Last Updated on February 23, 2023, 9:26 am ET
For Fair Use/Fair Dealing Week 2023, Association of Research Libraries (ARL) senior director of Scholarship and Policy, Judy Ruttenberg, interviewed Jonathan Band, counsel to the Library Copyright Alliance (LCA), which consists of ARL and the American Library Association (ALA). Jonathan drew on his deep substantive knowledge of the application of copyright law to information technology to share his views on fair use and artificial intelligence in this brief interview:
Judy: As recently as 2022, the US Copyright Office denied registration of a work based on its authorship solely by machine or AI. Even major scholarly publishers are clarifying that AI cannot be an author. Many are looking at the results of new, generative AI tools and suggesting this position or distinction might not be sustainable. Can you help us understand what it would take for the Copyright Office to change its position?
Jonathan: The requirement of human authorship is a basic principle of US copyright law. This is well established in the case law (think of the case involving the photograph snapped by Naruto the monkey) as well as Copyright Office rules. It is theoretically possible, but unlikely, that courts would overturn this precedent and start interpreting the phrase “original works of authorship” in 17 USC §102(a) to include works generated by computers. As a practical matter, congressional action would be needed to extend copyright protection to works generated by AI.
However, if the parameters provided by the human using the AI are sufficiently creative, it is conceivable that the human could convince the court that she is the author of the resulting work. Likewise, if the AI was sufficiently idiosyncratic that the works it produced reflected the creativity of the human AI developer, the developer might succeed in claiming authorship in the resulting work. But then computer scientists probably would not view the AI as real AI.
But I must push back a bit on the question. Why is not extending copyright protection to the results of AI “not sustainable”?
Judy: I think you’ve answered that question, actually, with the explanation of creativity of the AI parameters. I’ll leave it to computer scientists to adjudicate whether that’s real AI.
We are seeing generative AI services scrape the web for potentially copyrighted content to train its machine learning. Is this a fair use?
Jonathan: The case law in the US so far certainly indicates that this ingestion is a fair use: for example, Authors Guild v. Google and Authors Guild v. HathiTrust. The legal scholarship also treats the ingestion as a fair use. And the Israel Ministry of Justice recently issued an opinion finding that the use of copyrighted works for machine learning purposes was a fair use under the Israeli copyright law’s fair use provision, which is based on 17 USC §107. But several cases have been filed recently (in the US) that challenge this proposition.
Judy: Will the efficacy of “in the style of” AI change the thinking about whether such use is “transformative”? Is there any nuance to this in terms of highly creative and stylistically unique art, such as images, painting, some writing/text, and music?
Jonathan: The question of whether a particular work produced by AI is infringing is different from whether the ingestion of works in order to perform an AI process is infringing. The ingestion could be a fair use, but the output could be substantially similar in protected expression to one or more of the ingested works. One would then need to conduct a fair use analysis, and the conclusion could very well be that work is infringing. If I requested an AI program to create a new Harry Potter novel, it almost certainly would be infringing unless it had a parodic purpose.
Judy: Is generative AI an interesting use case for looking at the conflict between terms of service and what’s allowable under the Copyright Act—in other words, fair use?
Jonathan: Absolutely. The terms of service of a website might prohibit the scraping of the site by a bot for AI purposes, while fair use would allow such copying. The question would then be whether the terms of service would be enforceable—whether the operator of the bot could be found liable for violating the terms of service. The case law is far from clear on the answer to that question.
ARL’s Scholarship and Policy team will continue to inform the research library community about the application of copyright law to new technologies.
