Spotify is full of AI music
fastcompany.com
fastcompany.com
Ghost artists on Spotify
I'm done with Spotify. I've had the service since it was beta in the Nordics, maybe 2008? It was amazing to be able to stream almost every piece of music there is.
I love music, I'm an amateur musician and I respect artists. This treats music as a fungible commodity. That's not what music is to me, so I choose not to support that.
I also understand the predicament they are in. I had friends who worked at Rdio, and I briefly worked at a startup doing music recommendation until the big 3 quoted us how much royalties they wanted in advance to stream their catalogue one time to learn from.
The record labels are the root cancer on the industry, but the alternative for Spotify might be to become a better label, not become another bad actor.
This is only going to accelerate. Until the labels die, the universe will find every way to route around them.
Hopefully concert venues are next.
If you're not familiar, the company has a corner on the market for concert venues and all associated services (booking, ticket resale, food and liquor vendors, point of sale, etc.) They use this to inflate ticket prices.
"Elevator generic background music is trash. Good riddance. "
"No, X and Y in the genre are good bands"
"So you agree it is generic elevator music?"
How I read it, the second comment is saying that AI music competes with machine-replicable human music, a subset of human music, but not all.
Frankly, while Brian Eno's Music for Airports was monumental in the history of ambient music, in 2024 people unfamiliar with it aren't going to be able to tell it apart from select AI clips (as much as that will hurt to admit). That's less about it being generic and more about how good AIgen can mimic humans. Just like how humans couldn't tell apart human from AIgen in impressionist art: https://www.astralcodexten.com/p/how-did-you-do-on-the-ai-ar... -- we wouldn't use that example to say that human art is just generic slop.
On re-read, I do think the thread started off on the wrong foot by using the word generic. Ambient music is simply some of the easiest music to machine generate being minimalist with heavy use of synthesizers, and much of it was machine generated in the first place in a DAW.
As music historians note the profundity of their work with that explicit aim in mind
It was possible to do a credible job composing algorithmic music in many genres before the current AI frenzy. Before long, you won't be able to tell the difference between human-composed and machine-composed music in any genre.
It will be quite a while before that happens with novels and screenplays. Wouldn't bet against it in the long run, though.
Paint a picture without painting a picture!
Write a novel without writing a novel!
Compose a melody without composing a melody!
There was an absolutely insane comment on HN a few months ago where the poster said that they felt a sense of accomplishment after "producing" a song on Suno. I mean... how could that possibly be true? They didn't compose the melodic motif, figure out the chord progression, structure the song, or even write the lyrics. The world is getting a bit too kafkaesque for my liking.
You just pointed the thing and pressed a button. Yet you feel accomplishment.
Sure you might have chose a lens, depth and tried to point the things somewhat precisely. Then you sat at your computer to cut enough of the edges and fiddled with some curves so it feels nice for you. But ultimately you didn't make any of it. Machine made it all. You just guided it with the knobs provided for you. Yet you feel accomplishment if you make something good because you also made hundreds of it that were crap.
They "created" something that, until now, would have required skills they didn't have. You don't see how someone could feel a sense of accomplishment for the end result, the thing they poofed into existence and like?
We use tools to make things (even art) easier to create. I don't make my paints from scratch (anymore), or string my paintbrushes (anymore), or now sometimes choose my color palettes or general composition -- tools (like genAI) exist to let artists create art easier, and saying "you didn't put in enough thought/work/etc for it to count as art!" feels dismissive of the creative process and the joy people find in expressing themselves, regardless of how they got there.
I could see how someone with no background in music could feel a sense of accomplishment with this.
It is like a gen z version of playing the ironman or smoke on the water riff on guitar. It is just not that hard to play the ironman riff either.
As someone who has put thousands of hours into learning music, anything that gets people into music is a positive to me.
It seems like generative AI art and music really triggers tech people who have nothing to do with art and music.
