Record labels gang up to sue AI music generator duo into utter oblivion Big name record labels are together suing two AI startups for allegedly training their music-generating models on copyrighted tracks without permission, resulting in the software emitting audio that rips off commercial work. In legal action coordinated by industry body the Recording Industry Association of America, Sony, …
"Yesterday
All my[their] troubles seemed so far away"
I have no idea if "Yesterday" is one of the pieces of music in contention, but, McCartney's recollection about the origins of the melody, and his efforts to check that he was not plagarising someone else's work is equally applicable in the case of "AI" halucination and the checking/acknowledgement (or rather, the lack of that) on the part of these services...
Whilst greedy records execs are definitely Asshats. Anyone involved in setting up an AI to produce new music by scraping existing music and not paying a damn cent to the artists whose work they're stealing are even bigger Asshats!
The greedy execs might steal the lions share of the revenue generated by an artist, but the artist still gets paid. With these AI scum, the artist gets nothing, and can kiss goodbye to any future revenue as well. So screw them, I hope the Music industry wins this one and wins big...
Protect your local artists, and go and see a show...
" but the artist still gets paid"
I think you need to review how that "getting paid" works for the majority of artists that don't have major commercial successes. Hint : they are still in debt.
I think a more accurate analogy is a blood sucking leach that is an independant species vs a blood sucking leach that is a non-symbiotic parasite.
It seems to me, my memory being a little foggy lately, that somebody somewhere talked about an AI-based system that claimed to be capable of producing all possible tunes.
And that it was a finite and calculable number. A few trillion or something. Doubtless based on a few rules about scales, notes, duration, etc and excluding non-Western musical scales.
(Probably just my memory is hallucinating, but neither you nor I can't be sure.)
Googling of course lists the "top 7", "top 10", "top 160" AI music makers. And nothing about an all-possible tunes algorithm.
Oh wait, here is a reference: 2405.15103 which explores the difference between white noise and music. It includes a reference to "all possible simple 8 note melodies." Enjoy the rabbit hole. I'm going to bed.
The problem with that is that it's similar to taking a dictionary, making all possible sets of eight words, then claiming that no book can be copyrighted anymore because I've copyrighted every sequence of words. It's been clear from the start that copyright is based on originality of the whole thing, not tiny parts. I can't write "The world was" (copyright me, you have to pay me to use it), but neither can I invalidate your copyright even though you said "the world was" and I said it earlier.
A lot of music copyright cases are unconvincing to me, because they focus on exactly this kind of tiny thing and try to claim ownership of it. Usually, when I read about one musician suing another (or usually a company or family rather than the musician themselves), I think they're making a stupid argument and deserve to lose. This case does not follow that rule. This is not someone with a similar chorus. This is someone taking whole sets of files, in their entirety, and using them wholesale to make money. If these companies had their models stolen, I would not be permitted to use them at my leisure because they are copyrighted. That applies equally well to their training data. I often disagree with RIAA actions and statements, but once again, this is an exception.
> It's been clear from the start that copyright is based on originality of the whole thing, not tiny parts.
That "whole thing" vs "tiny parts" can be far from straightforward, however, as became abundantly clear with the sampling boom of the 80s.
To take an interesting example, Janet Jackson very effectively samples Joni Mitchell's Big Yellow Taxi in her 1997 hit Got 'til It's Gone. Joni Mitchell is credited as a songwriter (as she should be - the sample is the hook), so I assume she receives royalties on it. (In fact Jackson sought permission from Mitchell, and has apparently said that she wouldn't have released the track if Mitchell hadn't liked it - which it turns out she did.) Then Jackson was successfully sued by the singer Des'ree for similarity to her track Feel So High - justifiably so; Got 'til It's Gone has the same chord sequence and almost identical melody in the verse as Des'ree's song.
I upvoted, because I agree with the ethics you're espousing, but this statement isn't right :
"copyright is based on originality of the whole thing, not tiny parts".
in that the _legal_ definition of "song copyright" is those three notes.
Case after case after case used that. THEN we get to Blurred Lines.
Sometimes, but many such cases have failed because the thing that was "copied" was so small or so common that it did not deserve protection. Each time, it's just an experiment to see what the jury thinks. There are also distinctions depending on what exactly was copied. For example, a sample of a recording is considered copyrighted, while a note is not. While they might seem similar as they take about as much time (fractions of a second) to play, they should be and are treated differently. What I was trying to point out is that releasing a pattern of notes will not invalidate copyright on a subsequent song that also plays that pattern.
Also I'd add that a sequence of notes isn't the creative work. You've got your billions of combinations of notes, but that doesn't include the speed and rhythim in which they're played. Then you have to add in the different kinds of instruments you might use to play those notes - plus the words you put against them. Or don't. Then to complicate things still further you've got simple sequences of notes that you might choose to repeat, or not. Add all of this in and you're in to uncountably huge numbers of different combinations.
Even for a really simple example, take a covers band like Hayseed Dixie. They started off doing bluegrass covers of AC DC, then branched out into lots of other songs. I assume they'll be paying credits to the songwriters from their CDs - and it's clear the songs are the same - but just changing the instruments and rhythm also changes the song.
