On February 2, US District Court of Delaware Judge Stephanos Bibas handed down a partial summary judgement in favor of Thomson Reuters, in its copyright infringement suit against Ross Intelligence, a legal AI start-up. It’s one the first cases to deal with the legality and training of AI tools, which often use copyrighted data that is scraped without permission or license.
Similar suits against OpenAI, Microsoft and other AI giants could be pending in the courts and could have similar questions regarding whether or not AI tools can claim “fair use” of copyrighted materials. In a statement to the Verge given by Thomson Reuters spokesperson Jeff McCoy the company said: The copying of Westlaw’s content was not considered “fair use.”
But, as the judge pointed out, this case concerned “non-generative AI” and not a generative AI like an LLM. Ross shut down its operations in 2021. It called the lawsuit “spurious”but said it could not raise enough money to continue operating while involved in a legal fight.
As previously reported byWired, today Judge Bibas wrotein his decision that “None[of Ross’s]possible defenses hold water” against accusations for copyright infringement. He ultimately rejected Ross’s defense of fair-use, relying heavily upon the factor of how Ross’s use of copyrighted materials affected the market value of the original work by building a competitor. Thomson Reuters filed a lawsuit against Ross for his use of the Westlaw search engine. Westlaw indexes material that isn’t copyrightable, like legal decisions, but also mixes it with its own content. Westlaw headnotes, which are summaries written by human editors of legal points, are a signature feature designed to make the expensive Westlaw subscription appealing to lawyers.
Ross used the annotations and the headnotes to build a legal search engine. “The data was then converted into numerical data about the relationships between legal words, which it fed into its AI,” Bibas wrote. The ruling explains how, after Thomson Reuters refused to license Westlaw content, Ross turned around and bought 25,000 Bulk Memos containing questions and answers from lawyers using Westlaw headnotes. This data was used to train the AI.
Ross’ CEO Andrew Arruda claimed that Westlaw data was merely “added noise,” and that the tool “aims to extract answers directly from law using machine-learning.”
www.aiobserver.co
