Federal judge rules in AI company Anthropic’s favor in landmark copyright infringement lawsuit brought by authors : NPR

The anthropogenic website and the mobile phone application are presented in this photo on July 5, 2024. A judge ruled in favor of the AI ​​company in a copyright violation case presented last year by a group of authors.

Richard Drew / AP

hide

tilting legend

Richard Drew / AP

AI companies may have the legal right to train their important language models on the works protected by copyright – as long as they obtain copies of this legally work.

It is the result of a first decision of its kind by a federal judge in San Francisco on Monday in a case of violation of the current copyright which opposes a group of authors against a large company of AI.

The decision is important because it represents the first substantive decision on how fair use applies to generative AI systems.

The doctrine of fair use makes it possible to use works protected by copyright by third parties without the consent of the copyright holder in certain circumstances such as the illustration of a point in a press article. Allegations of fair use are generally invoked by AI companies trying to plead in favor of the use of work protected by copyright to train their generative AI models. But the authors and other complainants of the creative industry pushed with a series of proceedings.

The authors face anthropogens

In their collective recourse in 2024, the authors Andrea Bartz, Charles Graeber and Kirk Wallace Johnson allegedly allegedly used the content of millions of pounds protected by copyright to train large language models behind their chatbot, Claude, including at least two works from each applicant. The company also bought paper books and scanned them before ingesting them in its model.

“Rather than obtaining authorization and paying an equitable price for the creations he exploits, the anthropes are picked up,” said the authors’ complaint.

In the ordinance on Monday, the main American district judge William Alsup supported the argument of anthropic, declaring that the use of books by the complainants by the company by the company to form their model of AI was acceptable.

“The use of training was fair use,” he wrote. “The use of the books in question to form Claude and his precursors was extremely transformative.”

The judge said that the digitization of books purchased in the form printed by Anthropic could also be considered a fair use, “because all that Anthropic has done was to replace the printed copies that she had bought for her central library with more practical and consultable digital copies for her central library – without adding new copies, creating new works or existing copies.”

However, Alsup also recognized that not all books were paid. He wrote Anthropic “downloaded millions of pounds protected by copyright for free from digital form from pirate sites on the Internet” as part of his efforts “to raise a central library of” All Books in the World “to keep” Forever “.” “”

Alsup has not approved the point of view of Anthropic “according to which the copies of the pirated library must be treated as training copies” and allow the piracy complaints to proceed to the trial.

“We will have a test on the hacked copies used to create the central library of Anthropic and the resulting damage, real or statutory (including the will),” said Alsup.

Bifurcated responses

The bifurcated decision of Alsup has led to responses in the same way of the persons involved in the case and the stakeholders of the industry.

In a declaration at NPR, Anthropic praised the recognition of the judge that the use of works to form important language models was “transformer – spectacularly thus”. Society has added: “In accordance with the objective of copyright to allow creativity and promote scientific progress, the major anthropic language models are formed on works not to run in advance and reproduce or supplant them, but to turn a hard corner and create something different.”

However, Anthropic also declared that he disagreed with the court’s decision to proceed to a trial.

“We believe that it is clear that we have acquired books to one goal – to build models of big languages ​​- and the court clearly judged that the use was right,” said the company.

A member of the complainant’s legal team refused to speak publicly about the decision.

The guild of the authors, a large defense group for professional editors, shared a declaration: “We do not agree with the decision that the use of hacked or digitized books for the training of large language models is fair use,” said the press release.

In an interview with NPR, the CEO of the Guild, Mary Rasenberger, added that the authors do not need to be too concerned about the decision.

“The impact of this decision for the books of books is actually quite good,” said Rasenberger. “The judge understood scandalous hacking. And this comes with statutory damages for violation of intentional copyright, which are quite high by book.”

According to the Copyright Alliance, the American copyright law declares that copyright violation can lead to legal damages up to $ 150,000 per work. The decision declares anthropic hacked more than 7 million copies of books. The damage resulting from the next trial could be enormous.

The part of the case focused on the responsibility of Anthropic for the use of hacked work should be tried in December.

Other cases and a new decision

Similar proceedings have been brought by other leading authors. Ta-nehisi Coates, Michael Chabon, Junot Díaz and the actress Sarah Silverman are involved in current cases against AI players.

On Wednesday, the American district judge Vince Chhabria ruled in favor of Meta in one of these cases. A trial for copyright violation was brought by 13 authors, including Richard Kadrey and Silverman. They continued Meta for having allegedly used pirated copies of their novels to form Llama. Meta claimed fair use and won because the authors did not present the proof that the use of Meta of their books had an impact on the market for their original work. However, the judge declared that the decision applies only to the specific work included in the trial and that in future cases, the authors making similar complaints could win if they argue stronger.

“These decisions will help technological companies and copyright holders to see where judges and courts are likely to go in the future,” said Ray Seilie, lawyer based in Los Angeles with Kinsella Holley Iser Kump Steinsapir, who focuses on AI and creativity. It is not involved in this particular case.

“I think they can be considered a victory for the brief of the AI ​​community, as they create a precedent suggesting that AI companies can use legally obtained equipment to train their models,” said Seilie.

But he said that it does not mean that IA companies can immediately go out and scan the books they buy with impunity, because decisions are likely to be on appeal and business could potentially end before the Supreme Court.

“Everything could change,” said Seilie.