Three authors have sued OpenAI, creator of the generative artificial intelligence (AI) interface ChatGPT, whom they accuse of having used the content of their books to power the conversational robot, we learned on Monday from judicial source.
• Read also: Artificial intelligence: the UN faced with the awakening of robots
• Read also: Artificial Intelligence: Humanoid robots say they can rule the world
• Read also: A conversational AI would have “encouraged” a man who wanted to kill the queen
They also filed a lawsuit against Meta, which is accused of doing the same for its own generative AI software, LLaMA.
The humorist Sarah Silverman, author of the autobiography “The Bedwetter”, as well as Christopher Golden and Richard Kadrey, two writers known in the United States in particular for their works of fantasy, asked a federal judge in San Francisco to qualify this class action proceedings, which would allow others to join.
The three authors have no direct proof that OpenAI actually used their books to feed its language model, i.e. to create software, ChatGPT, capable of responding to requests in everyday language and generate text.
But they indicate, according to the subpoena document seen by AFP, having asked ChatGPT to offer a summary of the works and having obtained, in return, a “very fair” content and in line with the story of each book, even if “ some details are wrong”.
ChatGPT not being fed in real time directly from internet sources and relying on a limited amount of information, unlike Bard, Google’s interface, these summaries would mean that elements relating to these books have been entered into the software.
Concerning the interface of Meta, LLaMA, the three writers point out that the Menlo Park group admitted having used online bookstores, in particular Bibliotik, which offers digitized books without the authorization of their authors or their publishers.
Unlike OpenAI, Meta has only given access to LLaMA to a limited number of users and has not yet announced a launch for the general public.
The three authors do not mention, unlike ChatGPT, having asked LLaMA to produce a summary of their works.
At the end of June, two other writers, the Canadian Mona Awad and the American Paul Tremblay (“The cabin at the ends of the world”) had already summoned OpenAI before the same court.
Asked by AFP, OpenAI declined to comment and Meta did not respond immediately.
These legal actions are in line with other procedures that aim to police the practices of generative AI software developers, who are fed with immense amounts of data, some of which is protected by intellectual property law. .
In January, artists thus attacked, in the form of a group action, Midjourney, Stable Diffusion and DreamUp, three AI models formed thanks to billions of images collected on the Internet.