If AI replaces musicians, does the entire plugin industry die with them?

[Tiles]

Do you have a court ruling or any solid legal source that confirms that claim in the way you're stating it?

And even if you assume there are contested parts of training data, that still doesn't justify the blanket statement that "the data was illegally downloaded". Most large-scale training datasets are built from a mix of licensed, public domain, and publicly accessible sources, and the legal situation varies by jurisdiction and is still being actively discussed.

Turning that into a universal claim doesn't really follow.

[BONES]

something like art rock places a much greater emphasis on the artistic side.

I don't know that even that is true. Given that we all know what Art Rock sounds like, it is as formulaic as any genre, with no more or less freedom to create art. I think there is art and product in every genre. If your music is targeting a genre or a scene, be it the Art Rock crowd, Jazz wankers or whoever, then you're making product. OTOH, if you're making music for the sake of art, even if it falls into a narrow, defined genre, it can still be art. It's purely down to the creator's intent and most of the time only the creator can know what they are trying to do. If you want to make art, make art. If you want to grow your audience, make product.

Maybe you don't know how Suno got access to the massive amount of audio material they illegally downloaded and used for generating their LLM datasets. Unless they had a trillion or so dollars to pay to each and every content owner for the tracks .

And you do? Did they hack any websites to get behind paywalls? Did they use BitTorrent sites to obtain pirated material? Or did they do what we all do and skim data from freely available, public sites? If they used a legitimate account to gain access to Spotify's library, for example, then they were doing nothing illegal. If they used the millions of songs from Bandcamp, where they are freely available for anyone to listen to, then they have done nothing illegal.

[Tiles]

something like art rock places a much greater emphasis on the artistic side.

I don't know that even that is true. Given that we all know what Art Rock sounds like, it is as formulaic as any genre, with no more or less freedom to create art. I think there is art and product in every genre. If your music is targeting a genre or a scene, be it the Art Rock crowd, Jazz wankers or whoever, then you're making product. OTOH, if you're making music for the sake of art, even if it falls into a narrow, defined genre, it can still be art. It's purely down to the creator's intent and most of the time only the creator can know what they are trying to do. If you want to make art, make art. If you want to grow your audience, make product.

Fair enough. I think it's both. Art and product are not mutually exclusive, and most music contains a bit of each.

[TechHaus]

Does anyone have any good sounding suno tracks they can share? (good audio quality - not "performance")

[slackhead]

If they used the millions of songs from Bandcamp, where they are freely available for anyone to listen to, then they have done nothing illegal.

Bandcamp's terms of service which any user of the site agree to abide with explicitly prohibit the use of any of its songs to be scraped or otherwise used to train an ai model.

[ksandvik]

Do you have a court ruling or any solid legal source that confirms that claim in the way you're stating it?

And even if you assume there are contested parts of training data, that still doesn't justify the blanket statement that "the data was illegally downloaded". Most large-scale training datasets are built from a mix of licensed, public domain, and publicly accessible sources, and the legal situation varies by jurisdiction and is still being actively discussed.

Turning that into a universal claim doesn't really follow.

Do you know how the US copyright rules work? Do you know why various artists that use samples have other pay royalties to the originators?

[TechHaus]

OMG bracing for this one....


Is AI Taking over Pop Music?

[Touch The Universe]

That's incoherent. A person listening to a song is different than an ai "listening". There input generates output that can be applied in ways mere "feelings" or enjoyment can't.

[Grizzellda]

I doubt she(?) sees us as worthy of her talents. But she(?) has stuff on YouTube if you're interested. It was nothing like what I had expected, given the way she(?) carries on.

Are you questioning her sexuality?? Ha ha ha ...good one BONES! but I actually think she is a real female, I have called her a ''Gloria Steinem'' type of feminist, and if that is true, well then good for her!!!

Well, she seems older, prolly has used ''substances''.

Is she even posting lately? Don't think she wants to talk to me. Purely and simply... I have outcooled her!!!

Welcome to Urantia, jancivil!!!

[Tiles]

Do you have a court ruling or any solid legal source that confirms that claim in the way you're stating it?

And even if you assume there are contested parts of training data, that still doesn't justify the blanket statement that "the data was illegally downloaded". Most large-scale training datasets are built from a mix of licensed, public domain, and publicly accessible sources, and the legal situation varies by jurisdiction and is still being actively discussed.

Turning that into a universal claim doesn't really follow.

Do you know how the US copyright rules work? Do you know why various artists that use samples have other pay royalties to the originators?

Yes, I do.

And you're mixing up two different things.

Sampling in music is the direct use of a recognizable portion of a copyrighted sound recording. That's why it triggers licensing and royalties: you're literally taking parts of the original audio and reusing them. You use the samples.

Training a model is not that. It does not store or replay specific tracks as samples. It learns statistical relationships from large datasets and generates new outputs based on those patterns. It is not about reusing specific material, but about learning from many examples at scale.

The core question here is pretty simple: why should "learning from existing works" suddenly become illegal just because the system doing it is not human?

Either way, the analogy doesn't really hold. The legal mechanism you're referring to is about reuse of identifiable material, not pattern learning at scale.

Different mechanisms, different legal questions.

[ksandvik]

The patterns learned are not learned, rather used. As such you are dealing with samples, whether they are mixed or not. If this was midi information reshuffled I would agree, but Suno did not steal midi, rather music under copyright rules.

[ksandvik]

In addition Suno had to agree to this theft -- as they had other sign contracts with various record labels and are chasing other contracts. Also there are lawsuits against Suno from indendent artists. This is a messy state.

At least Suno produced music is not copyright-protected, which is a good thing.

[Grizzellda]

Yes, I do.

And you're mixing up two different things.

Training a model is not that. It does not store or replay specific tracks as samples. It learns statistical relationships from large datasets and generates new outputs based on those patterns. It is not about reusing specific material, but about learning from many examples at scale.

It is basically gonna compute / analyze ''stuff '' or ''data'' (see Negativeland's song...''content'', or something as such. The song, if I recall, was essentially saying that all we are in this modern age...is content.

That is all we are. Content to be infinitely REMIXED, and ultimately regurgitated, alright???

Hey D.J.!!! fix it in the mix, ALRIGHT???

[ksandvik]

Exhibit three: there are now AI tools that could trace back actual songs used for the AI generated music. So there's a direct 1:1 reference to original track and the synthesized Suno AI track.

Note: I'm not against remixing and such, that's fine and has happened all the time -- the difference is that the originators were both acknowledged and paid. This is not what Suno does with the stolen property.

[koalaboy]

The patterns learned are not learned, rather used. As such you are dealing with samples, whether they are mixed or not. If this was midi information reshuffled I would agree, but Suno did not steal midi, rather music under copyright rules.

I can't speak for Suno, but that's not how AI works (or can do), and you're focusing on one or two companies rather than the technology. AI is absolutely capable of synthesizing every waveform, and putting them together into 'learned' patterns. Technically it's stochastic maths with attention and other fancy features, but fundamentally it's learning, just like you or I (or as representative as we can explain).

In the same way that the infamous "million monkeys with a million typewriters" could eventually turn out the works of Shakespeare, or the Library of Babel contains everything that could be written, AI is effectively accessing smaller versions, with learned patterns on how to take different parts and put them together (you know, like a human learns to take a beat and mix it with a melody), and then getting a result.

You can keep believing that they just 'steal' samples if you want (and older/lesser applications of AI may do that) but that's not how AI models work, and you'd be much better off learning about the actual theory than constantly blaming AI for something that it isn't.