Rick Rubin on AI (& now Graeme Revell, too)

[jancivil]

Now it might kind of resemble Calder's mobiles, weights in space that behave in their various ways owing to gravity...

If I understand what you're saying, music for you as the modernist painters would say is non-objective, and any emotion it invokes in a listener being incidental? If this is the case the dark side to this is novelty for novelty sake. I'm not saying that's what you're doing, but I think that can be/is a danger. Some of the stuff coming out of academia right now is completely ridiculous. If I don't understand what you're saying then…

I don't know the reference as pertains to "the modernist painters... non-objective". If objective there means more or less the same as 'representational', in a sense, yes. I would more think of the word 'subjective' for that. In what I do, if a subject seems to appear I suppose that would be incidental. I just call it music qua music.

I'm not following 'novelty for novelty's sake' as the dark side of my thinking, at all. I think it doesn't follow.

Novelty, for me to seek novelty in music would indicate a couple of things: that I'm making the music in reference to some assessment of other music's newness in some way and that I'm competing with that. In terms of the harmonic vocabulary in my awareness this is pretty much impossible anyway. We experienced Anthony Braxton years and years ago, Cecil Taylor well outside the bounds of convention also a long time ago. Edgard Varese in the 1920s-30s isn't dated in terms of harmonic or sonorous vocab, or in terms of orchestration, eg., Deserts, what is that, ca. 1954... if there is a point for anybody to compete and seek to supersede this in terms strictly of newness in any way this would be completely stupid and pointless.

In terms of rhythmic ideas, I can say that certain combinations of time signature and the ramifications of reordering subdivisions in say Tigran Hamasyan's music can be news specifically to me but the kind of thinking generally isn't. The breaking news to the world in this regard happened in 1913.

The reference to academia is also lost on me. There would have to be examples, and I'm not a music critic in the first place.
I don't reside anywhere near academia, I'm not even very scholastic. I've written two papers in my life and it was fake it til you make it all the way. The second was on a JS Bach suite and it's safe to say nothing new was found in that.

I have managed a couple of things we might be at a loss to locate precedence for, for reasons. But maybe not. I set to work and there's no point but the ideas that come, I don't have time for it to be otherwise.

[vurt]

i guess it's only fair to post examples of the non ai stuff i find, the old fashioned way, a friend suggests it.
this guy deserves lots more views imo.

[jancivil]

I do kind of hope I can surprise myself, but this is in terrms of my own tendencies.

There are no logical fallacies or contradictions or anything else in anything I've said.

Yes there are. See I can just say stuff too…but there actually are…a lot.

in and as pertains to this thread his remark is the very picture of the Dunning-Kruger Effect.
There are glaring strawmen in the "Graeme Revell" quotes that are proudly presented, one of them foillows his example in no way whatsoever. Presenting that dross to the group - "and now:" - is fallacious appeal to authority anyway.

[bermudagold]

i guess it's only fair to post examples of the non ai stuff i find, the old fashioned way, a friend suggests it.
this guy deserves lots more views imo.

"make some time for the people you love...then some time for the people you don't"...cant go wrong with that

reminded me of this for some reason

[BONES]

Yes there are. See I can just say stuff too…but there actually are…a lot.

Yet you seem unable to identify even one. I've given you an opportunity to back up your bullshit with an example but again, you have failed. Hardly surprising.

It's amazing how many people think they are anarchists when they are actually nihilists.

The two things are not mutually exclusive, however I have a very strong set of moral principles so I am anything but a nihilist. I am a misanthrope which is an entirely different thing. If I got hold of all 6 Infinity Stones, I'd do exactly what Thanos did with them, no hesitation.

will.i.am is on a full media blitz defending AI...of course because he is heavily invested in multiple AI startups...

Maybe he's heavily invested in AI because he believes strongly in its value?

but he is another one making this comparison to sampling...the two are not materially analogous and they know it...

