USA Trends and Developments Contributed by: Steven Schindler, Katherine Wilson-Milne, Thomas Kline, Eden Burgess and Aaron Haines, Schindler Cohen & Hochman LLP
4. Modification or arrangement of AI-generated outputs A work involving AI-generated material may be eligible for copyright if the human author modifies the work through selection, co-ordination and arrangement of the AI-generated materials. The overall work may be eligible for copyright, but the AI-generated materials are not copyrightable on their own. This conclusion is consistent with the Office’s deci - sion regarding Kristina Kashtanova’s copyright regis - tration application for the graphic novel Zarya of the Dawn (US Copyright Office, Decision Cancelling Reg - istration (Registration # VAu001480196) (21 February 2023)). The Office issued a limited registration that provided Kashtanova with copyright protection only for the novel’s text and its arrangement (authored by Kashtanova), but not the images produced by AI. In another application, Jeffrey Allen refused to iden - tify aspects of his work “Théâtre D’opéra Spatial” that were created by AI (US Copyright Office Review Board, Decision Affirming Denial of Registration (Cor - respondence ID: 1-5T5320R; SR # 1-11743923581) (5 September 2023)). The Office denied Allen’s regis - tration because he refused to disclaim AI-generated components of the work. Allen appealed the decision to a federal district court, where it remains pending (Allen v Perlmutter, No 24-02665 (District of Colo - rado)). While the above cases have already become seminal in the short history of AI and copyright, the Report cautions that AI-related determinations are fact-spe - cific and must be made on a case-by-case basis. Copyright infringement guidance Just as the law regarding AI and copyright registration is developing, so too is the law governing AI and copy - right infringement. In Part 3 of its Report, the Office confirmed that using copyrighted materials to com - pile an AI training data set and develop an AI training process implicates the right of reproduction held by the copyright owners. AI outputs can also implicate the copyright owner’s right of reproduction when the output “replicates or closely resembles copyrighted works” (Report, Part 3 at p 31).
The Report explains that the primary defence in these situations comes from the doctrine of “fair use”, con - sisting of four factors, and notes that the greatest weight is given to the first and fourth factors. 1. The purpose and character of the use, including whether such use is of a commercial nature or is for non-profit educational purposes This factor weighs in favour of fair use when the work is transformative of the original copyrighted materials. Although “transformative” is not well defined, an AI training model’s use of copyrighted material is likely transformative when the output is used for research or in a closed environment. On the other hand, an AI training model using copyrighted materials to create substantially similar outputs is unlikely to be trans - formative (Report, Part 3 at p 46). The Report notes that AI developers can implement restrictions on their training models to ensure more transformative outputs. For example, the model could be trained to “reject[] requests for excerpts of copyrighted works or even refuse[] to generate expressive works” (Report, Part 3 at pp 46–47). Regarding the output’s commercial nature, the analy - sis does not depend on whether the AI model serves a commercial purpose, but on whether the output serves a commercial purpose. 2. The nature of the copyrighted work AI training models that use more expressive copy - righted materials such as visual art – rather than func - tional works like computer code – will have less justi - fication for copying being fair use. 3. The amount and substantiality of the portion used in relation to the copyrighted work as a whole The Office found that AI developers often use entire works to create training models, a practice that weighs against a fair use defence, but also acknowledged that this concern may be diminished when the outputs have a transformative purpose or little to none of the copied material is made available to the public. 4. The effect of the use upon the potential market for, or value of, the copyrighted work The Office acknowledged that AI models can greatly damage the potential market for copyrighted works. It
603 CHAMBERS.COM
Powered by FlippingBook