For creative individuals and content creators, concerns over rights to their work, especially in the context of the evolving generative artificial intelligence (AI) industry, have recently been brought into sharp focus. A court decision delivered in February offers some insights into the type of evidence they need to present in order to successfully challenge copyright infringement.
In the legal battle, a group of authors, one of the most notable being comedian Sarah Silverman, were up against OpenAI, the creators of ChatGPT, a well-known chatbot. The case revolved around the claim of vicarious copyright infringement. However, the ruling on February 12, issued by Judge Araceli Martínez-Olguín, dismissed this claim. This dismissal has crucial implications for potential future claims—it signals that future plaintiffs must examine the final product created by the bot, rather than just the act of copying itself.
Part of the lawsuit hinged on whether there was substantial similarity between Silverman and other plaintiffs’ original work, and what was created by OpenAI’s ChatGPT. While there was evidence that OpenAI had certainly copied from the plaintiffs’ works, attention was directed to the final outputs produced by ChatGPT, which were the alleged infringements.
The plaintiffs argued that ChatGPT’s outputs were derivative works, thus constituting infringement, but did not provide sufficient allegations demonstrating substantial similarity. In their original complaint, they declared that OpenAI directly copied their books into its AI model. However, they never claimed that the AI outputs were identical copies of any sections from their original books.
This case pencils out a fundamental principle of copyright law; it protects the exact expression of an idea, not the idea itself. And in the realm of AI and copyright, this principle stands to be a decisive factor. It points to the understanding that any new art is essentially a build on previous works.
ChatGPT creating summaries of the original works maxes out to the same legal position as when a biographer is guided and informed by previous authors’ works while etching a biography. Reading previous work isn’t infringement, as far as the new work is original in its expression.
Plaintiffs such as Silverman must furnish evidence showing that the AI tool outputs contain direct copies from their work. Such a condition may pose considerable challenges as the exact details of the evidence that would meet these requirements remains to be seen.
However, this is not across-the-board advice. There are cases in the legal pipeline that seem to be in a more advantageous position. This is particularly so when the AI tool shows near verbatim reproduction of original content. For instance, The New York Times and a group of music publishers have filed similar suits alleging direct copying in outputs.
As the legal battles involving AI and copyright law continue, potential plaintiffs need to review whether and how their work has been copied, and be ready to show substantial similarities or direct copying.
The legal citation for the case is Silverman et al v. Open AI, N.D. Cal., 3:23-cv-03416, decided on 2/12/24.