Why aren't they? Explain your position on the subject. Actually, don't bother, your bias is laid bare by the rest of your post. will.i.am is just being honest, suggesting that it would be hypocritical of him to criticise the use of AI, given his own use of sampling. He is not in any way advocating for AI in this article. His example of jazz musicians and sampling illustrates the similarities very well but he's not pushing AI's barrow, simply pointing out the similarities between Ai and sampling in his genre.

Yeah, it's definitely a false equivalence.

No it's not, you are simply drawing arbitrary lines to separate the two things. The equivalence is there and completely obvious to any unbiased observer.

sampling was highly controversial back in the day, and that controversy is a big reason people have to pay for or use royalty free samples, because it was determined to be copyright infringement.

Are you stupid? That's exactly the point of making the comparison - it was highly controversial at the time but we got past that, we worked it out and now it is a common practice that nobody bats an eyelid at. We will get to the same point with AI. There is nothing at all "false" about the equivalence.

Repurposing existing music in different contexts via sampling is an incredible art form; doing via AI is just being Jive Bunny

Explain the difference to me because, frankly, I ain't seeing any, beyond the absurd belief that when a human does it, it has merit but when a machine does it, it has none. Or if you could explain how a human magically imbues a piece of music with something a machine does not, I'd be keen to hear that explanation, too. It all seems wrapped up in religion to me, which we know to be a crock of shit.

this guy deserves lots more views imo.

Why? It sounds completely generic and boring to me. More stuff I'd hear in the change rooms at work.

There are glaring strawmen in the "Graeme Revell" quotes that are proudly presented, one of them foillows his example in no way whatsoever. Presenting that dross to the group - "and now:" - is fallacious appeal to authority anyway.

For a start, I only used one Graeme Revell quote and it wasn't intended to illustrate my point so much as to shed light on the motivations for others'. If there is a straw man there (and there isn't), it belongs to Revell, not me. Beyond that I made no appeal to anything, I simply provided a link to an interview, in the interest of furthering the discussion. But great work derailing the thread with more irrelevant bullshit.

[eassae]

You're adorable.

False equivalence:

One is borrowing/stealing, depending on your perspective, work created by other artists in the service of transforming the original art into a new form of art.

The other is taking an output trained on other artists work, to help a multibillion dollar valued tech bro actually live up to that valuation, all while plagiarizing the average. The only art happening here is where the tech bro mined his data.

I'll let you decide which is which.

Are you using Chat GPT to answer?

[eassae]

Also, if you want to continue to say there isn't a difference, sampling was ultimately determined to be theft without an agreement and compensation from and for the original artist. Are you admitting that A.I. music is theft and all artists scraped deserve compensation?

[BONES]

And we are already seeing similar agreements being put in place with AI, aren't we? So what happens to your argument when all that's worked out, when record and publishing companies are happy that they are getting their cut and all the legal battles go away?

But no, I don't believe it's theft and I doubt it could ever be proven in court to be such, for the simple reason that it isn't taking other people's work and re-using it, it is doing what we all do and using what it has learned to influence what it creates

As long as it is only being trained on material that's publicly available on the internet, I don't see that anyone has any right to expect any kind of compensation. I've used this analogy before but anyone can go to a public library and read 1,000 books for free. If that person takes all that accumulated knowledge and experience and uses it to write their own books, do they have any responsibility to the authors they trained on? No, of course they don't so why should AI be held to a different standard?

this is how superficial a level of creation it is...

You mean exactly like most Hip-Hop and a lot of modern R&B? The only thing that's changed is he's sampling something he created in Suno, rather than an old record. Hardly the end of the world.

...is this what we want flooding radio and streaming platforms?

Is it any worse than what's been flooding them for the past 30 years? I couldn't care less.

One is taking work created by other artists and turning it into something thoroughly plagiarised to help a multi-billion dollar media conglomerate (once a humble record company) keep their shareholders happy and make some money for the artist. The other is taking an output trained on (not taken from) other artists' work, to help a multibillion dollar valued tech bro actually live up to that valuation, whilst also helping artists to realise their creative goals and allow them to possibly make some money.

