Streaming Services Urged To Clamp Down on AI-Generated Music (ft.com) 108
Universal Music Group has told streaming platforms, including Spotify and Apple, to block artificial intelligence services from scraping melodies and lyrics from their copyrighted songs, according to emails viewed by the Financial Times. From the report: UMG, which controls about a third of the global music market, has become increasingly concerned about AI bots using their songs to train themselves to churn out music that sounds like popular artists. AI-generated songs have been popping up on streaming services and UMG has been sending takedown requests "left and right," said a person familiar with the matter. The company is asking streaming companies to cut off access to their music catalogue for developers using it to train AI technology. "We will not hesitate to take steps to protect our rights and those of our artists," UMG wrote to online platforms in March, in emails viewed by the FT. "This next generation of technology poses significant issues," said a person close to the situation. "Much of [generative AI] is trained on popular music. You could say: compose a song that has the lyrics to be like Taylor Swift, but the vocals to be in the style of Bruno Mars, but I want the theme to be more Harry Styles. The output you get is due to the fact the AI has been trained on those artists' intellectual property."
So? (Score:5, Insightful)
The question is: What music did Taylor Swift, Bruno Mars, Harry Styles "train" on?
Surely they didn't grow up and become famous without ever listening to music.
Re: (Score:1)
oh, a rude anon. how... worthwhile an opinion.
If you're going to be rude, at least do it without hiding behind being anonymous coward.
Re: (Score:2)
Whether you know their music or not, Taylor Swift, Bruno Mars, and Harry Styles are legitimate musicians, by any standard.
Re: (Score:1)
Re: (Score:3)
That definition is silly. Really, really silly. You don't even understand the basic difference between "musician" and "composer".
Re: (Score:2, Interesting)
Re: (Score:2)
Your goalposts are far removed. You want to redefine musicians to your use case, but it's simply not the definition the world understands. You're on an island.
Oxford definition:
Musician: a person who plays a musical instrument or writes music, especially as a job
Composer: a person who writes music, especially classical music
So a pure cover band are musicians, but not composers, despite not writing anything themselves.
Re: (Score:2)
Re:So? (Score:5, Interesting)
If anything the law should be moving in the opposite direction. Artists already have the right to do cover versions of any song they want, only having to pay a "mechanical" licence fee to one of the orgs that collects them, under US law. Sampling is not covered though, so you can do your own near identical version of the same song, but not take a bit of the original song and transform it into something new without getting the permission of the copyright holder and paying whatever they demand.
You could argue that AI is somewhat similar to sampling. The same data is ingested, but used to train an AI that produces output, rather than being manually edited into a new song. Which seems to suggest that it is probably copyright infringement, if not licenced.
When it comes to sound-alike songs, it usually boils down to proving that the artist knew of the other song. I expect lawyers will be demanding to scan hard drives for training data in the near future.
There has already been a bit of a backlash against some AI generated music, from other artists. There are AI rappers who are controlled entirely by corporations. They couldn't be less authentic, and are accused of appropriating hip hop culture.
Re: (Score:3)
There may be a few highly specific parts of a song that are unique, but permutations of those can side step the issue by being close, but not exact. Most people are familiar with Yakety Sax,
Re: (Score:2)
There may be a few highly specific parts of a song that are unique, but permutations of those can side step the issue by being close, but not exact.
Yeah like Marvin Gaye vs. Blurred Lines.
Ohwait...
Re: (Score:3)
The problem there is that some Judges who try these cases appear to be insane. Consider silence. There are basically an infinite number of examples of pre-existing things with exactly the same cord progression and lyrics as "4:33" by John Cage. That's because there are literally no lyrics and no chords. It's just silence for 4 minutes and 33 seconds sampled from natural silence. Nevertheless, people have been constructively decided against in court (constructively in the sense that they actually settled, bu
Re: (Score:1)
That's not the question. The questions are:
Do AI researchers have a license that allows them to transform the work into an AI model?
Is transforming the work in this way transformative enough to count as fair use?
