Summer Associate Bridget Moore also contributed to this article.
As companies—and more recently, courts—have struggled to address the role of artificial intelligence (AI) in innovation, legislators are embroiled in a struggle of their own. Over the past two years, the Senate and House have held public hearings to address how, if at all, AI should be regulated and to what extent IP rights should inhere in AI-assisted inventions and creative works. Most recently, in April 2024, the House Judiciary Subcommittee on Courts, Intellectual Property, and the Internet held its third in a series of hearings addressing those very questions. Unsurprisingly, the witnesses and legislators expressed varying views.
This article summarizes those views and sets forth five key takeaways relevant for industry players who currently use AI to drive innovation or who are contemplating doing so.
- New technology, old problems: our IP laws have always had to evolve to accommodate new technologies.
All seemed to agree that, although AI is groundbreaking technology, the problems that legislators now face are not unique to that technology. As Representative Jerry Nadler (D-NY) emphasized, AI is simply the most recent development in a centuries-long philosophical inquiry into the essence of human creativity, stemming all the way back to Descartes’ first principle: “I think, therefore I am.” The question for Congress, then, is whether existing IP law can effectively govern this new technology.
The answer from the witnesses seemed to be “yes.” For example, Sandra Aistars, Clinical Professor at the George Mason University Antonin Scalia Law School, argued that the Supreme Court’s test for copyright originality set forth in in Feist Publications, Inc. v. Rural Telephone Service Co., 499 U.S. 340 (1991), can easily be applied to AI and reflects an appropriately technology-neutral view of the law. That test asks whether a human author contributes a “minimal amount of creativity” to a work. If so, the author should be awarded a copyright. Any lesser degree of protection, Professor Aistars maintained, would “relegate” creative human input “to the category of synthetic data.”
Echoing these views, Joshua Landau, Senior Counsel for the Innovation Policy, Computer and Communications Industry Association, noted that the authorship questions facing Congress today are analogous to those raised decades earlier with the advent of computer technology. In that context, the Copyright Office provided guidance that applies with equal force to AI. The key inquiry is “whether the ‘work’ is basically one of human authorship, with the computer merely being an assisting instrument, or whether the traditional elements of authorship in the work were actually conceived and executed not by man but by a machine.”
- The subcommittee expressed the desire to be “forward leaning” with regard to granting IP protection for AI-assisted works.
The subcommittee expressed a strong desire to be “forward leaning” in granting IP protection to AI-assisted inventions and creations. Some legislators, moreover, feel that it is important for Congress, in particular, to do so. Ranking Member Hank Johnson (D-GA), for example, said that he felt a “little queasy” about ceding legislative authority to the courts or agencies to determine how the law should apply to AI.
Others expressed the view that strong legislation granting IP protection for AI-assisted innovation is critical for the United States to compete with other countries. Chairman Darrell Issa (R-CA), for example, asserted that a failure to extend such protection would threaten to dismantle the system that has “made the United States the most innovative and successful country in the history of mankind.” The legislators agreed, moreover, that strong protections were needed to combat China’s IP theft. With that said, Mr. Landau raised concerns about potential policy implications of making AI outputs eligible for protection. As he noted, China has already flooded the United States with dubious trademark applications, and more than half of U.S. patents are issued to foreign inventors, providing an avenue for foreign actors to “weapon[ize] patents against the American economy.”
- Legislators are stumped about where to draw the line between AI-assisted innovation worthy of protection and AI-created innovation that ought not be protected.
The question that eluded both legislators and witnesses alike is where to draw the line between AI-assisted innovation and AI-created innovation. Or, as Chairman Issa framed it: “[h]ow far can we go in protecting creations by man but assisted by machine?”
The witnesses expressed divergent views on this issue. For example, Kristelia Garcia, Professor of Law at Georgetown University, endorsed the view reflected in current Copyright Office guidance that works wholly or substantially generated by AI do not merit copyright protection. Thus, while a human-authored prompt ingested into an AI without further control or interaction would fall short of human authorship, repeatedly altering and editing the output of the AI could be viewed as the product of the human’s own creativity and intellectual conception. Rewarding this process with copyright protection would align with the policy goals of incentivizing creativity.
Professor Aistars, by contrast, criticized the Copyright Office’s guidance. In her view, the guidance inappropriately focuses on the user’s control over the AI even though the user had no role in programming or training it. Professor Aistars thus proposed a subjective test that asks whether the output of the AI stays true to the user’s authentic artistic voice. If so, copyright protection may be appropriate.
As for patents, Mr. Landau proposed simply extending the Copyright Office’s authorship inquiry to issues of inventorship. When a machine simply supports the process of human invention, patent protection should remain available. In contrast, Ms. Claire Laporte of Ginkgo Bioworks, Inc., suggested that an inquiry into the level of AI involvement is inappropriate, particularly given the Patent Act’s mandate that “patentability shall not be negated by the manner in which the invention was made.”
- At least some legislators believe that liability for training AI models on copyrighted works should go hand-in-hand with copyrightability for the output of such models.
Some legislators questioned the witnesses about the interplay between the fair-use doctrine for training AI and copyrightability of AI-generated output. For example, Chairman Issa expressed concern that, if fair use does not allow AI models to be trained on copyrighted works and copyright is not granted on the output of AI, then AI developers would not have a revenue stream to pay royalties to copyright owners whose works are used for training. In his view, fair use at the training stage and copyrightability at the output stage go hand-in-hand: if fair use applies, then the output of AI should be copyrightable; if the output is not copyrightable, fair use should not apply.
- The witnesses all agree that the legislature should not take action at this time … except perhaps to undo the PTO’s and Copyright Office’s guidance addressing AI.
Finally, even though the witnesses disagreed about the appropriate tests for determining inventorship and authorship, they were unanimous in their pleas for Congress not to rush legislation. Professor Garcia urged Congress to wait and see how courts resolve these issues, thus reducing the risk that legislation spawns unintended (and undesirable) consequences.
Ms. Laporte echoed similar sentiments in the patent context. Patent litigation and prosecution are already complicated enough, she noted, and do not need the additional complexity and confusion that premature legislation would create. With that said, Ms. Laporte urged the Committee to undo the PTO’s guidance addressing AI. In her view, the guidance places too much weight on AI tools used during the innovation process. She noted that, at least in the biotech space, the patent laws should be concerned with the patentability of the end product, not on the tools used en route to that product. She also noted that the PTO guidance does not define AI and thus could be read to apply broadly to any computer modeling tool that researchers routinely use. Professor Aistars, as noted above, expressed similar views as to the Copyright Office’s guidance.
In the end, while legislators seem hesitant to take the “wait-and-see approach” advanced by the witnesses, all agree that the United States should take steps to protect its competitive edge in the global economy.
Sterne Kessler is following developments in artificial intelligence and IP.