Perhaps you really are as stupid as you appear to be? I've fixed up your pathetic attempt, to remove the bias (or at least even it out). Take out the bias and the equivalence is completely obvious.

BTW, what were you trained on, if not the work of other artists? It is totally disingenuous to even suggest that we have not learned much of what we know, what we put into our own creative processes, from listening to and playing the work of others. e.g. You hear a drum lick you like and you play or program it into your DAW or you get a session muso to reproduce it for you in the studio. Or any of a million other things that inform your creative decisions, all based entirely upon the work of others, sometimes consciously but often deeply subconscious. None of us is creating in a vacuum, everything we do is informed by what we see and hear as we go through our daily lives.

[jancivil]

I came to sequencing via computers out of being a musician for what we'd call the requisite ten years before you stand up as though to be counted, and then figuring to compose music of my very own, meaning I had ideas I wanted to bring to fruition, in a process of discovery.

I never had the impulse to grab someone else's whole riff and base something off that, although I would operate a television and gather found sound and edit to form rhythmic stuff, mainly absurd things. There was no sequencing application, so anything apart from recording in real time would be something like that. I focused on tape music my last semester in school, as well. Contiguous music was live, end of story.

But like 95% of what's in my music is about deliberate choices of discrete notes forming lines that I want to take from an idea to something concrete. If there's involvement with other musicians it can be collaborative or directed but the whole exercise is I have ideas. If I record the MIDI, record a physical instrument, draw/paint in notes I'm being specific. I've done a fair bit of covers, but the point was this is our take on this thing, or just me writing on top of Satie as though it's what jazz does to show tunes or that (EDIT: there's no pretense it isn't the Satie comp, it's not Under Pressure now known as Ice Ice Baby). Transforming a score into something else to suit my ideas but maintaining congruence with the source, I felt like a wizard.

Resort to the Large Language Model working out some verbal prompt is this other whole thing. While one supposes this could be used experimentally and very creatively, this and these arguments are definitely no such thing. It screams laziness and a lack of real interest in the task of music making.

[Bombadil]

And joy, thereof.

[jancivil]

yeah, I don't get how depriving oneself of the fun of discovery without a net gets to be a thing.
fear of failure?

[v4p0r]

But no, I don't believe it's theft and I doubt it could ever be proven in court to be such, for the simple reason that it isn't taking other people's work and re-using it, it is doing what we all do and using what it has learned to influence what it creates.

Well, Dall-e for instance didn't ever create coherent images for my prompts. If it were doing what I did, it should have drawn frames of a jumping yellow duck. I know the situation is not always like this, but if I were to decide whether to compensate the original authors of the images it gave me, I would compensate them, if the works were not licensed for the training. Whenever I tried AI like that, the added value was the number of different images I can ask it for, but not value added to the work. With code it seems to be the same. It works when every alteration is a parametrization of some sort. And when AI is asked for something difficult, it fails. But you can easily do it yourself in a reasonable time.

Not sure what the situation is with music, because this is not so much of a requirements thing, and it would need a lot of time to listen to all those results and find missing value. Isn't a musician pretty much free to make every mistake and not be judged? I think this difference might show when deciding whether AI music is a legitimate derivative or not.

[csscp]

Most of the conversation around use of generative AI in arts, in the present case music, is centred on the concept of learning from training data. Musicians without academic and technical expertise might take the concept of training in AI at its face value and conclude that AI learns from musical material just like humans do. However, that is simply not true. Human learning in general does not consist of keeping track of statistical regularities in the stimuli; this way of learning makes up a very limited cases of learning. The utilization of statistical regularities in learning music in humans is even more limited - music as a collection of systems involve way more layers, abstractions, rules, heuristics to be remotely captured by presentation of statistical regularities in an artificial network-like structure.

A layperson may mistake forming of impressions of musical styles with learning the style itself. The most distance such a layperson will be able travel in the musical universe would be basically a very limited imitation of their influences and therefore regurgitating the same old material with no new substance and it's indeed what generative AI does. However, they should not take what they do with regards to the compositional process as the singular process and equate what generative AI does as a compositional tool to the essence of how humans have created music.