"Kids these days starting off programming a web app with a language model. In my day we had to take adderall and debug assembly code just to print an underscore to a dot matrix printer. That is the right way to start and anything else is WRONG"
I have seen amazing works of art and music from people with zero technical knowledge. It may be rough or weird but the jagged edges is what makes it art.
I wonder how people will feel about this long-term as it blurs the line between generated from scratch and tool-assisted. Clearly the intent is to make it easier for producers to get the results they want. How easy is too easy?
I believe in the near future there will be more discussion about specific parts of songs being derived from AI even if the rest is traditionally composed. Because that is the natural result of creating more efficient tools to solve specific needs that artists have, when "generate my song for me" is not an option they would ever consider (given today's rapidly changing norms, at that point it becomes an issue of integrity over efficiency).
I think it is also interesting how revered Vocaloids have been in certain fandoms stretching back to the late 00's in spite of their artificiality. Whereas a lot of the autogenerated content today is derided for being artificial, "not the real thing." I guess because the vastly wider gap and set of limitations of Vocaloid compared to human singers causes creative workarounds that wouldn't be considered otherwise. But the creator is the one who needs to plan and execute those workarounds, as it confirms to their artistic preference.
Maybe in the future the AI art tools that are seen to be most ethical are the ones that preserve the artistic vision of the user. An aspect that artists are lamenting the loss of. Less outsourcing and more deliberate collaboration.
Doesn't promote AI songs, has high-res, lossless sound quality and pays artists more.
Also they pay to the rights-holders exactly the same as Spotify, which is 70% of your subscription. It just happens that Spotify has on average a lower revenue per stream, but for your particular use they both pay the same thing.
The algorithm still suggests some forgotten classics. On YT music I can find entire albums of somewhat more obscure yet excellent artists that are completely missing from Spotify.
I pay for Youtube with a family sub. Good value for me.
My Analog Journal is a great YT channel to zero in on old world music hidden behind the marketing noise of American pop
As for new artists yeah, making art to pay bills is not making art. It’s boring capitalism. IMO it’s why music by humans sucks now. Back in the day there was financial demand to unicorn off one album.
Marketing and advertising, VC are a box. Middle men with no skills but keeping that hand out
On an AM family plan, if they are really paying out 0.01 per stream... then anyone streaming more than 19 songs/day is a money loser for apple.
Spotify has the same licensing agreement.
The difference is that Spotify has a free tier (lower revenue from those users), their users listen more, and they may also have more customers in lower-income countries with lower subscription rates.
Which is to say - of you took a specific category of users - say, US-based paying subscribers and kept their listening activity the same - every streaming service would pay out roughly the same amount of royalties per-stream for that group of listeners.
This has been true for at least half a century. Probably many centuries.
I would bet every dollar of my net worth that they are critically wrong on this; its such a beautiful little package that some VP can sell a higher-up to explain away why something they're responsible for isn't performing as expected. I've heard dozens of people complain about how crowded the Spotify app is becoming. I don't know anyone who actively and regularly uses Spotify as a podcast platform; when Joe Rogan was exclusive to Spotify for a bit, I heard a couple people complain about that, which means they were (begrudgingly) using it; but that isn't the situation anymore.
How these complaints convert into behavior changes, subscription cancellations, etc; I don't know.
After cancelling Spotify I surprisingly ended up on Tidal, and it's been amazing so far. Feels like how Spotify used to feel before they started shipping the org chart like all big companies eventually do. Plus its cheaper, offers higher quality streams, and gives bigger payouts to artists.
It was sending the staff crazy to listen to it all day.
Netflix was well on the same course; but instead made what was absolutely not an obvious pivot at the time, to just become a studio like those dozen other guys. In retrospect, it was a genius move because their access to fine-grained data on what their users watch became a differentiating factor to feed back into what the studio greenlights. But, it would have been the right move even if they didn't capitalize on that data, because there genuinely is no business to be built being a middle-man who has to pay for content. Spotify knows this, but they don't have the same escape hatch Netflix had; but they're tearing their app and users apart trying to find one.
