Lever CohenHis first encounter with Google’s creative artificial intelligence surprised him. “Dames [Hassabis, CEO of Google Deepmind] And his team presented a research project about genAI and music and my head was off my shoulders,” said Cohen, global head of music at Google and YouTube. Billboard in November. “I walked around London for two days excited about the possibilities, thinking about all the problems and acknowledging that genAI in music is here – it’s not even close.”
While some major labels are touting YouTube as an important partner in the emerging world of music and AI, not everyone in the music industry is as enthusiastic about these new efforts. That’s because Google trained its model on a large set of music — including copyrighted major label recordings — and then went on to show it to rights holders, rather than first. Permission should be sought, according to four sources with knowledge of the search giant’s push into generative AI. and music. This could mean that artists “opting out” of such AI training – a key requirement for many rights holders – is not an option.
YouTube made sure to sign unilateral licenses with some parties before rolling out a beta version of its new genAI “experiment” in November. DreamTrack, the only AI product it has released publicly so far, allows select YouTube creators to soundtrack clips with pieces of music on shorts, based on text prompts, featuring the voices of popular artists. May include duplicates. (A few major label acts participated, including Demi Lovato and Charli XCX.) “Our superpower was our deep collaboration with the music industry,” Cohen said at the time. But the negotiations, which many in the business see as the ideal setting for broader, label-wide licensing deals, have been ongoing for months.
According to multiple sources familiar with the company’s label negotiations, negotiating with a company as big as YouTube was made difficult because it already took what it wanted. Meanwhile, other AI companies are pushing ahead with their own music products, putting pressure on YouTube to advance its technology.
In a statement, a YouTube representative said, “We are committed to working with our partners across the music industry to develop AI responsibly and in a way that monetizes participants. , could be awarded long-term opportunities for controls and potential genAI tools. Content down the road,” declined to get specific about the license.
GenAI models require training before they can generate correctly. “AI training is a computational process that aims to deconstruct existing tasks to mathematically model them. [they] works,” Google explained in comments to the US Copyright Office in October. “By isolating existing works, the algorithm creates the ability to predict how new works should be assembled.”
Whether or not a company needs permission before starting this process on copyrighted works is already the subject of several lawsuits, including Getty Images v. Stability AI and Authors Guild v. OpenAI. In October, Universal Music Group (UMG) was among the companies that sued AI startup Anthropic, alleging that “in the process of creating and operating AI models, [the company] Illegal copying and dissemination of copyrighted works in large quantities.”
As these cases progress, they are expected to set the precedent for AI training — but that could take years. Meanwhile, many technology companies seem to be heeding the Silicon Valley rallying call to “move fast and break things.”
Although rights holders deny copyright infringement, tech companies say their activities fall under “fair use” — the US legal doctrine that allows unlicensed use of copyrighted works in certain circumstances. Gives. News reporting and criticism are the most common examples, but TV show recordings for later viewing, parodies and other uses are also included.
In its comments to the U.S. Copyright Office, Anthropic wrote, “A variety of cases support the proposition that copying a copyrighted work as an intermediate step to creating a non-infringing production constitutes fair use.” can be formed.” “Innovation in AI depends fundamentally on competence. [large language models] to learn in a computational sense from the widest possible variety of publicly available content,” Google said in its comments.
“When you think about creative AI, you mostly think about companies that are taking a very advanced approach — Google, OpenAI — with sophisticated models that require a lot of data. is,” says. Ed Newton-Rex, who resigned as Stability AI’s vp of audio in November as the company grappled with copyrighted works. “In this community, where you need a lot of data, you don’t see many people talking about the concerns of rights holders.”
when Dennis Coker, president of global digital business and U.S. sales for Sony Music Entertainment, speaking at a Senate forum on AI in November, rejected the fair use argument. “If a creative AI model is trained on music to create new musical works that compete in the music marketplace, that training is not a fair use,” Coker said. “Training in this case cannot take place without the consent, credit and compensation of artists and rights holders.”
UMG and other music companies took a similar stance in their lawsuit against Anthropic, warning that AI firms should not be “excused from complying with copyright law” because they claim they are “a part of society.” will provide a facility of immense value.
“Undoubtedly, Anthropic would be a more valuable company if it could avoid paying for the content it acknowledges,” UMG wrote at the time. “But that should hardly compel a court to give him a get-out-of-jail-free card for wholesale theft of copyrighted material.”
This season, getting major labels on board like Google and YouTube did last year with the Dream track — after training the model, but before releasing it — could be a step forward from the music industry’s point of view. Is. At least it’s better than nothing: In 2004, Google began scanning large volumes of books, now known as Google Books, without asking permission from copyright holders. The Authors Guild sued Google alleging copyright infringement, but the case was eventually dismissed — nearly a decade later in 2013.
While music business-backed AI-related bills have already been proposed in Congress, the two sides are currently pushing back against each other. Newton-Rex summed up the different mindsets: “What we in the AI world think of as ‘training data,’ the rest of the world has long thought of as creative output.”
Additional reporting by Bill Donahue.