Musicians have long been paid every time their work gets used. Whether it's a vinyl or CD sale, a stream, a radio play, a cover version, or a niche use like karaoke, the industry has built up agreements over decades about what counts as "use" and what that use is worth.
Now a new kind of use is forcing the question open: feeding recorded music into the systems that train artificial intelligence. According to IEEE Spectrum, the music industry is grappling with how to extend its established compensation model to cover music used as AI training data.
The logic the industry leans on is simple, IEEE Spectrum notes: the more something is used, the more value it generates — and, by extension, the more its creators should be paid. That principle has underpinned royalties across formats for generations. The challenge with AI is that training doesn't look like a sale or a stream. A model may ingest enormous catalogs of music to learn patterns, without any single play that the existing systems are designed to count and pay out on.
That gap is where the debate sits. If musicians are accustomed to getting paid each time their creative work is used, then training an AI on their recordings is arguably a use too — one that current agreements were never written to address. Working out how to attribute and price that use is the unresolved problem the industry is now confronting.
Why it matters: as AI tools increasingly generate music, the question of whether the artists whose work taught those systems get compensated will shape both musicians' livelihoods and the rules for how creative work is valued in the AI era.