Thomson Reuters Wins Initial Legal Battle Over Artificial Intelligence, Copyright, and Fair Use.
Are AI tools legal?
On Tuesday, Judge Stephanos Bibas of the U.S. District Court in Delaware issued a partial summary judgment in favor of Thomson Reuters in its copyright infringement lawsuit against Ross Intelligence, a legal artificial intelligence startup. This lawsuit, filed in 2020, is one of the first to address the legality of AI tools and the way they are trained, often using copyrighted data obtained without a license or permission.
Similar lawsuits against companies like OpenAI and Microsoft are developing in the courts and could raise similar questions about whether these AI tools can invoke a "fair use" defense for protected material.
In a statement provided by a spokesperson for Thomson Reuters, Jeff McCoy, the company expressed its satisfaction with the court's decision, asserting that the editorial content of Westlaw, created and maintained by its editors, is protected by copyright and cannot be used without their consent. The company emphasized that copying its content did not constitute "fair use."
It is noteworthy that the judge mentioned this case involved "non-generative" artificial intelligence, in contrast to generative tools like language models. Ross Intelligence shut down in 2021, calling the lawsuit "frivolous," but acknowledging it could not secure enough funding to continue amid the legal battle.
In his ruling, Judge Bibas argued that "none of Ross's potential defenses holds up" against the copyright infringement allegations, rejecting Ross’s fair use defense by considering how the use of protected material affected the market value of the original work, given that it built a direct competitor. Thomson Reuters sued Ross for its use of the Westlaw search engine, which indexes a substantial amount of unprotected material (such as court decisions), but also intermixes it with its own content. For example, point-of-law summaries, created by human editors, are a distinguishing feature that makes the costly Westlaw subscription attractive to attorneys.
By developing a legal search engine, Ross transformed annotations and summaries into numerical data about the relationships between legal terms to feed its artificial intelligence. After its attempt to license Westlaw content was rejected, Ross approached another company, LegalEase, and purchased 25,000 Collective Memory question-and-answer briefs drafted by attorneys using Westlaw summaries, which it then used as training data.
Ross’s CEO, Andrew Arruda, claimed that Westlaw's data was "additional noise" and that his tool "seeks to recognize and extract answers directly from the law through machine learning." However, after analyzing how the 2,830 Collective Memory questions, summaries, and judicial opinions were related, the judge concluded that the evidence of copying was "so obvious that no reasonable jury could consider otherwise."
