tech

Music made by generative AI has been on the horizon as an issue for a couple of years, and the industry started playing close attention when the “Fake Drake” track hit streaming services in April 2023. The part of the issue that gets the most attention is, of course, that part that involves celebrities — especially Drake, who objected when his voice was spoofed by AI and then used AI software to spoof the voices of other rappers.
That’s just the tip of a particularly dangerous iceberg, though, according to a new study commissioned by global collective management trade organization CISAC and conducted by PMP Strategy. Vocal imitations are fun, but how much time can you really spend listening to Frank Sinatra sing Lil Jon? The bigger issue is what generative AI means for new music — first for passive listening, presumably, and eventually for the entire business.

Trending on Billboard

Using quantitative and qualitative research, PMP concluded that a full $4 billion could be lost to composers and publishing rightsholders in 2028 — 24% of the revenue they collect through CISAC-member organizations. (This gets complicated: The study measures revenue that comes to them through collective management organizations, which includes performing rights and, in most cases, mechanical rights royalties.) This doesn’t even count the recording business, or revenue from synch licensing. By 2028, generative AI music will be worth $16 billion and the services that create it will bring in $4 billion in revenue. One previous study commissioned by SACEM and GEMA reached somewhat similar conclusions — in that case, that 27% of revenue would be at risk.

The music business is driven by new pop music, so there’s a tendency to focus on stars like Drake. But publishers and songwriters also depend on background music played in public — at bars and in stores — and streaming services have made a business out of utilitarian music, sounds to help listeners focus, relax or sleep. That might be where the impact of generative AI is felt first — the restaurant that plays AI music to avoid paying a performing rights organization, a playlist that avoids copyrighted music, a low-budget film that uses music generated by an algorithm. By 2028, the study predicts that generative AI could cut into 30% of digital revenue, 22% of TV and radio revenue and 22% of compositions played in public. Eventually, presumably, AI could generate hits as well.

To some extent, AI is inevitable.

“There’s no way we can or should stand against AI — it can be a wonderful tool,” said songwriter and ABBA frontman Björn Ulvaeus, who serves as president of CISAC, at a Nov. 3 online press conference to announce the results of the study. Many composers already use it as a tool, mostly for specific purposes. (I thought of using AI to write this column but I got nervous that it would do a poor job — and terrified that it would do a good one.) “Creators should be at the negotiating table,” Ulvaeus said. “The success of AI isn’t based on public content — it’s based on copyrighted works. We need to negotiate a fair deal.”

That’s only possible if generative AI companies are required to license the rights to ingest works, which by definition involves copying them. That seems to be the case already in the European Union, although the AI Act says rightsholders need to opt out in order to prevent the unauthorized use of their work to train AI software. In the U.S., until Congress turns its attention to AI, this is a matter for the courts, and in June the RIAA, on behalf of the major labels, sued the generative AI companies Suno and Udio for allegedly infringing their copyrights.

The study lays out a picture of how generative AI will develop between now and 2028, in both the music and audiovisual sectors — it will affect streaming revenue the most, but also change the market for music in TV and film. And although few people spend much time thinking about background music, it provides a living, or part of a living, for many musicians. If that business declines, will those musicians still be able to play on albums that demand their skills? Will studios that are booked for all kinds of music survive without the background music business?

The study predicts that the generative AI business will grow as the creative sector shrinks, as some of the money from music goes to software — presumably software trained on copyrighted works. “In an unchanged regulatory framework, creators will not benefit from the Gen AI revolution,” the study says. Instead, they will suffer “the loss of revenues due to the unauthorized use of their works” to train AI software and the “replacement of their traditional revenue streams due to the substitution effect of AI-generated outputs.”

Making sure composers and other creators are compensated fairly for the use of their works in training generative AI programs will not be easy. It would be hard for creators and other rightsholders to license a onetime right to use material for training purposes, after which an AI model can use it forever. The current thinking is that it makes more sense to license these rights, then require AI programs to operate with a certain level of transparency to track the works they reference in response to a given prompt. Then the owners of those works can be paid.

This is going to be hard. Getting it right means starting immediately — and the obvious first step is clarifying creators’ rights to be compensated when their work is used to train an AI.