In rejecting an AI company's fair use defense for using Thomson Reuters' Westlaw headnotes to train its competing legal tool, Judge Bilas, the District of Delaware judge in Thomson Reuters Enterprise Centre GMBH and West Publishing Corp. v. Ross Intelligence Inc., Case No. 1:20-cv-613-SB, was careful to emphasize that the ruling was not generally applicable to generative AI as the platform at issue did not create new creative content and instead used the training materials to enable it to answer questions.
While such a distinction could arguably be used to distinguish the current litigation involving generative AI systems' use of existing copyrighted music as training materials and then create wholly new works, the parties in that matter have nonetheless moved to introduce the ruling as supplemental authority.
Outside of the creative space, the ruling could be seen as more applicable to LLMs that answer questions from users (rather than generating creative works). In Thomson, Judge Bilbas noted that the fact that the defendant's system was intended to compete with that of the plaintiff meant that it was not transformative and cut strongly against fair use. More general LLMs are less likely to be in such direct competition with the sources of their training data. But the ruling also implies that to the extent there is a market for the licensing of such training data, it could be difficult to establish a fair use defense. The publishers in the generative AI litigation, too, are focusing on this aspect of the ruling.
Thus, while on its surface the Thomson ruling makes efforts to avoid broad proclamations about whether the use of training data for AI systems amounts to fair use, it may have nonetheless done exactly that. The coming decisions on the subject will be watched closely for how they interpret, embrace, or reject this early ruling.
he wasn't issuing a decision that people would take as a general proclamation against AI
www.law360.com/...