Where did you get this from? All available answers when searching seem to point out "Time Spent Listening" being the north-star metric for Spotify, not whatever you said. Did you just make this up?
Seems like the whole online industry is eliminating itself and cheering all the way down.
https://en.wikipedia.org/wiki/Dead_Internet_theory
When I first heard about it on /x/ about 6 years ago I thought it was a bit of a morbid joke around how if all your human interaction is online you can't definitively state that the universe exists outside your front door. But recently it's started to look like more of a true statement that, at least as far as the internet goes, it will soon be impossible to know if anybody you interact with actually exists.
FWIW I'm actually looking forward to this happening, for every social activity that is often claimed to have been better before the internet (ie dating, casual debates about politics, etc) has a potential to return to its pre-internet status quo as people come to the conclusion that none of it is real and gradually drop out. Internet will still exist as a valuable tool for transmitting data but the concept of online communities and social networks will suffer and degrade until there's nothing but spam like usenet did.
I much prefer Maggie Appleton‘s Dark Forest writing and talk on the matter:
> The dark forest theory of the web points to the increasingly life-like but life-less state of being online. Most open and publicly available spaces on the web are overrun with bots, advertisers, trolls, data scrapers, clickbait, keyword-stuffing “content creators,” and algorithmically manipulated junk.
> It's like a dark forest that seems eerily devoid of human life – all the living creatures are hidden beneath the ground or up in trees. If they reveal themselves, they risk being attacked by automated predators.
Links:
- https://maggieappleton.com/ai-dark-forest
- Talk: https://youtu.be/QPoM-h1fK8M
edit: formatting
But after reading about this phenomenon, I also started to dig deeper into which artists were actually suggested each week. And to my surprise there were quite a bit of artists I've never heard of, and were hard to find further information on.
Hopefully Spotify won't touch weekly release playlist (the one which refreshes every friday) of artists you already listen to and have saved into personal playlists.
Money quote:
> Spotify’s own internal research showed that many users were not coming to the platform to listen to specific artists or albums; they just needed something to serve as a soundtrack for their days, like a study playlist or maybe a dinner soundtrack. In the lean-back listening environment that streaming had helped champion, listeners often weren’t even aware of what song or artist they were hearing. As a result, the thinking seemed to be: Why pay full-price royalties if users were only half listening? It was likely from this reasoning that the Perfect Fit Content program was created.
The "ghost" thing is interesting too, that sounds almost like the industry from before The Beatles, when bands were just people hired by the record companies to record songs written by others, and the companies owned pretty much everything.
> Spotify Employees Say It's Promoting Fake Artists to Reduce Royalty Payments to Real Ones
[...]
> According to a shocking new book, Spotify has been promoting so-called "ghost artists" so it can avoid paying its piddling royalties to real artists.
> In an excerpt from "Mood Machine: The Rise of Spotify and the Costs of the Perfect Playlist" published in Harper's, author Liz Pelly reveals that the streaming platform has a secretive internal program that prioritizes cheap and generic music.
> Dubbed "Perfect Fit Content" or PFC for short, this program not only involved a network of affiliated production firms creating tons of "low-budget stock muzak" for the platform, but also a team of employees who surreptitiously placed tracks from those firms on Spotify's curated playlists.
I think the mods should change this post to link to the Harper's piece, or even the Futurism one.
More broadly, it is interesting how Spotify is following the path blazed by supermarkets, then by video streaming services, in starting to sell (and promote, and prioritise) own-brand content. (Of course the Futurism summary is itself an example of a related problematic practice ...)
The recommendations used to feel like magic, now it's very limited in what it considered similar to something I'm listening to, like it fixates on one attribute of a song, but not the broader context or genre.
