AI Generated Metal...

[narad]

Popularity would be perfect for record labels seeking to make maximum profit, but not really for any other situation. Generic sounding "goodness" doesn't lead to musical innovations. AI will copy what it is programmed to deem desirable. Musical innovation is achieved through someone doing something for the very first time, not by following existing patterns.


My comment above provides the fundamental reasoning, but here's some examples:

- Would AI have thought "what if guitars had steel strings instead of nylon?"?

- Would AI have thought "what if I plucked the strings with a shard of tortoise shell instead of my fingers?"?

- Would AI have thought "what if I slide this metal tube across the fretboard?"?

- Would AI have come up with over the top whammy use (dive bombs etc.) after someone invented the Floyd Rose to achieve improved tuning stability?

- Is AI able to invent new guitar techniques, such as sweep picking, bends, vibrato, hammer-ons / pull-offs, legato, staccato etc.?

- Is the lead guitarist of an AI band going to lose their fingertips in a factory accident, then subsequently change their tuning and string gauge to compensate for this, creating a new tone that inadvertently leads to a new genre?

- Is AI capable of inventing a new vocal technique / style? (e.g. death growl, throat singing, scat, yodelling or even Jack Black's inward singing haha!)


AI could theoretically have come up with something "innovative" like Nightwish (if they didn't already exist), by combining operatic vocals with metal, but only if operatic vocals and metal music already existed. It couldn't have invented the operatic singing style or the metal music style.

Well I mean...yes? You can use AI to invent new timbers from interpolations or extrapolations from existing sounds. This can even include sounds that are not originally musical in nature, which in theory makes it much more general starting point than your typical musician. If you really want it to be vocal, you could incorporate a model of the vocal tract and explore variations on possible sounds there, though I don't think that's on anyone's to-do list.

Your list seems really stuck in the context of physical inventions that humans then utilized, as some sort of innovation. But any innovation in musical tool could then be incorporated into a model. A new sound might need some human feedback to steer the model towards utilizing it in pleasing ways (since it would likely not have reasonable model of how humans perceive or enjoy such sound combinations) but nothing stands out as "dealbreaker".

 

[jaxadam]

You can use AI to invent new timbers from interpolations or extrapolations from existing sounds.

*timbres

 

[narad]

*timbres

That's the keyboard I say tambers in my head.

 

[Neon_Knight_]

Well I mean...yes?

Artificial Intelligence can do this without clear direction from human intelligence?

If humans decide to explore new vocal sounds by creating a model for AI to work with and then telling the AI how to make the vocal sounds pleasing, would that actually be innovation by AI or computer-generated music created by human intelligence? It's hardly the same as a human independently making a new vocal sound and then independently incorporating it into their music, in the hope that some other humans will appreciate it. The "new" vocal sounds produced by the AI would be contributing no more to the creative process than a non-intelligent EQ randomiser would be contributing to a guitarist discovering a a new guitar tone through trial and error of randomly selected EQ settings.

I can see how AI could be "creative" within the very limited confines of synthesised music, but I don't find that music at all pleasing in the first place.

Unless programmed to know what a guitar sounds like and what it sounds like when a guitar string is bent (which is only possible if it has already been physically done), AI could only produce a synthesised note with a synthetic representation of a bend.

Unless an AI has been programmed with "this is how it sounds when a metal tube is dragged across the strings of an electric guitar", how could it create this sound? If no-one has thought to try the slide guitar technique yet, how would they have thought to programme an AI with the information necessary to achieve it?

 

[narad]

Artificial Intelligence can do this without clear direction from human intelligence?

If humans decide to explore new vocal sounds by creating a model for AI to work with and then telling the AI how to make the vocal sounds pleasing, would that actually be innovation by AI or computer-generated music created by human intelligence? It's hardly the same as a human independently making a new vocal sound and then independently incorporating it into their music, in the hope that some other humans will appreciate it. The "new" vocal sounds produced by the AI would be contributing no more to the creative process than a non-intelligent EQ randomiser would be contributing to a guitarist discovering a a new guitar tone through trial and error of randomly selected EQ settings.

I can see how AI could be "creative" within the very limited confines of synthesised music, but I don't find that music at all pleasing in the first place.

Unless programmed to know what a guitar sounds like and what it sounds like when a guitar string is bent (which is only possible if it has already been physically done), AI could only produce a synthesised note with a synthetic representation of a bend.

Unless an AI has been programmed with "this is how it sounds when a metal tube is dragged across the strings of an electric guitar", how could it create this sound? If no-one has thought to try the slide guitar technique yet, how would they have thought to programme an AI with the information necessary to achieve it?

Well we're way off the deep-end now and I don't think too much consideration has been given by the research community to both creating new sounds via AI, and utilizing them in AI generated songs. I can't imagine it doing this without human feedback, though I suppose after a certain amount of feedback it may have some predictive power about what types of sounds are useful, and how to use them.

But when you take a step back and say, okay, if I had <insert favorite genre defining album> and think what makes this a new genre, timbre can be a part of it, but there's gotta be something in the music theory side of things that is an important step forward. If the melodies, rhythms, tempos, harmonies, etc., are all the same as some other genre, it's hard for me to think of that as the birth of a new genre. Like Meshuggah played on a Minimoog isn't a new song or genre, etc. It's in these categories that I think a lot more attention is placed, that people place most of their playing and creative time, and that will soon be automate-able.