What happens in a human mind when they listen to music is irrelevant in 2023. Maybe if you could download a human mind and fix it into new media it would matter, but you can't.
Unregulatable. Cat's out of the bag, givvit up. (Score:5, Insightful)
Human artist brains borrow patterns also, as others pointed out. The Beatles borrowed heavily from the patterns and styling of Buddy Holly, Everly Brothers, and Chuck Berry, for example.
So why is a human allowed to borrow general patterns but not AI? And who can tell the difference? True, one can probably detect an AI-made sound file itself at this time, but it's also possible for a human to generate AI variations, and then manually perform and record the one they like.
My point is that human composition pattern borrowing itself is probably indistinguishable from bot borrowing. And even if a way was found to detect bot swipes from human swipes, one can train the bot on the detector itself to get around it.
Re: (Score:3)
Human artist brains borrow patterns also, as others pointed out. The Beatles borrowed heavily from the patterns and styling of Buddy Holly, Everly Brothers, and Chuck Berry, for example.
So why is a human allowed to borrow general patterns but not AI? And who can tell the difference? True, one can probably detect an AI-made sound file itself at this time, but it's also possible for a human to generate AI variations, and then manually perform and record the one they like.
My point is that human composition pattern borrowing itself is probably indistinguishable from bot borrowing. And even if a way was found to detect bot swipes from human swipes, one can train the bot on the detector itself to get around it.
This in a nutshell, TBH... ALL OF IT
Re: (Score:2)
Sure, but the AI researcher didn't have a license for the original work that granted them the rights to "perform it" for AI.
Re: (Score:2)
Sure, but the AI researcher didn't have a license for the original work that granted them the rights to "perform it" for AI.
This is the part of a larger conversation that foreshadows 'augmented humans', which are only a few years away. Enhancing our own capabilities with cybernetic agents is not something which will be easy to regulate. The only difference here is that the AI agent in its current form is external to the humans.
Just because we have ears doesn't mean we're not allowed to use ear enhancers.
And just because we have brains doesn't imply that we're not allowed to have knowledge enhancer agents to help [some among
Re: (Score:2)
If cybernetics fix the work into a tangible medium, like today's AI does, then it's still going to a problem.
Re: (Score:2, Interesting)
My point is that human composition pattern borrowing itself is probably indistinguishable from bot borrowing.
No, because these bots can only interpolate, whereas humans can extrapolate (not to mention experiment to find elements that are not derivative at all).
Re:Unregulatable. Cat's out of the bag, givvit up. (Score:4, Insightful)
That sounds suspiciously like the arguments for why computers will never beat human grandmasters at chess. Or the later arguments for why, Ok, chess sure, but never go. Now, someone did recently beat a top-ranked go-playing AI, but it's telling how he did it: He used another computer to find a flaw in the play style!
If music were some infinite canvas maybe I could buy that argument. But we're not talking about an infinite canvas, we are talking about a large, but finite set that is greatly constricted by the poorly defined rules of what is copyrightable. Basically, every piece of copyrightable music can be mechanically reproduced with no Artificial Intelligence at all, just brute force. Throw in some rules to eliminate combinations that are unpleasant to humans and you greatly reduce the set. Copyright, and intellectual property in general, are one of those areas where the legal concepts just do not match reality.
So interpolation and training are not even required. Just producing random combinations and checking them against some rules that eliminate unpleasant sound combinations is enough. Modern AI generation techniques just speed up the process.
Re: (Score:2)
Throw in some rules to eliminate combinations that are unpleasant to humans and you greatly reduce the set.
Is Beethoven's music pleasant or not? [youtube.com] Answer!
So interpolation and training are not even required. Just producing random combinations and checking them against some rules that eliminate unpleasant sound combinations is enough.
Beyond your (false) idea that music need to be pleasant, I'd like to see you even come up with rules for eliminating unpleasant sound combinations. I think you'll find it's not as simple as you claim. It's even harder when the music involved has symantec meaning.
Re: (Score:2)
Beethoven is probably not a good example, as someone who composed in a clearly recognizable style.