[BONES]

I came to sequencing via computers out of being a musician for what we'd call the requisite ten years before you stand up as though to be counted, and then figuring to compose music of my very own, meaning I had ideas I wanted to bring to fruition, in a process of discovery.

Yeah, I don't think I ever wanted to be anything other than a songwriter and performer. I bought my first synth in 1982 and as soon as I had a sequencer, I started creating my own original music. The musician thing never held any appeal for me at all, despite my closest school friends all having a high level of formal training (or perhaps because of that).

I never had the impulse to grab someone else's whole riff and base something off that

Perhaps not consciously but that's not to say it didn't happen at a deeper level. It's almost impossible not be to influenced by the things you hear around you. I once wrote a song that, two years later, I realised had a prominent riff from an obscure 12" EP that I had forgotten I even owned. It sounded identical and I never played my song again.

It screams laziness and a lack of real interest in the task of music making.

I am more than happy to cop to the lack of real interest but I've been working way harder getting all this AI shit working in Studio Pro than I've ever worked on my own original ideas or anything my bandmate has sent me. Way harder!

yeah, I don't get how depriving oneself of the fun of discovery without a net gets to be a thing.
fear of failure?

Lack of interest. It's certainly not fear of failure, given that we've achieved an order of magnitude more than I could ever have even dreamed of. I just don't get any enjoyment from playing an instrument. To me it's like a hammer or a screwdriver - it has a purpose and when I need to build something, I'll pick it up but it has no value to me beyond that, it's just a tool. it's actually taken a long time to come to this realisation; I kept buying new synths all the time and never using them. e.g. I had an Elektron Analog Keys for around 4 years and never used it on stage even once. I'd set it up with the intention of using it but, ultimately, it was more hassle than it was worth so it spent 90% of the time I owned it in storage. So I've sold most of the hardware I owned and I'll probably give the rest away when/if I find someone who wants it. I have no intention of ever buying any more.

... but if I were to decide whether to compensate the original authors of the images it gave me, I would compensate them

It doesn't work that way. It's not cutting out a duck from another image and pasting it into yours. It's making something original.

if the works were not licensed for the training.

Why should "the works" be licensed for AI training? You and I don't have to pay a license fee to get a book from the school or university library, do we? It's an argument that has never made sense to me. It seems to be nothing more than a shameless money-grab.

A layperson may mistake forming of impressions of musical styles with learning the style itself. The most distance such a layperson will be able travel in the musical universe would be basically a very limited imitation of their influences and therefore regurgitating the same old material with no new substance and it's indeed what generative AI does.

Except it's not. The music it's been producing for us is not in any way "limited" and it if it is imitating anyone in the genre, it's nobody I've heard before. Yes, I feel like it will run out of new "ideas" eventually but that's true of the vast majority of human artists, too. e.g. When was the last time your favourite band from the 80s put out a decent record? It does happen but not with anything like the frequency it did in their hey day. But you never know, a new version might offer a whole new experience.

However, they should not take what they do with regards to the compositional process as the singular process and equate what generative AI does as a compositional tool to the essence of how humans have created music.

That's kind of the point - it does things differently, often things you might not think to do yourself, which broadens your palette and inspires new ideas of your own. What would be the point of it if it was as stupid as people are?

[v4p0r]

if the works were not licensed for the training.

Why should "the works" be licensed for AI training? You and I don't have to pay a license fee to get a book from the school or university library, do we? It's an argument that has never made sense to me. It seems to be nothing more than a shameless money-grab.

They should be licensed because they are created without understanding. If I gave instructions to an artist to draw three frames of an object acting in a certain way, and then I get back only two frames, the first one and the third one, and I never can instruct him to draw the second one, I simply must conclude that he cannot draw. And when he cannot draw, he probably gave me copies. And then I would need a license for those two copies.