“They would only get into trouble if they summed up the story incorrectly and made it defamatory, although that had not happened before. “It’s something they would actually be at legal risk for, especially if they don’t make the original source clear enough and people can’t easily go to that source to check it out,” he says. “If Perplexity’s changes make a story defamatory, 230 does not cover it as interpreted by that case law.”
In one instance observed by WIRED, the Perplexity chatbot falsely claimed, although clearly linking to the original source, that WIRED had reported that a specific police officer in California had committed a crime. (“We have been upfront that the answers will not be 100% accurate and may cause hallucinations,” Srinivas said in response to questions for an article we published earlier this week, “but a central aspect of our mission is to continually improve accuracy and user experience. “)
“If you want to be formal,” Grimmelmann says, “I think this is a set of propositions that could pass muster with a motion to reject a number of theories. “I’m not saying that he will win in the end, but if the facts support what Forbes and WIRED claim, the cop – a group of potential plaintiffs – these are things that, if proven and other facts that are unfavorable to Perplexity, could lead to liability.”
Not all experts agree with Grimmelmann. Pam Samuelson, a professor of law and information at the University of California, Berkeley, writes in an email that copyright infringement is “the use of another person’s expression in a way that undermines the author’s ability to obtain adequate compensation for the value of the unauthorized use. One sentence literally probably does not constitute an infringement.”
Bhamati Viswanathan, a research associate at New England Law, says she is skeptical that the summary crosses the threshold of substantial similarity that is typically necessary for a successful infringement claim, although she doesn’t think that’s the end of the matter. “He certainly shouldn’t pass the sniff test,” she wrote in an email. “I would believe that this should be enough to get your case past the threshold for a motion to dismiss – especially given all the evidence you have of copying actual material.”
Overall, however, he argues that focusing on the narrow technical merits of such claims may not be the right way to think, as tech companies may adapt their practices to respect outdated copyright laws while grossly defeating their purpose. She believes an entirely new legal framework may be necessary to correct market distortions and promote the core goals of US intellectual property law, including allowing people to financially benefit from original work, such as journalism, in order to encourage them to produce it – which theoretically benefits society.
“In my opinion, there is a strong case for the intuition that generative AI relies on large-scale copyright infringement,” he writes. “The initial question ante is: where are we going? And the more important long-term question is: how to ensure the survival of creators and creative economies? Ironically, artificial intelligence is teaching us that creativity is more valuable and desirable than ever. But even as we recognize this, we see the potential to weaken and ultimately eviscerate the ecosystems that enable creators to earn a living from their work. This is a puzzle we must solve—not finally, but now.”