Beyond your (false) idea that music need to be pleasant,
I don't think that music has to be pleasant. However, this is in the context of copyright complaints from large producers of music. As a general rule, their commercial interests are going to be over music that people want to listen to. Painful cacophonous noise typically has poor commercial appeal.j
I'd like to see you even come up with rules for eliminating unpleasant sound combinations. I think you'll find it's not as simple as you claim.
I don't need to, and I never claimed it was simple. You're barking up the wrong tree if you think it can not be do
Re: (Score:2)
I don't need to, and I never claimed it was simple. You're barking up the wrong tree if you think it can not be done
I don't think the rules can be codified using a neural network. We'll need something more complex than that.
Well, if you compose some music that has symantec meaning, make a video.
That's easy, just look at Beethoven's ninth symphony. It even has words for part of it. Joseph Kerman also pointed out that some of Beethoven's string quartets have meaning, although it's not clear what. Michael Tilson Thomas did the same with Beethoven's fifth symphony.
Re: (Score:2)
I don't think the rules can be codified using a neural network. We'll need something more complex than that.
Is this a religious or "spiritual" thing? It's sounding a little that way, just want to be sure.
That's easy, just look at Beethoven's ninth symphony. It even has words for part of it. Joseph Kerman also pointed out that some of Beethoven's string quartets have meaning, although it's not clear what. Michael Tilson Thomas did the same with Beethoven's fifth symphony.
I think you missed the joke about the video going "VIRAL".
Re: (Score:1)
What's "extrapolating" a tune? Could you give an example?
GPT is perfectly capable of adding a degree of randomness into results. In fact, randomly going with occasional low-grade node "matches" is part of what has made it so powerful. Prior techniques tended to maximize all nodes.
Re: (Score:2)
Re: (Score:1)
Sorry, I just don't know what means in terms of melodies. I'd like a specific example, if you don't mind.
I'll give it a try: suppose the melody goes down-up-up, down-up-up, and thus seems to be slowly climbing. Extrapolation of that pattern could be 3+ sets of down-up-up.
However, that's pretty trivial for AI to do, assuming the pattern generation is based on offsets instead of absolute notes, which would be an obvious course of action, or at least an obvious variation to try, in terms of encoding pattern st
Re: (Score:1)
You can turn their own words on themselves:
> "You could say: compose a song that has the lyrics to be like Taylor Swift, but the vocals to be in the style of Bruno Mars, but I want the theme to be more Harry Styles. The output you get is due to the fact the AI has been trained on those artists' intellectual property."
OK, so, if **I** compose a song that has lyrics like Taylor Swift, vocal style of Bruno Mars, and theme more Harry Styles, having learned all those things from listening to those guys, the
Re: (Score:2)
What if say Paul McCartney did that, without assistance from machines? Borrowing and mixing styles has always been in human composing.
Perhaps a law can be written such that if one is "caught asking a bot" to mirror styles, they can be sued. But otherwise, nobody will know.
Actually, the Ghost Busters theme writers were sued for intentionally copying the style of "I Want a New Drug", but that's because the producer explicitly asked it to be parroted. If that intent request never got out, nothing would probabl
Re: (Score:1)
PunGPT
Re: (Score:2)
IANAL, but the problem isn't at time of performance of a new work. The problem happens when the bot listens to the work, as AI researchers don't have a license that explicitly allows it, and we don't know yet if it falls under fair use.
Re: (Score:1)
A law/ruling that says "auto-pattern-learning bots can't listen to any song without obtaining proper permission" is a rather heavy-handed regulation. I doubt it would fly in the US; but who knows, lawmakers and judges do random-seeming stuff sometimes.
Re: (Score:2)
That's not the question. The questions are:
Do AI researchers have a license that allows them to transform the work into an AI model?
Is transforming the work in this way transformative enough to count as fair use?
What happens in a human mind when they listen to music is irrelevant in 2023. Maybe if you could download a human mind and fix it into new media it would matter, but you can't.
Nothing a little formaldehyde won't fix. You tangible medium, you.
