The Universal Lexicon: Generative AI Chatbots and Fair Use of Works in a Training Corpus

For centuries, research publications, encyclopedias, and lexicons have helped advance knowledge by drawing freely on expressive works, without seeking licenses for most of what they reference or “use.” Today, that longstanding practice is under new and intense scrutiny. One pressing question is whether chatbots powered artificial intelligence (AI) should be treated differently. Major copyright holders, including The New York Times, leading authors, and powerful music rights organizations, have filed lawsuits against companies such as OpenAI, Anthropic, and Google. They argue that the training and output of AI systems infringe the exclusive rights of copyright holders, principally the rights of reproduction, distribution, and preparation of derivative works. The Supreme Court’s decision in Andy Warhol Foundation for the Visual Arts v. Goldsmith has raised the stakes in this copyright controversy by narrowing the meaning of transformative fair use and casting fresh doubt on whether copyright law still permits the creation of generative AI (gen-AI) models as modern lexicons of knowledge. This article examines prior copyright claims against lexicons, scholarly works, and directories, among other technologies and creations, and explains why the right to release a universal lexicon of culture and thought must be defended. At stake is not simply a technical question of doctrine, but the future of tools that help society organize and expand human knowledge. First, the purpose of a lexicon or grand encyclopedia differs fundamentally from that of novels, songs, or other creative works, or even individual news articles, press releases, or webpages. Courts have long recognized that reference works, like news reporting and scholarship, serve purposes that fair use protects, precisely because they promote the progress of knowledge rather than mere entertainment. Second, the amount and substantiality of what is used in AI outputs can often be negligible. Any fragment that may appear in output is generally tiny, both in absolute terms and as a proportion of the original work, and rarely, if ever, recreates the expressive heart of that work. Third, efforts to prohibit research-based creativity by invoking hypothetical licensing markets threaten to dismantle fair use itself. They invite a circular reasoning process in which any unpaid use is treated as illegal harm simply because it could have been licensed. Some courts have rightly rejected this logic and have insisted that fair use requires rightsholders to demonstrate a real and direct connection between a specific use and actual loss, whether in sales, profits, or other market value. Mere claims that a license might have been sold are not enough. What follows is an analysis of why generative AI models fit within the longstanding protections of fair use, and how courts should apply similarity-related principles of fact, genre, and research to defend the creation of these modern lexicons.