The recent update on the Silverman v. OpenAI case has got people thinking about the end of the world. Okay, maybe not the end of the world, but it sure feels like it for OpenAI. They’re facing claims from Sarah Silverman, Paul Tremblay, and others, who are accusing OpenAI’s ChatGPT platform of copyright infringement. And let me tell you, OpenAI’s response is filled with catastrophe and references to Michael Jordan and the future of artificial intelligence.
Now, if you haven’t been following, let me catch you up. Silverman and Tremblay are among the folks who have filed lawsuits arguing that using their creative content to train big language models like ChatGPT violates the U.S. Copyright Act. OpenAI, on the other hand, is trying to dismiss most of the claims. They’re arguing that ChatGPT’s output isn’t “substantially similar” to the works of the plaintiffs, so there’s no infringement. But the plaintiffs are firing back, saying that OpenAI copied their stuff without permission and that the language models are derivative works that incorporate their content. And since ChatGPT’s answers rely on these language models, the plaintiffs claim that the output is infringing too.
OpenAI has responded with a doomsday scenario, saying that the future of artificial intelligence could hang on how the court decides this case. They argue that if the plaintiffs’ theory stands, every single answer given by ChatGPT would be considered infringement. So, for example, if you ask ChatGPT who the greatest basketball player of all time is, and it says “Michael Jordan,” that would supposedly infringe on Silverman’s book, even if MJ wasn’t mentioned in there. OpenAI thinks this approach is bonkers and wants the court to focus on the fact that there’s no substantial similarity between ChatGPT’s output and the plaintiffs’ works. They say it goes against copyright law and how courts evaluate these claims.
To drive their point home, OpenAI uses an illustrative example involving a film student studying Citizen Kane. The student watches the film, takes notes, and creates their own film as a tribute to Orson Welles. OpenAI argues that just because the student directly copied the original movie doesn’t mean their film is substantially similar to Citizen Kane. It’s an interesting analogy that showcases OpenAI’s perspective.
OpenAI also points out that the plaintiffs’ own claims don’t hold up. The ChatGPT outputs aren’t exact copies of the original books, and they often bear little resemblance to them. So, OpenAI argues, they can’t be infringing. They even mention that the plaintiffs’ included outputs in their complaints still contain the copyright information they claim OpenAI “removed.” Ouch, that’s a fatal blow to their claim under the Digital Millennium Copyright Act.
Ultimately, OpenAI thinks the plaintiffs are trying to fit their claims into a legal framework that just doesn’t support them. They believe that accepting these arguments would defy established legal precedent and sound the death knell for large language models. OpenAI insists that these platforms aren’t out to replace the original authors in the marketplace. They’re simply using publicly available material to gain a better understanding of human knowledge.
Now, we’ll have to wait and see how the court views all of this. There’s a motion hearing scheduled for December 7, 2023. So mark your calendars, folks, because this is going to be interesting.