“They would only face repercussions if they compressed the narrative inaccurately and turned it into libel without cause. This is a scenario where they could genuinely face legal consequences, especially if they fail to attribute the original source adequately, making it difficult for individuals to verify,” he remarks. “If Perplexity’s modifications result in the story becoming defamatory, Section 230 does not protect against that, according to various legal precedents interpreting it.”
In a scenario witnessed by WIRED, Perplexity’s chatbot did incorrectly assert, although prominently referencing the original source, that WIRED had stated a specific California police officer had committed a crime. (“We have been transparent that responses won’t always be accurate and may deviate from reality,” Srinivas commented in response to queries for the recent story, “but a key aspect of our mission is to enhance accuracy and enhance the user experience.”)
“If we want to be formal,” Grimmelmann remarks, “I believe this presents a series of assertions that could withstand a motion to dismiss based on several legal theories. I’m not saying it will prevail eventually, but if the allegations from Forbes and WIRED, as well as the police officer – among various possible plaintiffs – are substantiated by the facts, they are the sort of claims that, if proven and other circumstances turn out unfavorably for Perplexity, could result in accountability.”
Not every expert shares Grimmelmann’s viewpoint. Pam Samuelson, a law and information professor at UC Berkeley, explains in an email that copyright violation pertains to “the utilization of someone else’s expression in a manner that diminishes the author’s capacity to receive proper compensation for the worth of the unauthorized use. One verbatim sentence is unlikely to constitute a violation.”
Bhamati Viswanathan, a faculty fellow at New England Law, expresses doubt if the summary surpasses the threshold of significant resemblance often essential for a successful infringement claim, although she does not view that as the conclusive factor. “It should definitely not meet the smell test,” she conveyed in an email. “I would argue that it should suffice to progress your case beyond the motion to dismiss threshold – particularly considering the compelling signs of actual content being replicated.”
Overall, she contends that concentrating solely on the specific technical merits of such assertions may not be the most suitable approach, as technology firms can amend their methodologies to comply with the wording of outdated copyright legislations while still blatantly breaching their intent. She believes a completely new legal structure may be required to rectify market distortions and uphold the fundamental objectives of US intellectual property legislation, including allowing individuals to reap financial rewards from original creative endeavors such as journalism in order to motivate them to create more – potentially benefiting society in theory.
“In my opinion, there are strong arguments supporting the belief that generative AI is based on extensive copyright infringements,” she states. “The crucial question at hand is, where do we proceed from here? And the broader issue in the long term is, how do we guarantee that creators and creative economies endure? Ironically, AI is educating us on the heightened value and demand for creativity. Nevertheless, even as we acknowledge this, we perceive the threat of undercutting, and ultimately eradicating, the ecosystems that support creators in earning a livelihood from their creations. This is the dilemma we need to resolve – not in the distant future, but immediately.”
