(San Francisco) An American judge said on Monday that Anthropic company could lead to its artificial intelligence models (AI) with copyright -protected books without their permission, a decision that can mark a major step in the development of AI.
According to a federal judge of trial in San Francisco, the training of these models of CLAUDE with books, purchased or not, is allowed by the doctrine of “the reasonable use” of American laws on copyright.
“The use of the books in question with the aim of causing Claude has caused spectacular developments and was of reasonable use,” wrote the judge in his decision.
He described the AI ​​as “among the most revolutionary technology that many of us will see in their lives” and compared the learning of an AI to that which human beings perform by reading books.
Huge amounts of data are necessary to cause linguistic models generative, such as Chatgpt, one of Claude’s rivals.
Anthropic, valued at $ 61.5 billion and widely supported by Amazon, was founded in 2021 by former OpenAi engineers, the company that developed Chat GPT. It promotes, more ostensibly than its competitors, a responsible development of AI.
Many musicians, artists, media and writers – such as Andrea Bartz, Charles Graeber and Kirk Wallace Johnson with Anthropic – brought legal actions against several AI companies having used their data without permission or payment.
These companies generally defend themselves by brandishing the doctrine of reasonable use, arguing that IA training on large amounts of data radically transforms original content and is necessary for innovation.
Through the voice of a spokesperson, Anthropic welcomed that “the judge has recognized that using work to train major models was a source of innovation”. This decision “is consistent with the objective of copyright legislation to allow creativity and promote scientific progress”, according to the same source.
However, the judge was careful not to grant Anthropic a white-single, believing that his practice of downloading millions of pirated books to constitute a permanent digital library was not compatible with reasonable use.
Because in addition to having downloaded hacked books, the company bought books to scan them and keep them in digital format, according to justice documents.
And Anthropic was aimed at assembling a library “of all the books in the world” to train his AI models at leisure, according to the judge who considered that it was a violation of copyright, regardless of the objective.
The decision rendered on Monday is preliminary and the file will then be the subject of a civil lawsuit to determine whether it is advisable to inflict on anthropic any damages.
“The judge’s decision is a mixture,” said Keith Kupferschmid, head of the American NGO Copyright Alliance. “In some cases, artificial intelligence companies should be satisfied with the decision and in other cases, copyright should also be,” he said.
