protected at all under British patent law, and if not, (2) which human is properly named on a patent application for an AI-derived invention. The first may prompt debate, as some might argue AI technology simply cannot invent without human contribution. But as AI develops, this could easily change. In both the US and UK, it will be important for lawmakers to consider whether, as a matter of policy, patent law should incentivize AI-inventorship and patentability in the interest of national competitiveness, especially considering its growing use and adoption. While philosophically human inventorship has enjoyed a special status, there are valid concerns that precluding patent as a protection for AI-derived technology may stymie the next wave of innovation. Turning to the second question, it may also seem obvious to name the human who used the AI as the inventor on an application, but the UK Supreme Court probed deeper, considering whether the AI’s developers or even those who trained the AI models should more properly be named inventor. In this way, inventorship could become a kaleidoscope. One final aspect that was not addressed was whether the owners of the AI’s data sets could possibly have a claim of inventorship too. If the UK Supreme Court chooses to address these issue directly, the impact could be profound. Maybe it’s soon the time we just ask the AI.
(To read more about Dr. Thaler’s US efforts before the Federal Circuit and more thorough dive into AI-inventorship, see our last post here.)