Report from TechSpot
In Brief – Federal Judge Stephanos Bibas has delivered a significant ruling in a copyright case pitting Thomson Reuters against the now-defunct legal services startup Ross Intelligence that claimed to have developed an AI-enabled legal service. Judge Bibas ruled that Ross’s system was developed using thousands of Thomson Reuters’ Westlaw case summaries without paying licensing fees, and that copies of those summaries were provided to Ross’s users. Of note, the judge took pains to point out that, “Ross’s AI is not generative AI (AI that writes new content itself). Rather, when a user enters a legal question, Ross spits back relevant judicial opinions that have already been written.” And therefore, he cautioned that this summary judgement ruling, which rejected Ross’s fair use copyright defense, was about non-generative AI. Bibas said that Ross’s fair use defense failed on two prongs of fair use analysis, namely that Ross’s service was a commercial venture that was not truly transformative, and that Ross’s service competed with Westlaw in the market and harmed Westlaw’s value. He said that the Supreme Court’s 2023 ruling in Andy Warhol Foundation v. Goldsmith guided his fair use determination.
Context – The fact that Ross’s service was inarguably not generative AI limits the value of the ruling on the huge questions around the legality of “training” the neural networks of major GAI models with non-licensed copyrighted material. In the EU, with its AI Act, regulators and AI expert groups will play key roles. In the US, copyright lawsuits targeting GAI giants will likely focus on fair use. Federal Judge William Orrick, overseeing cases involving image generating services trained on digital artworks, recently issued a ruling in which he explained that he is trying to ascertain how the GAI systems work. He will learn that they are not databases like Ross’s system. They do not store or retrieve copies. They “learn” from data and then produce new output. It will be interesting to see how courts react when GAI operators admit they don’t exactly know why their systems produce any particular output, hence the nagging existence of our favorite GAI concept, “hallucinations”.