Large technological companies form their AI models mainly on the work of other people, such as scientists, journalists, filmmakers or artists.
These creators have long opposed practice. Now, the American copyright office seems to have joined them.
The office published its latest on Friday in a series of reports exploring copyright laws and artificial intelligence. The report examines whether the content protected by copyright uses AI companies to train their AI models is qualified under the doctrine of fair use.
AI companies will probably not like what they read.
AI companies are desperate for data. Most of them believe that the more information can digest information, the better. But with this insatiable consumption, they risk submitting to copyright laws.
Companies like Open IA had to face a series of creators’ prosecution who say that the training of AI models on their work protected by copyright without permission undermines their rights. AI leaders argue that they have not violated copyright laws because training is fair use.
According to the new US Copyright Office report, however, it’s not that simple.
“Although it is not possible to prejudge the result in a particular case, the previous one supports the following general observations,” said the office. “Various uses of works protected by copyright in AI training are likely to be transformative. The extent to which they are fair, however, will depend on what has been used, on what purpose, for what purpose and with what controls on outings – which can all affect the market.”
The office has made a distinction between AI models for research and commercial AI models.
“When a model is deployed for purposes such as analysis or research – the types of uses that are essential for international competitiveness – the results are unlikely to replace the expressive work used in training,” said the office. “But make the commercial use of large works of work protected by copyright to produce expressive content that competes with them on the existing markets, in particular when accomplished thanks to illegal access, goes beyond the established limits of fair use.”
In the report, the office compared the results of artificial intelligence which essentially copy its training materials to outputs with added elements and new value.
“At one end of the spectrum, the formation of a model is the most transformative when the objective is to deploy it for research, or in a closed system which limits it to a non -substitute task,” said the office. “For example, the training of a linguistic model on a large collection of data, including social media publications, articles and books, for deployment in systems used for content moderation does not have the same educational objective as these articles and books.”
The formation of an artificial intelligence model to create outings “substantially similar to works protected by copyright in the data set” is less likely to be considered a transformative.
“Unlike cases where the copy of computer programs to access their functional elements was necessary to create new interoperable works, using sound images or recordings to form a model that generates similar expressive outings does not simply remove a technical obstacle to productive competition,” said the office. “In such cases, unless the original work itself is targeted for comments or parody, it is difficult to see the use as a transformer.”
In another section, the office said it had rejected two “common arguments” on the “transformative nature of AI training”.
“As indicated above, some argue that the use of copyright-protected works to train AI models is intrinsically transforming because it is not for expressive purposes. We consider this argument to be wrong,” said the office.
“Nor do we agree that IA training is intrinsically transforming because it is like human learning,” he added.
One day after the office published the report, President Donald Trump dismissed his director, Shira Perlmutter, a spokesperson for Business Insider.
“Saturday afternoon, the White House sent an email to Shira Perlmutter by saying” your post as copyright register and director of the American copyright office is immediately dismissed, “said the spokesman.
While Trump, with the help of Elon Musk, who has his own AI model, Grok, sought to reduce the federal workforce and close certain agencies, some have seen the moment of the dismissal of Perlmutter as a suspect. New York representative Joe Morelle, Democrat, addressed Perlmutter’s dismissal in an online statement.
“Donald Trump’s dismissal on the copyright register, Shira Perlmutter, is a cheeky and unprecedented takeover without legal basis. It is certainly not a coincidence that he acted less than a day after refusing to stadium the Elon Musk rubber to exploit works of works protected by copyright to form models of AI,” said the press release.
Large technology and AI companies have rallied around Trump from its elections, led by Musk, which has become the face of the White House office and administration efforts to reduce federal spending. Other technological billionaires, such as the meta-PDG Mark Zuckerberg and the CEO of Openai, Sam Altman, have also been up to Trump in recent months.
A White House representative did not respond to a request for comments to Business Insider.