MP3s made by NPCs? The rise of AI-generated muzak

Something weird is happening to music. Go on Spotify or YouTube and you’ll find a bizarre number of tracks that appear straight to your recommended section produced by artists with virtually no other releases. These songs sound eerily similar to the ones that came before them, each with nearly identical run times and one average listener a month. With impressively bland stock photo cover art, and titles like “Romilda Gebbia” and “Veneranda Caputa”, there’s something off about these songs, uncanny even, as if garbled in translation. Enter: AI-generated muzak.

“Many artists are coming forward saying there are seemingly anonymous artists with zero internet presence popping up [on Spotify] within days of X artist’s release with their own releases that sound strikingly similar to the original. Obviously, people rip off other artists all the time, there’s an entire industry dedicated to ‘type beats’, but what makes this seemingly different is the lack of human face or even a social media presence behind these unknown producers,” says Max Alper, AKA La Meme Young, a music educator and writer.

It’s nothing new that artificial intelligence is seeping into the music industry. Last month, fake tracks by Kanye and Drake went viral on TikTok, becoming a popular meme template for generative AI experimentalists adjacent to the spawning subgenre of absurdly funny deepfake videos on YouTube (see: Joe Biden and Donald Trump play Yu-gi-oh). “Heart on My Sleeve”, a song created by TikTok creator @ghostwriter, which combined the AI-generated voices of Drake and the Weeknd, picked up millions of hits on Spotify, TikTok, and YouTube before being taken down, prompting Universal Music Group to urge streaming platforms to block AI developers from using UMG artists to train the software. 

With Google now introducing a DALL-E-style text-to-music generator, we can only imagine how AI music is going to further muddy the waters. While the creation of generative music isn’t bad in itself – artists like Holly Herndon and Grimes have already demonstrated ways in which artists can work alongside such technology to empower and give them control over their likeness – there are valid fears of such sounds clogging up our feeds. Last month, Spotify removed tens of thousands of songs created and uploaded by AI music generator Boomy, though the company has created over 14 million tracks to date (which is around 13 per cent of the world’s recorded music).

Whether or not this will have actual sway over actual music enjoyers is yet to be seen. For the most part, the AI tunes available to us right now sound hilariously bland, like bot-generated muzak. But Alper asserts this might in fact be intentional: “Considering how so many of the ‘easy listening’ ‘chill’ ‘relaxing’ playlists are already marketed towards restaurants, office lobbies, yoga studios, etc and considering how much traffic these playlists get every month, it wouldn’t be absurd to speculate that we will start to hear more and more of these anonymous generative tracks in these contemporary muzak settings, if we aren’t already.”

For the uninitiated, muzak describes the sort of dulcet sounds found in hotel lobbys, elevators and members clubs. Initially deployed for its calming effect on listeners, the music designed for background purposes is intrinsically passive and uniquely bland, often deployed by companies to make people happier, more productive. By extension, the rise of AI-generated muzak comes at a time when the world is arguably scarier than ever, where climate change and invasive technologies threaten our position on earth. It’s ironic that something so ‘scary’ and unknown is also used to calm us of said fear. But given the pace at which generative art fills our feeds, the chances of it becoming boring ahh, and therefore easy listening – and fast – isn’t all that surprising. 

“If the human musicians working at music licensing companies are already on the chopping block as music AI software becomes sophisticated enough to replace their 30 second piano and vibraphone loops for a Toyota commercial, then this is simply another way to cut corners and save money overall,” says Alper. “Music licensing shareholders will pressure their companies to shift to AI generated sound to save money, and those same shareholders also have stakes in streaming platforms. If the stakes are already so low monetarily for the individual independent artists, why wouldn’t these people just cut them out of the equation entirely?”

There is however a potentially more sinister side to the rise of AI-generated muzak. Spotify has previously been accused of incentivising gaming its algorithm through stream farms (where you can purchase fake streams to receive more royalties) and generative sleep playlists. “They’ve even been accused of pumping their playlists with artists who release tracks exclusively through music licensing companies that Spotify shareholders have stakes in, such was the case with Epidemic Sound, also based in Stockholm,” adds Alper. While Spotify does have policies against fake streaming, it’s not that tinfoil hat to believe that there might be ulterior motives pedalled by the company that feed into its wider incentives. Back in 2021, Spotify founder Daniel Ek invested 100m euros (£85.2m) into defence company Helsing, which uses AI to support militaries in battlefield assessment operations. “One could speculate that streaming platforms such as Spotify are continuing in the incentivisation of gaming the algorithm as they’ve done before, only now the music is entirely generated by the rudimentary music AI software.”

There’s also wider concerns about machines learning from artists’ songs, unauthorised. Not to mention the sheer number of these songs flooding streaming services distracts from real humans and their compositions reaching our ears, which means increasingly bad payouts for artists. Mirroring the problem with AI image generators earlier this year, where artists had their styles learned by DALL-E or other visual AI generators without their consent, the legalities surrounding data sets are yet to be hashed out – and there’s no standard for compensating artists for AI training. “The larger implication here is that anything you upload to these platforms or the internet at large can potentially be fair game for machine learning,” expands Alper. “Labels and masters rights holders need to get their shit together *yesterday* to find a way to regulate this. I foresee generative AI ripoff-type beats falling under a new grey area of legality akin to sampling. If the artists in question are small fry, why not ask for forgiveness rather than permission when it comes to a dataset learning their albums? Will be much harder to win that case in court with a major label artist, but independent artists are already used to getting the shit end of the stick and this will just add more fuel to the dumpster fire.”

Cutting out the artist from the equation will of course make Spotify even more of a barren hellscape than it already is. Perhaps AI muzak will become its own sub-genre, no different to the countless generative soundtracks already out there. “If both artists and labels feel like they’re losing ground to faceless AI-generated music then that would encourage seeking alternative means of distribution, a revaluing of what music is, and possibly a spur on to create art more wondrous and captivating than a data modelled soundalike,” suggests Tom Furse, a member of The Horrors and AI enthusiast. As the technology advances, some commentators have referred to Big Tech’s rush to develop AI as an arm‘s race, the impetus is on us to explore new forms of artistic expression and break out the cycle of increasingly mid content.