Re: (Score:1)
Re: (Score:3, Informative)
The fundamental difference is that they are human beings, filtering music through their human experience, whereas the output of AI could be transformative, but never creative.
Copyright protects human creative expression, not the output of an algorithm. Even if a human "guides" the algorithm, the creative and expressive potential is so limited as to be virtually meaningless. With a human being, similarities to other musicians are accidental, and never exact copies, but with AI, the express intent is to
Re:So? (Score:4, Interesting)
I'm pretty sure you can't copyright a "style". If you could, there'd be no cover bands, no Elvis impersonators, heck, no 10 second songs etc on Youtube.
You can already pay a cheaper musician to do a cover of a popular song and make it sound as close as they can, and that wouldn't break copyright as long as they paid the songwriter license. Copyright is about a copy that sounds exactly like the original. As long as you can show it just "sounds like" and isn't like an MP3 copy - you're fine.
Many of the "sounds like" lawsuits I think are because they didn't license what they were sampling or copying, but I also think those turn out to be rare as you certainly don't win all the time when it's down to like 4 notes in a row.
Re: (Score:2)
Isn't there a program that created profiles of music that will "listen" to a song and tell the record execs if it is likely to be a hit where they encouraged musicians to write towards the algorithm? This seems like protectionism and fear for the music execs.
Re: (Score:2)
Secret Sauce (Score:5, Informative)
Taylor Swift is just a pretty-ish person they can slap on the box.
Re: (Score:2)
The music industry doesn't want everyone to find out that all pop music is and has been AI generated for years now.
Taylor Swift is just a pretty-ish person they can slap on the box.
Some evidence or time to put the tin-hat on?
Re: (Score:2)
I doubt she will even look in your direction.
If she wasn't somewhat pretty after being covered in make up, she would be a nobody,
Re: (Score:1)
Some evidence or time to put the tin-hat on?
Personally, I think he's wrong and that Taylor Swift writes her own music with maybe some HMS (Hit Music Science) reviews and maybe tweaking before it goes to master. However, I don't think it'll be too long before what he's saying is all too true. Personally, I would love to see most teachers, actors, artists, and musicians all have to get real jobs because AI ran them outta dodge (especially teachers). Guess you'll have to out-compete the AI, despite it's advantages. Kinda reminds me of the late 1990's a
Re: (Score:2)
I think to some extent there's been the formulaic pop, rock, country, whatever boy bands etc since at least the 60s and the Monkeys.
I also think that most IP based on copyright is a false idea of property and was broken during the Napster wars if not all the way back to tape recordings.
I also think that a lot of people are interested in Art because they like the artist - so that will always be like patronage, not commercial. For anyone who is just into "content" - they already have driven down cost to a ver
Re: (Score:1)
they like the artist - so that will always be like patronage, not commercial
Good point. I agree and think you'll probably be proved right on that. I'm also hoping that AI impacts artists the least out of the groups who seem to be impacted. The term "starving artist" is apropos, most of the time. Teachers, on the other hand, should have been already replaced by VHS in the 1980's. Unfortunately, life isn't fair and it'll probably be the other way around.
Re: (Score:2)
I'm not really sure why you think teachers aren't important - I'd argue everyone needs an education in todays environment, far less need any new art created (we have a functionally infinite backlog already created). I'm also not at all convinced that video only is actually effective for learning for all or even most people. If it was, we wouldn't have seen any learning deficit from remote schooling during COVID, but it's widely reported across many different countries.
And the benefit of AI vs Video is poten
Re: (Score:1)
I'm not really sure why you think teachers aren't important
That's okay. There would be no way for you to know that. The reason was that I was beaten by many of my teachers (about 30 incidents in total). In Texas, when I was a kid "corporal punishment" was very common and I'm a strong willed person with my own opinions (not shared by the teachers, usually). The "paddling" started in 3rd grade when I was 9 years old. I went to grades 1-3 in Anchoridge, AK (where they ignored me and didn't teach me to read). Then we moved to Texas and the beatings began (it's still l
Re: (Score:2)
My goodness, that is horrific.