I don't know if this phenomenon is related to the main article, but I've been wanting to rant about it for a while :-P
However, AI-generated fake podcasts are a different story—they're often frustrating and a waste of people's time.
Unfortunately, platforms like Apple Podcasts and Spotify are increasingly populated with these spammy, AI-created podcasts. For those interested, here's a small subset (4,000+) of such AI-generated fake podcasts available on Kaggle: https://www.kaggle.com/datasets/listennotes/ai-generated-fak....
To your point, I actually used to run CPU Bach (a 3DO game that tried to procedurally generate contrapuntal classical music) as a good background music generator while studying as a kid. Very few people remember it or the 3DO for that matter but it was quite impressive for its time!
The latest grift is posting AI "covers" of real songs to farm SEO hits while evading copyright strikes, which results in fever dreams like this:
I’d rather have human created music, even when it is background music and I’m doing a primary activity that needs focus.
In other contexts, where the stakes are even lower, such as elevator music or on-hold call music AI gen music will probably have less people sneer at it.
But not at the 10 bucks or whatever they charge now
Spotify says you can make a collection as the all powerful curator, and hear from it in what has been understood to be a random selection. They’re breaking the definition of shuffle.
Sure, there are some obvious cases. But I have been collecting AI songs that I like, and I can't tell the difference from human-made music.
I found that AI helped a lot, I can still write the story, the poem, the lyrics - and just let AI find a melody that represent the story as close as how I imagined it on my mind.
I mostly design and write the story first, then captures the story as poem or lyrics, then let AI “cover” it for me. I don’t want AI to write for me because the story and lyrics is what I cared about, and I like to put metaphors and hidden messages on it.
For example, this album that tells about the phases of life as seasons is AI generated.
“Nirvana Tribute ‘Smells Like Teen Spirit’ Girl Group 1960’s Tribute”
https://www.youtube.com/watch?v=qwSeVM1g9hs
At first I thought it was a real singer, but then I remembered how convincing the lastest text-to-music services are becoming.
“Stairway to Heaven” in a similar style:
I haven't used Spotify since.
Listening to retro music makes more sense to me these days.
Disclaimer: I’m not a musician.
One wrong move, and one export/import to another service and I’m gone.
I've experienced none of this on Deezer. They appear to still employ humans to curate playlists featuring actual artists, and as a nice bonus they support lossless audio like Spotify has been half promising for at least the last five years.
I feel like the introduction of AI music is shortchanging musicians: musicians have made the AI generation possible, it pushes them aside, and they don't get a cent from it.
Perhaps legal, but not in the spirit of fairness.
Much like visual artists, graphic designers, and soon authors.
And truth be told - musicians are an excellent example of how a hostile industry can fuck over the working man long before AI enters the equation. Established musicians needed to be managed, and that management typically came from a label that would sign you if you weren't rich enough to go indie. Then, you're effectively contracted talent that has little to no ownership of your own life work. You are expected to provide all of the emotional gravitas while retaining none of the artistic control, and it only got worse in a digital era that further highlighted how valueless physical music had become. Artists had to make their money off tours and merch sales while simultaneously honoring arbitrary record deals demanding X amount of music over Y period of time.
Yes it does, your own AI generated pieces are based in your prompt, not purely generated by AI. So they are meaningful to you, but still worthless to others. They can generate their own AI shit. What I mean is that AI content is useful if you had a hand shaping it.
Pretty much the past 5 years of model finetuning has been predicated on the idea that AI cannot reliably steer itself in the right direction. Models perform better in almost every way when you seed them with additional, non-unique and human-generated training data.
One year away, two max.
I only ever required attribution
didn't stop OpenAI, Microsoft, Google and Amazon shitting all over even that minimal condition
parasites
Or FLACs, if you can get them, which you can on bandcamp. I've got a terabyte of music in my offline playlist. It takes a few months to cycle through everything, then top it up with any new discoveries and latest releases before the next cycle starts.