Obviously there's an argument about creativity here, and then a different argument about what the law says and how that will be interpreted by different courts in different cases.
I imagine RIAA is just squashing these companies to gain control of them. There is definitely a possibility that the AI could be trained with modern (copyright) material- although the only good AI music I’ve heard has been 1940s throwbacks- but how can someone prove ownership of elements used in the creation of the derived work?
Imagine the analogous example in the vast amount of AI generated images trained on, no doubt, the enormous corpus of copyright images on Facebook, insta, Pinterest etc. Proving an individual AI image was created by a model trained on stuff including pages from my sketchbook that I uploaded to instagram is nigh on impossible.
Maybe RIAA want legislation that dictates training materials are logged and catalogued etc.
All musicians learn their craft by studying - and playing - existing pieces. They go on to improvise around those pieces, adding their own interpretation. You could argue that they have bought the sheet music, and so the music industry gets its cut, but you know that in reality they mostly haven't, and, anyway, that doesn't apply to those who learn a piece by ear.
At a high level, the process isn't so different from training an AI, the difference being that the AI can quickly learn to turn out dross that is indistinguishable from professional pay-for dross, and the dross merchants don't like it.
Take away the ability to train the model, their reasoning goes, and you take away the threat to their revenue stream. Further, if we acquire the technology, we can train it on our back catalogue, and finally rid ourselves of those needy, whiny, demanding musicians, replacing the "artistic" temperament with a machine, and nobody will even tell the difference. But first things first, we mustn't let just anyone do this - quick, call the lawyers!
In truth, though, I don't think they need worry too much. There will always be demand for live music, together with the associated merchandising opportunities - nobody is going to buy t-shirts featuring an AI, or go dance in a data centre.
"In truth, though, I don't think they need worry too much. There will always be demand for live music, together with the associated merchandising opportunities - nobody is going to buy t-shirts featuring an AI, or go dance in a data centre."
A large number of Hatsune Miku fans would disagree with you
is of late AI generated content is 100x better than the shite being written by "real" people.
Also, a fucktonne of it is very funny. There are a slew of brilliant songs on YouTube that really are as good as the real thing ... Google "**** off that ain't my job" as a starting point.
"[...] Our technology is transformative [...]
I suspect Shulman is missing the point here. He seems to be arguing that there is no copyright infringement because the output of the algorithm is original. Whereas, if I understand correctly, the lawsuit is targeting the training side of the process, i.e. you have no license to use a copyrighted song to train a LLM.
There are people who argue that because humans can learn from listening to a song, the LLM should be allowed to do the same. But the lawsuit isn't against the LLM; it's against its developers. Who are not listening to songs and learning. They are doing a different thing, a new thing, which so far has not been tested in court. It's up in the air whether it's allowed or not.
I don't know who gave "the record labels" jurisdiction over music but I think they're overstepping boundaries somewhat. They seem to think that AI owes them but the artists they 'own' owes no debt to literally thousands of yeas of musical output.
I've got news for them. Practically everything that's ever been composed is a derivative work of something that came before. One simple example that's easy to find is music that copies the chord progression of Pachabel's "Canon in D" but there's plenty of others if you want to look for them. Musicians learn the trade by learning songs and styles composed by others, they then incorporate those into their own compositions. AI does exactly the same; its main differentiator is its speed of both learning and composition.
There is room for creativity in any work but the line has to be drawn somewhere. Disney, for example, has made a fortune and then some by appropriating folk tales and dressing them up in contemporary themes. Fair enough, but then they want to 'own' the folk tale and set lawyers on anyone who dares to infringe on 'their' property. That's going too far. Same with music -- show me a truly original composition (that's not a gimmick like Cage's silence...) and I'll be very surprised.
Tricky.
On the one had, RIAA are truly asses and I hate to side with them no matter what.
Second hand, of course there's the concern about AI taking everyone's lunch. This is a real issue, I'm just not sure current regulations apply to this at all, RIAA is of course going to try to use AI as a new excuse to take away the public domain and fair use. Perhaps new regulations are needed but I'm not sure what precisely you would regulate. (I mean, you regulate AI using existing works for training... but laws need more precision than this.)
Third hand, as the Recording Industry Ass. of America do, they are completely ignoring free use and how copyright law actually works. Artists can sample with no issue whether RIAA likes it or not (although usually when it's a full-on sample they acknowledge the source out of artistic respect, and they usually do license these now viewing it as a slight "grey area".). Companies routinely (for TV shows) make "sound alikes" that sound VERY similar to an existing song but isn't that song to avoid having to license it. I, as a human, can read the entire collection of sci-fi and then write new sci-fi that would undoubtedly end up with some themes and bits of story line that were already in some other sci-fi at some point. For that matter, I could straight up write a new sci-fi in the exact style of an existing author, and as long as I don't try to pass it off as their work I'm still in the clear. Copyright doesn't cover the general style or theme, it covers copying the exact text (or audio or video). I, as a human, could listen to the last 100 yeas of music and write up new songs, that would undoubtedly have some musical themes, sets of chords, etc. from existing songs and the RIAA can't say dick about it. But if an AI does it they think they can.
