BBC News
An American judge ruled that the use of books to form artificial intelligence software (AI) is not a violation of the American law on copyright.
The decision came from a trial brought last year against the Société d’Ia Anthropic by three writers, a novelist, and two non-fiction authors, who accused the company of stealing their work to train its Claude AI model and develop a business of several billion dollars.
In his decision, judge William Alsup wrote that the use by anthropic of the authors by anthropic was “extremely transformative” and therefore authorized by American law.
But he rejected Anthropic’s request to reject the case, judging that the company should be judged on its use of hacked copies to build their equipment library.
Anthropic, a company supported by Amazon and the mother company of Google, alphabet, could face $ 150,000 in damages by work protected by copyright.
The firm holds more than seven million pounded pounds in a “central library” according to the judge.
The decision is among the first to weigh on a question which is the subject of many legal battles through industry – how large language models (LLM) can legitimately learn from existing material.
“Like any reader who aspires to be a writer, the LLM of Anthropic trained on the works, not to run in advance and to reproduce them or supplant them – but to turn a hard corner and to create something different,” wrote judge Alsup.
“If this process of training required reasonably to make copies within the LLM or otherwise, these copies were engaged in transformative use,” he said.
He noted that the authors had not claimed that the training had led to “counterfeit the introductions by abdomin” with replicas of their works generated for users of the Claude tool.
If they had done, he wrote: “It would be a different case”.
Similar legal battles have emerged on the use by the AI industry of other media and content, journalistic articles to music and video.
This month, Disney and Universal filed a complaint against the IA Midjourney images generator, accusing him of hacking.
The BBC is also take legal action on the unauthorized use of its content.
In response to legal battles, some AI companies responded by concluding agreements with the creators of the original documents, or their publishers, to obtain licenses of equipment for use.
Judge ALSUP authorized the defense of the “fair use” of Anthropic, paving the way for future legal judgments.
However, he declared that Anthropic had violated the rights of the authors by saving pirated copies of their books as part of a “central library of all the books in the world”.
In a statement, Anthropic said he was satisfied with the judge’s recognition that his use of the works was transformative, but in disagreement with the decision to take a lawsuit on how certain books were obtained and used.
The company said that she had remained confident in her case and evaluated her options.
A lawyer for the authors refused to comment.
The authors who carried the case are Andrea Bartz, a best-selling mystery thriller, whose novels include we warned and the last ferry, and non-fiction writers Charles Graeber and Kirk Wallace Johnson.
