The article examines the legal complexities surrounding the use of copyrighted books to train AI language models, highlighting that while the practice may seem illicit, current copyright law does not explicitly prohibit it. Attorney Cathy Gellis emphasizes the intricate nature of the issue, noting that copyright law centers on the act of copying rather than merely using or experiencing a work. This perspective is reinforced by Judge William Alsup’s ruling in a case against Anthropic, where he determined that training AI models on copyrighted texts is lawful, penalizing the company instead for obtaining the books from illegal online repositories.
The discussion further delves into the concept of fair use, particularly the transformative use of copyrighted material, which allows for its utilization without permission under certain conditions. Jason Henderson, an IP attorney, points out that courts are divided on how to apply fair use principles to AI training, often focusing on whether the use competes with the original work. The article references a case where Thomson Reuters successfully argued that Ross Intelligence’s use of its content to develop a competing AI platform was not transformative, thus not protected under fair use. Despite these legal precedents, the relationship between AI training and copyright remains unsettled, with ongoing litigation likely to shape future interpretations and applications of the law. Source