Generative AI has taken the world by storm since OpenAI launched ChatGPT in November 2022. But the buzz and excitement of GAI has come with difficult legal questions that threaten the new technology. Several lawsuits—some of which we have discussed in detail—have been filed against companies whose GAI products have been trained on copyrighted materials. Up until now, we have only been able to speculate how courts will handle GAI as the industry has held its collective breath.
Last month, the District of Delaware entered a summary judgment ruling that provides a potential glimpse into how generative artificial intelligence (“GAI”) will fit into U.S. copyright law. Although the decision left the key copyright issues for the jury to decide, it leaves us with valuable insight that could tip the scales within the growing list of lawsuits concerning the use and development of GAI tools.
Thomson Reuters Enterprise Centre GmbH v. ROSS Intelligence Inc., centers on the alleged unauthorized use of Thomson Reuters’s Westlaw platform. As legal practitioners know well, the Westlaw platform offers a database of judicial opinions in a digital format. Two key features are the subject of this litigation. The first is Westlaw’s “Key Number System,” which organizes opinions by the type of law. And the other is the “headnotes” system, which Westlaw uses to annotate opinions with short summaries of points of law that appear in the opinions.
Defendant Ross Intelligence is a legal-research start-up company. Ross created a natural-language search engine that would answer legal questions by returning relevant quotations from judicial opinions. Because the tool leverages machine learning and GAI, Ross sought out legal material to train the machine. Some of this material came from Westlaw’s Key Note System and headnotes, but the method by which Ross fed Westlaw’s information to its machine is disputed by both parties.
On summary judgment, Thomson Reuters argued that Ross infringed on copyright-protected Westlaw content in training Ross’s GAI legal-research tool. Additionally, both sides moved for summary judgment on the hot-button issue at the center of this case: whether Ross’s use of Thomson Reuters’s works constituted fair use. These motions led to the decision handed down last month.
Judge Stephanos Bibas mostly denied the parties’ motions on summary judgment. As to Thomson Reuters’s infringement claim, Judge Bibas found that too much remained in dispute for him to conclude one way or another. After extensive discovery, there remained genuine factual disputes as to the breadth and validity of Westlaw’s copyright and the extent of similarity between Westlaw’s headnotes and Ross’s training materials for its GAI machine.
Judge Bibas also found it inappropriate to rule on Ross’s fair-use defense at this stage, instead finding that the issue should go to a jury. As a reminder, courts use a four-factor test to determine if use of a copyrighted work is protected by the fair-use doctrine. These factors are: (1) the purpose and character of the use; (2) the nature of the copyrighted work; (3) the amount and substantiality of the portion used in relation to the copyrighted work as a whole; and (4) the effect of the use upon the potential market for the copyrighted work. Judge Bibas addressed the four factors in turn in denying the parties’ fair-use arguments.
The Court started with the purpose and character of Ross’s use, noting the need to balance the commerciality of the use with its transformative effect. Judge Bibas had no issue deciding that Ross’s use was commercial—indeed, one of Ross’s goals was to compete with Westlaw.
The difficult question, of course, was how transformative Ross’s use was. Thomson Reuters argued that Westlaw is a legal-research platform that synthesizes the law, and Ross simply used Westlaw’s syntheses to build its own legal-research platform that also synthesizes the law. Ross, on the other hand, described Westlaw’s headnotes and Key Numbers as annotations while arguing that it had transformed Westlaw’s headnotes by building a search engine that “avoids human intermediated materials.” Thomson Reuters, at 17.
The Court noted that the caselaw on “intermediate copying” could be useful here. In those cases, the users copied material as a minor step towards developing a completely different product. Despite using copied material as inputs, the final outputs were transformative enough to constitute fair use. See, e.g., Sega Enterprises Ltd. V. Accolade, Inc. But whether Ross’s use could be considered intermediate copying depends on the precise method by which Ross used Westlaw’s material—a crucial piece of information still in dispute at the summary judgment stage. Judge Bibas thus concluded that the transformativeness of Ross’s use should be decided at trial but offered guidance for the jury: Ross’s use “was transformative intermediate copying if Ross’s AI only studied the language patterns in the headnotes to learn how to produce judicial opinion quotes. But if Thomson Reuters is right that Ross used the untransformed text of headnotes to get its AI to replicate and reproduce the creative drafting done by Westlaw’s attorney-editors, then Ross’s comparisons to [the “intermediate copying” caselaw] are not apt.” Thomson Reuters, at 19-20.
Judge Bibas then discussed the nature of the copyrighted work. He noted that the scope of fair use is greater when “informational,” as opposed to “creative” works are involved, citing Hustler Magazine v. Moral Majority Inc. Judge Bibas went on to hint that Westlaw’s Key Number System and headnotes may not be “imaginative” enough to constitute creative works but stopped short of making this decision. Instead, he concluded that a jury must decide whether this factor weighs in favor of Thomson Reuters or Ross.
As to the amount and substantiality of Ross’s alleged copying, Judge Bibas also found that the issue required jury fact-finding. Small amounts of copying may fall outside the scope of fair use where the excerpts capture the “heart of the original work’s creative expression.” Google LLC v. Oracle Am., Inc. But large amounts of copying can still be fair use “where the material copied captures little of the material’s creative expression.” Id. Because the method by which Ross copied Westlaw’s works remains in dispute, Judge Bibas declined to analyze the scope of copying.
Additionally, “[t]he ‘substantiality’ factor will generally weigh in favor of fair use where . . . the amount of copying was tethered to a valid, and transformative, purpose.” Id. But the parties disagreed as to the validity of the purpose of Ross’s copying. Westlaw argued that Ross copied excessively; Ross contended that it needed a wide breadth of materials to train its AI. Again, Judge Bibas found this issue to be more appropriate for a jury to decide.
Lastly, Judge Bibas concluded similarly on the fourth fair-use factor: a jury must determine the effect of Ross’s use upon the potential market for Westlaw’s works. The Court first considered the potential market for Westlaw itself as a legal research platform. Although Westlaw and Ross do compete against each other in this market, Judge Bibas noted that the key question is whether Ross’s GAI product is a substitute for Westlaw. Transformativeness loomed large here: If Ross’s use was so transformative that it created “a brand-new research platform that serves a different purpose than Westlaw,” its product is not a market substitute. Thomson Reuters, at 25. But, again, Judge Bibas found this question inappropriate for summary judgment given the unsettled dispute over how Ross used Westlaw’s data.
Judge Bibas ended his fair-use analysis with a question that hits at the core of the GAI-copyright zeitgeist: “Is it in the public benefit to allow AI to be trained with copyrighted material?” Id. If so, the fourth fair-use factor may weigh in favor of fair use. Without answering his million-dollar question, Judge Bibas offered some guiding questions to be left to a jury grappling with GAI technology: “How transformative is it? Can the public use it for free? Does it discourage other creators by swallowing up their markets?” Id. at 25-26. How a jury answers these questions could determine the very future of GAI.
With this Thomson Reuters opinion—the first summary judgment decision discussing fair use in the context of GAI—we have a glimpse into what is in store. Transformativeness factored greatly in Judge Bibas’s opinion and will certainly play an outsized role in the outcome of the other GAI litigations. The perceived public benefit of GAI will also factor heavily in courts’ analyses (as it surely will in any GAI-related legislative efforts). The future awaits.