I can certainly see why you would be very skeptical of anything to do with the education system at all.
I think your experience is highly atypical, but maybe MY unconscious bias has colored my worldview to be more education-system-positive.
Re: (Score:1)
I think your experience is highly atypical
Having met others with similar histories, I'm not so sure, but hey, you are entitled to your opinion. Maybe you grew up in a different time and place and things were much better for you. I hope so.
my worldview to be more education-system-positive.
I love learning. I'm not so fond of education systems. However, I hope more people had your experience than mine. Otherwise, all we'd have is a world full of people seeking revenge, sympathy, or escape.
Re: (Score:2)
Yea, that is horrific, and I'm frankly surprised it happened, then again I probably should't be anymore. I had a very different experience with my education, which is why I tend to think education in general is important. I've also tried many times to "self motivate" or "self study" various topics for work (post college). And I just find that while theoretically, I should be able to go onto LinkedIn Learning, or Youtube, or whatever - I often don't know what I don't know, and I don't know how to put a "majo
Re: (Score:1)
I'm frankly surprised it happened
I'm surprised you're surprised since I've met many others with the same experiences. However, I hope you came up in a different time or place.
History is really useful to get exposed to
I was taught very little and what I was exposed to was whitewashed state-history. I took it upon myself to educate myself on history and just picked a starting point at 300BC and went from there. About 80-90 books later, I feel a lot better about my understanding of history. It wasn't easy, though, and took years. I got really bogged down from 900AD-1400AD. The crusad
Re:Secret Sauce (Score:4, Funny)
I don't know about Taylor Swift, but Lorde has been exposed as having been completely engineered using modern tech [youtu.be].
Re: (Score:2)
I know there were some "artists" created by studios (some in every possible way with no talent whatsoever e.g. Milli Vanilli [wikipedia.org]) and I know that many songs by famous artists are written by "ghost" song writers, but I know also that lot's of their success is their hard work and talent, just once famous then they can effort someone else writing for them, but it's "some" and "sometimes", and still making such claims especially calling names (Taylor Swift) warrants providing evidence, like I did above.
Re: (Score:2)
I remember reading a while ago that there were some composers/lyricists in some nordic country that were prolific in writing songs for a ton of pop acts. Basically the big studios are an orchestrated ghost writing operation where they make a person famous and then dictate the lyrics and music they sing. Not sure if that exactly rises to the level of "AI", but it's basically a music factory where you combine a personality under contract that you've made popular via huge marketing budgets with a song that y
Re: (Score:2)
As I answered above, indeed there are "ghost" song writers and some not talented performers created by studios, but it's not a general rule and most of them work hard to be noticed and support their success with talent, especially calling someone by name warrants evidence, hence me asking.
Well, AI will change the world in many ways, interesting times are coming (unfortunately it's also a Chinese curse), and indeed it doesn't warrant a bright future when children biggest dreams is to be a reach and famous si
Re: (Score:2)
Max Martin would like a word with you.
Re:Secret Sauce (Score:5, Funny)
Taylor Swift is just a pretty-ish person they can slap on the box.
Hey keep it PG, there are children listening to this.
Re: (Score:2)
thought some one had copyrighted (Score:5, Interesting)
Re: (Score:3)
You should see the amount of lawyers and court time it took for Happy Birthday to be clawed back into the public domain.
Re: (Score:2)
26 letters of the English alphabet produced millions of discrete books. There are 88 keys on a piano.
You only violate a copyright if you steal a musical idea or a sample. An idea is loosely defined, but basically it has to be enough of an identifiable sequence to be spotted by a listener as having been lifted. Think of the "Ice Ice Baby" or "Land Down Under" court cases.
If your sequence does not have enough character to stay in anybody's mind, somebody else playing it doesn't matter, and it wouldn't pass th
Re: (Score:3)
Nope. You have to prove "substantial similarity" AND you have to prove "access". The plaintiff has to prove that the defendant probably heard the work, and subsequently infringed upon it. A series of notes filed for copyright does not count, as there is no expectation that the defendant could have been exposed to it. This is to allow for the fact that sometimes music is similar through simple independent composition.
