IP Law Daily, COPYRIGHT NEWS: Predictability is low and ‘nobody knows anything’: A new report from the Copyright Office offers some frank talk about AI, (Feb 13, 2025)

By Matthew Hersh, J.D.

More empirical study is needed to determine whether generative artificial intelligence will be a boon or bust for human creators.

Artificial intelligence may reduce the production costs to consumers of creative works but “the effect on human creators is less predictable without sufficient information about the demand curve and cost function for creative works,” a new report by the Copyright Office has found. The report, written by a group of economist at the behest of the Office, notes that while AI may replace some human creators, the resulting lower price for creative works might well increase demand for those works and “thus increase demand in the labor market for creators, all else constant.”

The heavily academic treatise—rare for a Copyright Office that prides itself on user-friendly publications for a non-specialized public—reflects the work of an ad hoc committee of economic scholars convened by the Office to discuss “the numerous economic issues at the intersection of artificial intelligence and copyright policy.” The committee engaged in several months of substantive discussions, consultation with technical experts, and research, culminating in a daylong roundtable event on January 23, 2024. Brent Lutes, Copyright Office Chief Economist, edited the report with the support of economists from Harvard Business School, Carnegie Mellon University, the Massachusetts Institute of Technology, and several other institutions.

The report, titled Identifying the Implications of Artificial Intelligence on Copyright Policy: Context and Direct for Economic Research, takes on a wide variety high profile issues in the intersection between copyright law and artificial intelligence. It begins with one of the hottest topics, namely the copyrightability of AI-generated works. Whether it is “economically efficient” to extend copyright protection to AI-generated works, the report concludes, depends on the “net value proposition” of those works. If AI-generated works cause a net reduction in value, the report noted, it is not efficient to further incentivize their production through copyright protection. In contrast, a net positive value “may make it efficient to extend copyright protection to AI-generated works.” For example, where AI offers services or products that are creative or innovative in and of themselves—or that aid in the creativity or innovation of others—they may offer a net positive value to society. On the other hand, when generative AI platforms produce near-verbatim copies of existing work, “those works are no more valuable than the original—except possibly where the copies enhance public access to existing works.”

The report also takes on the equally high-profile top of the extent to which AI-generated works serve as viable substitutes for human-generated works. The substitution effect is “easy to conceptualize in the case of AI producing near-verbatim copies,” the court notes. Similarly, to the extent that creators can buy or commission works from others—the report analogizes to recording artists purchasing musical compositions from songwriters—an AI model may have substitutive effects if it “can produce songs of sufficiently high quality for substantially cheaper than a human songwriter.”

On the other hand, the report noted, not all uses would supplant revenue for human creators. “Some uses will reduce deadweight loss, replacing it with consumer surplus by allowing for additional consumption that otherwise would not occur,” the report notes. Similarly, the aspiring recording artist who may not be able to license a high-quality musical work from a human songwriter “may, with the help of generative AI, be able to produce a record that would not have otherwise been feasible.”

Ultimately, the report concluded, “empirical research is needed in order to better understand the relative magnitude of the value created and the value displaced by generative AI.” Moreover, the report acknowledges, “predictability is typically low—an understanding codified, for example, in the movie industry, by the expression ‘nobody knows anything.’”

Other chapters of the 66-page report take on a wide variety of issues in the AI space, including the economic impact of copyright infringement by AI-generated output; the impact on creative incentives in the area of name, image, or likeness rights; the effects of AI ingestion on copyright holders’ incentives; the economic questions surrounding developers’ access to training data (and the control of the use of copyright material in that training); and potential socioeconomic biases of AI policy.

The Copyright Office added a perhaps-unsurprising disclaimer to the report, emphasizing that it does not take a position on the ideas in the report and that “the inclusion of any particular idea merely indicates that the idea was discussed by roundtable participants and should not be taken as an endorsement of that idea or opinion by the Copyright Office, the principal contributors of a particular section, or any individual participants.”

MainStory: TopStory AINews Copyright IndustryNewsTrends TechnologyInternet GCNNews