'AI' is but the latest technology to sow fear in the denizens of the copyright rentier luxury chicken coop.
Prowling foxes have had a cumulative effect by progressively diminishing the size of the coop in which supposed copyright can be protected. During my lifetime, home recording off-air or from a disc caused anxiety for purveyors of analogue recorded music. Worry increased when cassette players displaced bulky reel to reel tape recorders, and became ubiquitous. Identical concerns arose when analogue 'video content' could be home-recorded, or purveyed on the 'black market'. In the light of things to come, unstoppable erosion of 'rentier' markets was negligible: no worse than 'evaporation' of stock in a supermarket, and easily accommodated within a monopoly pricing regimen supported by giving an impression of the analogue products being desirable luxury goods expensive to produce.
Meanwhile, the increasingly cheap and very efficient photocopier was projected to cause mayhem for publishers of academic paper-based materials. Restrictions imposed, and supposedly enforced by librarians, were irksome nonsense involving filling in forms. These were circumvented whenever possible by use of less zealously guarded copiers located outside library confines. Librarians themselves were no more enthusiastic than their 'readers', but obliged to be seen to enforce the copyright nonsense because otherwise their institutions would be hammered by lawyers.
The introduction of digital formats for music, static images, video, and text, diminished the size of the 'coop' considerably. Initially, CDs and DVDs, were arcane technology, but soon the hugely expanded market for office and home computers, these equipped with read/write CD/DVD attachments, demonstrated the easy and perfect reproducibility of 'content' abstracted from digital media, plus the cheapness of storage media.
Although a considerable amount of material allegedly owned by the rentiers did escape into the wild, 'bottom lines' were not affected in the slightest. Purveyors of digitised 'content' had many means, these including creative accounting, for making up any difference (if it should actually exist).
The arrival of the Internet in most offices and homes changed the game considerably. Not only was academic and entertainment 'content' beginning to be free for all who knew where to look, but so also was computational software. Although the writing was on the wall — at least in small text — the industries involved chose to ignore it and to double down on their existing modes of business whilst pursuing legal remedies to protect their specious 'rights'. 'Catch a mole' became the game. Rentiers attempted to cast wide nets to entrap wicked people who objected to monopoly distortion of open markets. The industries attempted technological solutions (e.g. DVD encryption) which promptly were overturned.
Setting aside so-called 'AI', which is promising much fun, big battles at present are waged over unauthorised streaming of sports and of films from online purveyors such as the Premier League, Netflix, and HBO: IPTV is the name of that game. For example, in Italy, drastic measures have been taken to prevent the 'intellectual property' of football leagues leaking out through illegal vendors offering much lower prices (aka competition). A recent article in “TF” suggests these responses are doomed to fail. At present, 'content'; encrypted by means such as “Widevine” is intercepted and 'rebroadcast' not encrypted via the Internet for the benefit of subscribers to 'pirate' IPTV services. This mode of copyright circumvention is, in principle, defeated by a tight regimen of on-the-fly pirate-site blocking. Under pressure/persuasion from the Premier League, the next British government will introduce legislation similar to that in Italy.
However, according to the TF article, no longer is in-depth knowledge required for end-users of Pirate IPTV to instead arrange their own decryption. There is at least one online repository, seemingly immune to take-down, offering open source decryption tools suitable for this purpose. Anyone with access to a bright teenage boy can have this quickly set up. More feared by the rentiers, is the prospect of these tools being seamlessly incorporated into free IPTV agents, as on Kodi.
One's stance on the propriety of viewing arbitrarily priced digital 'content' either at reduced cost from an intermediary or directly for nothing at all, is irrelevant to how matters shall transpire. Law is being broken. It will be increasingly so. For large tracts of the planet, that is the most sensible option (e.g. textbooks for Africa, academic papers for India, and entertainment for the impecunious elsewhere). The fate of 'intellectual property' law is collapse; historically, that is the way all inherently unenforceable laws end. Whilst retained, the law fittingly is mocked. In this instance, copyright law is an unwieldy edifice of Byzantine complexity, and perhaps understood in its entirety by only a handful of people (c.f. Sherlock Holmes brother Mycroft and other people who understand the 'Balkan problem').
The challenge from AI — a technology not containable in all legal jurisdictions — may be the last stray. The chicken coop will collapse. Its occupants, and the supposedly 'creative' people on whom they feed, must find other means to raise income. Of course, the truly creative shall, as always they did in the past.
To give you an insight into how these rentiers think look at the copying of sheet music. This is an absolute no-no for performance but the prohibition extends to copying pages to make page turns easier (a typical performance piece is a dog's breakfast of bits of paper -- assuming that the musician is still using printed music).
Fortunately the days of making insignificant editorial changes to a long out of copyright piece so that it can be copyrighted anew by 'the usual suspects' are behind us for the most part. It was a good run while it lasted but now there are excellent out of copyright editions of classic works that can be used to keep contemporary prices down. (I've got nothing against people earning a living -- we all have to do that -- but if you look at some of the list prices for sheet music 'back in the day' it was a major ripoff (even allowing for the extra costs of pre-computer engraving of music).
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