Re: (Score:2)
Re: (Score:1)
> If your [brute force] sequence does not have enough character to stay in anybody's mind
A lot lawsuits are by obscure artists whose ideas/works allegedly end up in well-selling recordings. Thus, notoriety doesn't appear to be a requirement to go to court.
Re: (Score:2)
Yep, Schoenberg: https://youtu.be/HDMUraUAAY4?t... [youtu.be]
Many popular songs are blatant ripoffs (Score:2)
If the record labels want to block ChatGPT music that has obvious imitations of existing songs, they open themselves up to the same scrutiny. Imitations have always happened, and you would be shocked how blatant it often is.
Re: (Score:3)
I'm all in favor of that. I prefer folk music. And there's a lot of it that's out of copyright. Most of the basic tunes date back to the Middle Ages in one country or another. The Smithsonian has (had?) a decent collection of uncopyrighted music, and I'm sure it's not the only place.
Re: (Score:2)
I was going to say. You can certainly 'borrow' stylistic elements in music without directly copying in the same way an author can ape the prose of another author using similar grammatical structures, vernacular, and voice without copying anything or even writing on the same subject.
Look at Skiffle music from the 60s or Post Punk from the late 90s - sooo much of it sounds very very similar. There are handfuls of somewhat distinctive vocalists in both genre + period examples, but I could easily find pairs
This may well be the hill labels chose to die on (Score:2, Troll)
Sort of the equivalent of M
Re: (Score:1)
Re: (Score:1)
so you didnt like his analogy but don't disagree with his point? good job modding you fucking retard.
They are not scrapping (Score:2)
They are listening, admiring and remembering, how could they avoid plagiarizing otherwise, if you can't compare your song to the 98263542 other songs?
An interesting issue (Score:5, Insightful)
All music is plagiarism - human artists have influences, they're taught rules, they pay attention to what others are doing. They do not magically birth new things from pure vacuum.
The differences with 'AI' are that it's generally not getting royalties, can churn out fresh creative product, hasn't yet been written so that it stays below a legal threshold of plagiarism, and just overall isn't quite as good as a talented human yet. ...But it's coming, and soon. Those last two points will be addressed, and the owners of the code and database behind the AI generators will be getting paid instead of the RIAA. And the RIAA isn't going to stand for that.
Universal Music Group (Score:2)
“Their music” says it all. How about the music of the artists you represent?
Like GitHub Copilot, but for music (Score:4, Insightful)
This is basically the same as the debate about GitHub Copilot [wikipedia.org], but in the field of music instead of software. Copilot is controversial because it's trained on code with FOSS licenses, and can sometimes emit portions of code that are direct copies of training input, without notifying the user of the FOSS license. Similarly, a music AI might emit lyrics or musical phrases that copy from their training input.
As long as it's just "in the style" of another artist, though, I don't see a problem. "Style" is an ill-defined general impression, not a specific copyrightable element, and it's not uncommon for human-written songs to be stylistically similar to songs by other artists, whether as intentional homage or just because both artists had similar ideas of what sounded good.
I think the key element is whether the AI can recognize the difference between "style" and actual copying. If you ask a human artist to paint a picture of "melting clocks in the style of Salvador Dali", the human artist understands that they shouldn't just copy Dali's painting [wikipedia.org] of the same, even though that image is the best possible response to the prompt given. It's not clear whether current AI models are able to make that distinction. But human artists can paint images of other melting clocks, clearly in reference to Dali but not copying any specific elements, and that's regarded as OK. It should likewise be OK for an AI model to do the same sort of thing, for music as well as for paintings.
Reality (Score:3)
I'm sorry to break it to you but it's technically not possible. Everything possible has already been scraped and even if it hasn't been it's still not possible to prevent AI companies from harvesting music and text while allowing mere mortals to listen to music.
Secondly, three years ago researchers wrote an algorithm which had created [vice.com] all possible melodies. You're a little bit too late.