 

[Neon_Knight_]

Well we're way off the deep-end now and I don't think too much consideration has been given by the research community to both creating new sounds via AI, and utilizing them in AI generated songs. I can't imagine it doing this without human feedback, though I suppose after a certain amount of feedback it may have some predictive power about what types of sounds are useful, and how to use them.

I wasn't just talking about human feedback, but being driven by human creativity to achieve a specific goal thought up by human intelligence. If AI is given only existing sounds and human feedback, with no other human intervention, I don't see how it could achieve a new sound. The creativity has to come from human intelligence.
AI music and AI-assisted music are two different things. AI-assisted could be innovative, but pure AI could not be.

But when you take a step back and say, okay, if I had <insert favorite genre defining album> and think what makes this a new genre, timbre can be a part of it, but there's gotta be something in the music theory side of things that is an important step forward. If the melodies, rhythms, tempos, harmonies, etc., are all the same as some other genre, it's hard for me to think of that as the birth of a new genre. Like Meshuggah played on a Minimoog isn't a new song or genre, etc. It's in these categories that I think a lot more attention is placed, that people place most of their playing and creative time, and that will soon be automate-able.

Reproducing an existing style / genre of music on a different instrument is exactly the sort of "innovation" that AI is capable of producing.
Things like inventing new techniques / new ways of using instruments, innovating the designs of instruments to achieve specific desired goals, inventing new instruments, creating a new vocal style etc. (i.e. the things required to create a new genre) are not things that AI can do imo.

 

[narad]

I wasn't just talking about human feedback, but being driven by human creativity to achieve a specific goal thought up by human intelligence. If AI is given only existing sounds and human feedback, with no other human intervention, I don't see how it could achieve a new sound. The creativity has to come from human intelligence.
AI music and AI-assisted music are two different things. AI-assisted could be innovative, but pure AI could not be.


Reproducing an existing style / genre of music on a different instrument is exactly the sort of "innovation" that AI is capable of producing.
Things like inventing new techniques / new ways of using instruments, innovating the designs of instruments to achieve specific desired goals, inventing new instruments, creating a new vocal style etc. (i.e. the things required to create a new genre) are not things that AI can do imo.

On the novel sound front, AI can work in the real audio space, in which case creating new sounds is easy. That's been done. I mean, it actually has to work quite hard to generate an existing sound from all the other noises that exist in the other combinations of 44k numbers per second, etc, when you think about it. And AI can take any sounds and interpolate between them to create new sounds. So can AI create new sounds? I think the answer is inarguably yes. Even when given existing sounds. Whether or not it would know how to use these new sounds it creates in musical ways is a separate issue, and while I think it might initially require human feedback, I also don't see why it would need prolonged human feedback. I mean, they rarely do, that's kind of the point.

So AI creates new sound -> tries to use sound in novel ways -> some suck and some are good and are judged that way by humans -> loop (now requiring less feedback on each iteration)

Sounds a lot more like AI music than AI-assisted music to me. I mean, I can't compose anything, but I can do that.

But yea, I think the posts are getting a bit repetitive, but basically I think the important core question is whether you will need humans to innovate in important ways, and while there are certain advantages to being human (being a better judge), most of what musicians do when they're "creating" is going to be equally well-served by AI. And sure, an AI is unlikely to invent a new sound that is analogous to a sound a human invented by physically manipulating some real world object. But anyone can invent a new instrument or a new sound or a new vocal style, independent of creating good music with it. So I don't see that as any sort of issue, personally.

Timbre and new sounds aside, I think there are general doubts that AI can do creative things stemming from some gut reaction or consideration of human beings as special. Creativity is just navigating through a large but finite space of possibilities to find novel places that are good. Humans cite inspiration as a principle for how they navigated that space, machines will just use models. And then we get back to the Alpha Go situation, of why it can seek novel moves that can even be hard to initially comprehend to humans, but are unquestionably good. There's no reason why Steve Vai can look at a bunch of 80s and 90s music and think, "aha! For the Love of God!" / and a machine cannot study similar music and wind up in a similar place (or better! or different in a cool way!).

 

[byzon]

It could easily be an Amon Amarth song lol! I like the fact that the AI made the female Orc play guitar instead of bass. A solo would have been beyond impressive, maybe on the next song lol!

But if this is really AI generated from start to finish, is there any point in composing music anymore?

Yes, the point is we get to do it. That’s where the enjoyment lies.

 

[jaxadam]

That's the keyboard I say tambers in my head.

*tambourine

 

[Voodoo Marshall]

They owe Amon Amarth a fucking royalty check!!

 

[Neon_Knight_]

They owe Amon Amarth a fucking royalty check!!

That's another can of worms. AI music, by its very nature, is at high risk of plagiarism and lawsuits. Imagine how Lars would respond if this had sounded like Metallica instead of Amon Amarth.

 

[bostjan]

That's another can of worms. AI music, by its very nature, is at high risk of plagiarism and lawsuits. Imagine how Lars would respond if this had sounded like Metallica instead of Amon Amarth.

I'm pretty certain Metallica replaced Lars with an AI, so it'd be the historic first of an AI suing another AI...

 

[nightflameauto]

I'm pretty certain Metallica replaced Lars with an AI, so it'd be the historic first of an AI suing another AI...

A current-level AI would nail the double-kicks live. He's last generation at best.

 

[jaxadam]

 

[BMFan30]

AI is dumber than a box of bricks that still falls for reverse psychology.