Another hearing took place before the Senate’s Intellectual Property Subcommittee this week, with artist/illustrator Karla Ortiz, Universal Music Group’s General Counsel Jeffrey Harleston, Emory Law professor Matthew Sag, Adobe EVP and General Counsel Dana Rao, and Stability AI’s Public Policy head Ben Brooks addressing issues relating to artificial intelligence (“AI”) and intellectual property, namely, the role of copyright law in the equitation. The hearing on Wednesday is part of a larger series that focuses on AI and IP; last month, the Committee hosted a hearing titled “Artificial Intelligence and Intellectual Property – Part I: Patents, Innovation, and Competition.”
Unsurprisingly, fair use – the legal doctrine that permits the unlicensed use of copyright-protected works in certain circumstances – dominated much of the discussion during the hearing this week, which was entitled, “Artificial Intelligence and Intellectual Property – Part II: Copyright.” The emphasis on fair use is unsurprising, of course, given the frequency with which it has been raised in connection with – or in response to – the growing number of lawsuits that are being filed over the training and output of generative AI models like ChatGPT, Stable Diffusion, etc.
As recently as this week, for instance, the plaintiffs that filed an unfair competition, negligence, invasion of privacy, copyright infringement, etc. case against Google and Alphabet over their allegedly unauthorized use of consumers data to develop/train AI models like the ones underlying the Bard chatbot appeared to be looking to get ahead of potential fair use arguments that Google and co. may make. According to the plaintiffs’ complaint, “The defendants’ wholesale collection and use of copyrighted material, with no option for copyright owners to opt out, far exceeds any reasonable interpretation of ‘fair use.’”
Another key takeaway from the hearing comes in the form of arguments in favor of the adoption of a federal right of publicity cause of action, which was raised by a number of the individuals testifying before the IP Subcommittee as a way to deal with some of the issues that are coming to light as a result of the development of generative AI models.
Against that background, there are a number of key points from the witnesses’ testimony statements that center on fair use and right of publicity that are worth noting. They are as follows …
AI and Fair Use
– Training and Outputs: Matthew Sag – who is a law professor in AI, machine learning, and data science, addressed the fair use doctrine at length in his testimony, stating that one of “the principal copyright questions” that law makers must consider is “the legality of using copyrighted works to train machine learning models, without express consent.” Training generative AI on copyrighted works is “usually fair use because it falls into the category of non-expressive [works],” he said, alongside the likes of “reverse engineering, search engines, and plagiarism detection software.”
– “Free learning”: Stability AI’s Ben Brooks similarly spoke to fair use, stating that training generative AI models is “an acceptable, transformative, and socially-beneficial use of existing content that is protected by the fair use doctrine and furthers the objectives of copyright law, including to ‘promote the progress of science and useful arts.’ These models learn the unprotectable ideas, facts, and structures within a visual or textual system, and that process does not interfere with the use and enjoyment of original works.” Such “free learning of facts about our world is essential to recent developments in AI,” according to Brooks, and “it is doubtful that these groundbreaking technologies would be possible without it.”
– U.S. leadership: Brooks also asserted that the U.S. “has established global leadership in AI due, in part, to a robust, adaptable, and principles-based fair use doctrine that balances creative rights with open innovation. Other jurisdictions, including Singapore, Japan, and the European Union, have begun to incrementally revise their copyright laws to create safe harbors for AI training that achieve similar effects to fair use.”
– Two critical questions: Reflecting on fair use, Adobe’s Dana Rao maintained that from a copyright perspective there are two core questions at play: (1) “Is the output image a copyright infringement of an image that was used to train the AI model? and (1) Is using a third-party image to train an AI model permissible under fair use?”
For the first question, she said that “the current technical understanding is that an output image is a new image ‘hallucinated’ from the user’s input text prompts and is not reusing copies of the images that were part of the training dataset to simply assemble a ‘composite’ output image. The input images are used to extract facts for training the model and its weights.”
For the second question, she said that “using an image to train an AI model would typically be considered a transformative use because an AI model, on its own, is a software program, which is very different than the original image. However, if the output of the AI model is substantially similar to a copyrighted work that it had ingested, the question remains whether fair use would be applicable, even though training the model itself may not have been considered a copyright infringement.”
Both questions are currently “the subject of several ongoing litigations and will eventually be decided in court or by Congress,” per Rao.
– “Ill-gotten” data: Artist Karla Ortiz said that AI companies, which are “reap[ing] untold billions in funding and profit,” have unsurprisingly “assured everyone that what they are doing is fair, ethical, and legal.” Part of the issue, however, is that “the artists who made the works that these AI’s rely on have never been asked for their consent [and] have not received any credit, let alone any compensation. In any other sphere, these practices would offend basic principles of fundamental fairness.”
Ortiz also asserted that this issue of whether such use of fair “has not yet been litigated,” noting that “the courts are beginning to weigh in,” including in the Andersen v. Stability AI case, in which “a federal judge sustained important parts of a complaint filed by coders challenging the use of their code as training data for Generative AI models without regard to the requirements of the open-source licenses that code was subject to.”
– Copyright law “violations”: Finally, while Universal Music Group’s Jeff Harleston did not explicitly refer to fair use in his written testimony, a read between the lines (and the record label’s efforts in the past) suggests that he and UMG do not view generative AI training or outputs as being shielded by fair use: “Copyright law has clearly been violated” by “AI-generated, mimicked vocals trained on vocal recordings extracted from our copyrighted recordings.”
A Federal Right of Publicity
On the issue of right of publicity, all of the witnesses save for Stability AI’s Ben Brooks, who did not address the issue in his written testimony, seemed to favor – or at least acknowledge – the potential benefits of for a federal statute.
– Fundamental rules for AI: Harleston stated that while “copyright law is largely fit for purpose,” there are “cracks in the foundation” that are important to address by way of “fundamental rules of the road that enable that development and ensure creators are respected and protected.” Specifically, he urged the Senators to enact a federal Right of Publicity statute, saying, “Deep-fake and/or unauthorized recordings or visuals of artists generated by AI could lead to consumer confusion, unfair competition against the actual artist, market dilution and damage to the artist’s reputation and brand – potentially irreparably harming their career. An artist’s voice is the most valuable part of their livelihood and public persona, and to steal it – no matter the means – is wrong.”
– Ortiz also spoke to the need to “pass laws expressly authorizing those who have had their data used to train AI models without their consent the right to vindicate those rights in federal court and to seek statutory damages,” noting that “this can take the form of passing a law authorizing a federal civil right of publicity cause of action.” She noted that “while under the laws of many states, a civil plaintiff may bring a case asserting violations of the right of publicity, it is often the case that many of those cases face difficulties due to preemption based on the Copyright Act or under state anti-SLAPP laws. A federal law would eliminate many of those hurdles.”
– Rao also held that “a federal right of publicity could be created to help address concerns about AI being used without permission to copy likenesses for commercial benefit. The potential for AI to be used for economic displacement is a critical problem to solve, and we believe this Committee should support a legislative solution for it.”
– “Hodgepodge” of state laws: “There are limits to what Congress can do to address” the issues posed by generative AI, Sag stated, asserting that he believes that “a national right of publicity law is needed to replace the current hodgepodge of state laws, and that we are overdue for a national data privacy law.” He further claimed, “If generative AI re-created someone’s distinctive appearance or voice, that person should have recourse under right of publicity. Congress should enact a national right of publicity law to ensure nationwide and uniform protection of individuals’ inherently personal characteristics.”
Share
