Generative AI Under Scrutiny in Thomson Reuters vs. Ross Intelligence Spar
Generative AI Under Scrutiny in Thomson Reuters vs. Ross Intelligence Spar
A court in Delaware made a decision that could give a glimpse into how US Copyright Law will treat works created by Generative Artificial Intelligence (GAI). The Court left the most important copyright questions for the jury to decide, but it also gave some valuable insights that could affect the many lawsuits about GAI tools.
Thomson Reuters Enterprise Centre GmbH sued ROSS Intelligence Inc. for allegedly using the Westlaw platform without permission. The Westlaw platform is a digital database of judicial opinions. This lawsuit is about two key features of Westlaw: the Key Number System, which organises opinions by type of law, and the headnotes system, which annotates opinions with short summaries of points of law.
Ross Intelligence is a new company that develops legal research tools. Ross created a search engine that can answer legal questions by finding relevant quotes from court decisions. Because Ross uses machine learning and GAI, it needed legal materials to train its machine. Ross used some of the information from Westlaw’s Key Number System and headnotes, but the two companies disagreed about how Ross got this information.
Thomson Reuters argued that Ross copied Westlaw content when it trained its GAI legal research tool. Both sides asked the judge to decide the key issue in the case: whether Ross’s use of Thomson Reuters’s works was fair. The judge’s decision was released last month.
Judge Stephanos Bibas mostly refused to decide the parties’ motions for summary judgment. He found that there were too many unanswered questions about Thomson Reuters’s copyright claim for him to make a decision. Even after a lot of discovery, there were still genuine factual disputes about how broad and valid Westlaw’s copyright is, and how similar Westlaw’s headnotes are to the training materials that Ross used for its GAI machine.
Judge Bibas also decided that it was too early to rule on Ross’s fair use defence. He found that this issue should be decided by a jury. Courts use a four-factor test to decide if the use of a copyrighted work is fair: (i) The purpose and character of the use; (ii) The nature of the copyrighted work; (iii) The amount and substantiality of the portion used concerning the copyrighted work as a whole; and (iv) The effect of the use upon the potential market for the copyrighted work.
Judge Bibas considered all four factors in denying the parties’ fair use arguments.
The Court first considered the purpose and character of Ross’s use. It had to balance how commercial the use was with how transformative it was. Judge Bibas easily decided that Ross’s use was commercial since one of Ross’s goals was to compete with Westlaw.
The difficult question was whether Ross’s use of Westlaw’s content was transformative. Thomson Reuters argued that Westlaw is a legal research platform that organises the law, and Ross simply used Westlaw’s organisation to build its legal research platform. Ross, on the other hand, said that Westlaw’s headnotes and Key Numbers are just annotations and that it had transformed Westlaw’s headnotes by building a search engine that “avoids human-made materials.”
The Court said that the law on “intermediate copying” could be helpful in this case. In intermediate copying cases, users copy material as a small step towards making a completely different product.
Even though they use copied material as inputs, the final outputs are different enough to be considered fair use. But whether Ross’s use can be considered intermediate copying depends on exactly how Ross used Westlaw’s material, which is still in dispute.
Judge Bibas, therefore, observed that the jury should decide whether Ross’s use was transformative, but he offered some guidance: Ross’s use was transformative intermediate copying if Ross’s AI only studied the language patterns in the headnotes to learn how to produce quotes from court opinions. But if Thomson Reuters is right that Ross used the exact text of the headnotes to get its AI to copy and reproduce the creative work of Westlaw’s attorney-editors, then Ross’s comparisons to the intermediate copying cases are not relevant.
Next, Judge Bibas talked about the type of copyrighted material. He pointed out that the rules for fair use are broader when it comes to “informative” rather than “creative” content, using the example of Hustler Magazine v. Moral Majority Inc. Judge Bibas also suggested that Westlaw’s Key Number System and headnotes might not be considered “imaginative” enough to be seen as creative works, but he did not make a final decision on this. Instead, he stated that a jury should determine whether this aspect supports Thomson Reuters or Ross.
When it comes to how much Ross may have copied, Judge Bibas said that it’s a matter for a jury to decide. Sometimes, even if you copy just a little bit, it might not be fair use if you’re taking the most important and creative parts of the original work. On the other hand, even if you copy a lot, it might be fair use if what you copied isn’t the most creative part. Since there’s a disagreement about how Ross copied Westlaw’s stuff, Judge Bibas didn’t examine the extent of the copying.
Moreover, when it comes to the “substantiality” factor, it usually supports fair use if the amount copied was connected to a valid and transformative reason. However, the parties had different opinions about whether Ross’s copying had a valid purpose. Westlaw said that Ross copied too much, while Ross said it needed a wide range of materials to teach its AI. Once more, Judge Bibas believed that this was something a jury should determine.
Finally, Judge Bibas made a similar decision regarding the fourth factor of fair use: whether Ross’s use affects the potential market for Westlaw’s works. The Court initially looked at the potential market for Westlaw as a legal research platform. While Westlaw and Ross are rivals in this market, Judge Bibas pointed out that the main question is whether Ross’s GAI product can replace Westlaw. The idea of transformation is significant here: If Ross’s use was so transformative that it created a completely new research platform with a different purpose than Westlaw, then it wouldn’t be a market replacement. However, Judge Bibas, once again, thought that this question was not suitable for a quick decision due to the ongoing dispute about how Ross used Westlaw’s data.
Judge Bibas wrapped up his fair-use examination with a question that gets to the heart of the copyright and AI debate: Is it a good thing for the public if AI can be educated using copyrighted content? If the answer is yes, then the fourth fair-use factor might support fair use. Without giving a definite answer to this critical question, Judge Bibas provided some guiding questions for a jury dealing with GAI technology: How much does it change the content? Can the public use it without cost? Does it prevent other creators from succeeding by taking over their markets? The way a jury responds to these questions could shape the future of GAI.
