Press Clipping
AI now has a key to Spotify’s songwriting fortress

It’s impossible to predict what sanctioning AI’s access to a vast trove of recorded music will do to the music business. Finally, a test case of how businesses fearful of AI hope to co-opt it.

What kind of music will Spotify’s suggestion algorithms serve its millions of subscribers on a rainy Sunday afternoon: A Lana Del Rey ballad, or machine-generated ambient noise? Well, eventually, maybe a combination of the two.

In a new twist in the music streaming wars that’s seen record labels fight the encroachment of artificial intelligence on pop stars’ turf, Universal Music Group on Tuesday announced it would partner with Berlin-based artificial intelligence start-up Endel to craft soundscapes designed as background listening for activities such as sleep, relaxation or increased focus.

The software will be allowed to use Universal’s catalogue of existing albums and artists to produce these functional soundscapes. It’s a truce rather than a lasting peace, and one with potentially noisy consequences.

This clearly marks a “can’t beat ’em? Join ’em” chapter in the race to contain AI disruption. UMG has this year been particularly outspoken in cracking down on the excesses of machine-made music, pushing Spotify and other streaming platforms to do more to fight audio deepfakes, AI data-scraping and a deluge of non-musical music from whale song to white noise that’s eating into labels’ market share.

The deal suggests a growing recognition that a purely hostile strategy against AI won’t work. Tech is improving at a rapid pace and looks like another competitive threat in a market where indie labels and DIY artists account for about 35 per cent of the available music, with UMG at 30 per cent, according to research firm MIDiA.

Why not partner with a company that’s willing to play by the rules? Endel has already made a soundscape album with songwriter James Blake, and an alliance could help UMG get a slice of a “functional” music market that’s already delivering 10 billion streams a month. Releasing a “sleep” or “chill-out” version of Taylor Swift or Drake’s latest album could enhance their star power, not rob artists of their royalties.

And from Endel’s point of view, it’s recognition that even a machine able to spit out infinite music can do far better with the (authorised) star power of a human. The start-up’s ambient soundscapes are composed by software but based on the musical building blocks, or stems, of songs created by human artists; the UMG deal opens up an Ali Baba’s cave of those stems for AI consumption.

Endel co-founder Oleg Stavitsky tells me: “I hope this is a new chapter that shows how music companies can embrace and use AI.” He’s keen to stress that this will be done in partnership with artists, whether they’re actively involved in the composition or not, although ultimately, it looks like rights-holders like Hipgnosis Songs Fund will have a big say.

Where things get less clear-cut are what it means to everyone else. For platforms like Spotify and Deezer, AI music is a double-edged sword: for all the downsides of potential fakery and non-musical flotsam and jetsam, it offers them the opportunity of expanding available product and cutting deals outside the remit of record labels.

And for musicians who aren’t part of rock’s aristocracy, cutting through the noise is going to get a lot harder in an industry that’s already winner-takes-all: some 95 per cent of streaming royalty payouts are generated by the top 15 per cent, or 200,000 artists. If the UMG-Endel deal reinforces the industry’s lop-sided economics and delays fairer revenue-sharing models, that will not be a good outcome.

And in the longer term, it’s not clear where this type of deal will lead – what if handing AI the history of recorded music ends up accelerating, not delaying, the end-game of generative content that’s better than the real thing? UMG has praised Endel as a model of “ethical AI”, but these are very early days for understanding how guardrails around this technology should work.

Alphabet CEO Sundar Pichai, writing in the Financial Times, gave the example of providing some kind of identification label for when AI has been used to generate content – expect the music industry to eye this as a vital defence mechanism.

For now, it’s impossible to predict what officially sanctioning AI’s access to a vast trove of recorded music will do to the music business. But it could be a test case of how businesses fearful of AI hope to co-opt it – while accepting the risk of inviting the fox into the hen house.