“It’s very Orwellian how the tech industry manages to change terminology on us. It’s not data and content to us, it’s music, it’s photographs; it’s not file sharing, it’s stealing – it’s simple.” – Ashley Irwin
The House of Representatives’ Subcommittee on Courts, Intellectual Property and the Internet today held the first of several planned hearings about the impact of artificial intelligence (AI) on intellectual property, focusing in this initial hearing on copyright law. The witnesses included three artists, a professor, and an attorney with varying perspectives on the matter, although the artists all expressed similar concerns about the potentially dire effects of generative AI (GAI) applications on their respective industries and careers.
Both Ashley Irwin, President, Society of Composers and Lyricists, and Dan Navarro, a Grammy-nominated songwriter, singer, recording artist, and voice actor, took particular issue later in the hearing with the tech industry’s habit of referring to the works it uses to “train” GAI as “data”.
“I appreciate the technology and the technical use of terms, but my music isn’t data,” Navarro said.
Irwin agreed, and suggested the tweak in terminology may be deliberate. “I’m a little stressed that we’re calling this ‘training,’ because in my mind we train athletes or animals, we don’t train machines,” Irwin said. “We equip machines with data, or as we like to call it, ‘music.’ It’s very Orwellian how the tech industry manages to change terminology on us. It’s not data and content to us, it’s music, it’s photographs; it’s not file sharing, it’s stealing – it’s simple.”
Chris Callison-Burch, Associate Professor of Computer and Information Science, University of Pennsylvania; Visiting Research Scientist, Allen Institute for Artificial Intelligence, said that his first foray into the large language models behind OpenAI “pitched my career into an existential crisis.” He said the technology seemed to have solved many of the problems he had spent his career researching and he wondered if there was any role for academic research going forward. However, he has since “calmed down” and has a largely optimistic outlook for the ethical use of the technology, but said that without quick and forward-thinking action to legislate in case of emergency, careers such as a paralegal could “go the way of the lamplighter.” He said in his written testimony that Congress’ role “could be as simple as providing job retraining for displaced professions, or as complex as creating a new [Works Progress Administration] WPA,” which was established to provide jobs for victims of the Great Depression.
Sy Damle, a partner with Latham & Watkins and former General Counsel of the U.S. Copyright Office, had a more optimistic view for the future of AI and copyright, and said the existing laws are sufficient to set the bounds for GAI, as the law has done for new technologies before this, including the VCR, Napster and software APIs. Damle argued at one point that the GAI learning process is similar to the human learning process and what matters is the output in terms of applying copyright law. “The copyrighted works are being used not to create a collage, but to learn statistical facts about the works themselves,” Damle said. “It’s a very similar process to the way humans learn.” However, the difference is that GAI requires billions of pieces of works to reach the same level of knowledge that a human does after reading just three or four books.
In March of this year, the U.S. Copyright Office announced a new statement of policy on “Works Containing Material Generated by Artificial Intelligence” following an increase in requests to review works created GAI. In addition to the more high-profile examples, like the Zarya of the Dawn and Thaler cases, the Office said it has also received “applications that have named AI technology as the author or co-author of the work, or have included statements in the ‘Author Created’ or ‘Note to Copyright Office’ sections of the application indicating that the work was produced by or with the assistance of AI. Other applicants have not disclosed the inclusion of AI-generated material but have mentioned the names of AI technologies in the title of the work or the ‘acknowledgments’ section of the deposit.”
Damle noted that the guidance deals with “one extreme” on a spectrum of scenarios in which GAI can be used to create a work. “There’s going to be a big gray area where there’s a human and AI together creating output,” Damle said. “I think that’s going to be a very common situation going forward.” He added that where a human has some control of the output, the situation can be compared to a photographer’s interaction with and use of a camera.
Jeffrey Sedlik, President & CEO of the PLUS Coalition and a photographer himself, said that “photographers are used to the advancement of technology; we anticipate it will continue to develop, we accept that. Our concern is our works are being used without our permission.”
The purpose of the IP Subcommittee hearing was to determine whether there’s a need – or a way – to introduce legislation that would help to compensate for or protect artists from the type of widescale threat of infringement some are afraid GAI poses. The artists all agreed they would not object to their works being used for training GAI if they were compensated, but how that would be implemented eluded most of the witnesses.
Representative Deborah Ross (D-NC) asked Callison-Burch if the industry practice keeping a record of how and whether copyrighted works are being used to train GAI would help in determining a potential compensation scheme for artists. “The tricky part is there are a trillion words worth of text in training data sets, and each author represents a vanishingly small portion of that,” Callison-Burch said. Since the outputs would be based on the volume of each artist’s contribution, checks would be infinitesimally small—something like two cents—he added.
Ross also noted, however, that Sam Altman, the CEO of OpenAI, had been on the Hill this week and indicated that new versions of OpenAI are contemplating ways to compensate copyright owners for content and style. “When we’re working on new models, if an AI system is using your content or style, you get paid for that,” Altman said, according to Ross. “And I hope he’s going to follow through on that, because Mr. Altman has a lot of money [with which] to pay [artists],” Ross commented.
Representative Kevin Kiley (R-CA) said the GAI space is evolving so quickly that the topics discussed during the hearing could be moot before the Subcommittee even gets to the point of implementing solutions. “I worry that many of the issues we’re talking about here might be obsolete within a matter of years, as technology evolves,” he said.
Subcommittee Chair Darrell Issa (R-CA) said he worries that GAI could eventually have broad implications across all areas of IP and essentially swallow creativity and innovation. “Computers can generate an infinite amount of combinations of copyrighted material. They could create a body of copyright that could eclipse future innovation, pushing out tens of hundreds of trillions of songs and variations of art,” Issa said.
Irwin said in his testimony that it ultimately just comes down to a “consent, credit and compensation” model that will ensure artists continue to be incentivized to create.
“You put in a lot of hours from a very young age to do what you do and get good at it…. If the machines can do it that quickly, what is the incentive for us to keep going? We love doing it, but at some point, the love doesn’t feed your family – there has to be a way for us to coexist.”