Re: Reality (Score:2)
Copyright is a TEMPORARY PRIVILEGE (Score:2)
Try "the songs which Universal did not create but which Universal paid for a temporary copyright privilege over".
After a reasonable 5 years, that copyright privilege should end, to promote creativity and the creation of NEW music, by an AI or humans.
Until the law is fair, we - and the AIs - will pirate.
Re: (Score:2)
You're talking about what SHOULD be the case, vs the reality of the law. And unfortunately, companies like Disney have much more money to fight court cases than you or I. That's why perpetual copyright is on the horizon. :(
Re: (Score:2)
Slashdot agreeing to call these things "our copyrighted songs" makes it seems like Universal have some kind of moral right to assert over legitimate use.
Whereas calling them "the songs which Universal has temporary copyright privilege over" makes it clear who has the moral rights here.
Re: (Score:2)
Copyright was not intended to protect the creator, it was to keep progress moving forward by putting thing out in public. It has since been perverted.
That's nice. They can demand. (Score:5, Insightful)
But there are too many ways to access the music that don't involve bulk scraping of streaming services. Cat's out of the bag, baby. Never coming back.
Personally, I don't think you can meaningfully differentiate "creativity" between what AI produces and what average artists produce when "trained" on the same materials. If you want to say that AI isn't "creative" then you need to say why people "are" and AI is "not". And you shouldn't just take a nearly religious approach and simply say that computation can't result in creativity.
If it passes the plagiarism test - i.e. you can't draw the direct line - then I don't know how you declare it somehow "less".
Nobody wrote the first blue's song ever. Like transitionary fossils in evolution, the gradations between genres are mostly imperceptible.
Right now, music produced by AI is... not great but not terrible. People produce worse. And very soon it'll be "good enough" for jingles, light background music, budget-sparse movies, and the like. Great will remain the purview of that top tier of composers and musicians for a long while yet...
HOWEVER - the greats will age out, and those coming up will not have had the opportunities to make a living and become "great". Jingle writers, session musicians, ghost writers, the invisible working stiffs - those jobs won't pay like they used to, and the price is that the next generation of evolved giants will mostly not appear.
Yuck.
Re: (Score:2)
Well, I though you could make an approximate meaningful distinction "between what AI produces and what average artists produce", and I guess that's still true, but then I remembered John Cage and his copyright on silence, and also his "music for goldfish", where the musicians had to play based on goldfish swimming in a tank. So I guess that the AI wouldn't even be more "creative" in ways people didn't like.
Re: (Score:2)
Nobody wrote the first blue's song ever. Like transitionary fossils in evolution, the gradations between genres are mostly imperceptible.
True, but someone did write the first Stairway to Heaven.
More to the point, there are elements that go into blues music such as the pentatonic scale and a 12 bar progression. Someone, somewhere discovered the pentatonic scale or wrote the first song to use a 12 bar progression.
It's impossible not to get philosophical about this because it is an epistemological question, not a fact-baed one: what is the nature of "creativity?"
Is all creativity a combination of existing ideas or does there exist the possibili
Re: (Score:2)
"Stairway to Heaven" is an interesting example. There's a reason the lawsuit over the similarities to "Taurus" lasted as long as it did. There's no question they toured with Spirit in 1968, and heard the song where the similarities are claimed. The courts decided they didn't infringe, but it would be silly to say they weren't influenced.
I'm not saying creativity doesn't exist - I'm saying it's ALWAYS additive. ALWAYS incremental.
But the question at play is whether AI can be creative. My question is whether
What about the opposite case (Score:1)
It seems plausible that pretty soon a model will exist that's fast enough to go ahead and produce vast amounts of every chord progression and rhythym conceivable - or at least enough to get a match on a hit someone else releases a year from now. What then? Who owns that copyright?
Next up: ban radio (Score:2)
Air gap (Score:2)
How can the streaming services stop that? Audio watermarking?
Not sure that would affect AI training much. My guess is the AI generated output, having been abstracted then respecialized, wouldn't definitively include the watermarks in its output.
derivative work (Score:1)