Artificial Intelligence (AI) generated music

[mambo4]

2 Thoughts:
1: It's another nail in the coffin of the "recording artist" as a valid profession
The industry will enthusiastically pay nothing if they can package and sell it.
Back to humans playing live music as the primary income stream.

2: I've been watching the "AI /Robot Revolution" for over a decade and while AI /robots do keep checking off items in the list of things " computers will never do", I am starting to get the feeling that this may be like nuclear fusion, a revolution that is perpetually 10 years away.

I'm hopeful though for a future where nobody *has* to work to live comfortably, and we can choose our burdens.

 

[wintremute]

Meh. Most human output in the arts (music, painting, writing, etc.) is complete dreck. I suspect most AI-generated art will meet the same standard.

That's a great point. If you take any given year and look at the songs that were put out, 99.9% of them are meh, and we hold onto that tiny 0.1% and say it was a good year.

 

[MonetBass]

AI is only as good as the database behind it. With storage being relatively cheap these days, those databases can be vast. But even so, they are still finite, and relatively small compared to the human mind.

Putting variables into an AI system to generate the music I hear in my head seems quite inefficient. "Let's try this... NO, no, that's not what I wanted. How about this... ugh, NO!" etc. (Can you tell I'm a reformed programmer?) Whereas I can sit down and knock out that piece/song quickly using minimal equipment.

Part of what makes art, music or literature interesting to me is imagining the creator's mindset when they were creating. That won't exist with artificially created works.

Does this way of creating have merit? Certainly. Will it replace humans creating art, music and literature? I doubt it.

 

[wintremute]

2 Thoughts:
1: It's another nail in the coffin of the "recording artist" as a valid profession
The industry will enthusiastically pay nothing if they can package and sell it.
Back to humans playing live music as the primary income stream.

2: I've been watching the "AI /Robot Revolution" for over a decade and while AI /robots do keep checking off items in the list of things " computers will never do", I am starting to get the feeling that this may be like nuclear fusion, a revolution that is perpetually 10 years away.

I'm hopeful though for a future where nobody *has* to work to live comfortably, and we can choose our burdens.

1. I agree. A business will almost always go with the cheapest solution. Look at all those self-checkout machines at grocery stores now.

2. You may be right on that asymptotic "never quite get there" theory. Some people have pointed out that coming up with lyrics and singing will be next to impossible for AI, but that was not something I considered. Often I ignore the singing in a song, or listen to it as another instrument, so I didn't think of that aspect when creating a song. Do you think an "everything except the vocals" song is attainable?

3. We are getting closer and closer to being able to feed and shelter everyone, and we will have to make a choice: will we choose to give people food and shelter, or will we choose not to.

 

[wintremute]

AI is only as good as the database behind it. With storage being relatively cheap these days, those databases can be vast. But even so, they are still finite, and relatively small compared to the human mind.

Putting variables into an AI system to generate the music I hear in my head seems quite inefficient. "Let's try this... NO, no, that's not what I wanted. How about this... ugh, NO!" etc. (Can you tell I'm a reformed programmer?) Whereas I can sit down and knock out that piece/song quickly using minimal equipment.

Part of what makes art, music or literature interesting to me is imagining the creator's mindset when they were creating. That won't exist with artificially created works.

Does this way of creating have merit? Certainly. Will it replace humans creating art, music and literature? I doubt it.

I can sit and write a song, and sometimes when I play what I hear in my head it turns out...it should have stayed in my head. However, I am not a great artist. But here I am using Midjourney to make really cool image (in my opinion). Will there be AI software to allow people who cannot play an instrument to make songs that they enjoy? Will they be able to make songs that the masses will enjoy?

 

[wintremute]

I had to reply because about a month or two ago, the algorithm has thrown this at me after some refining.
View attachment 4862916
Very similar to one of your images and I think this shows the limitations of this tech quite well.

I like that! I agree that's it's pretty similar to what I generated. However, I think that's because that style is kinda the default setting. Luckily for me, I like that style.

Here's an example of different styles this guy produced: https://weirdwonderfulai.art/wp-con...Guide-min-by-robomar.ai_.art-Insta-scaled.jpg

An even better example is what this person did with circles: https://towardsdatascience.com/exploring-midjourneys-art-style-using-circles-a461a78e7196

I think it's like someone else said, the limitations are going to be based on what the person enters.

 

[wintremute]

AI generated music will be the next thing and also another way of making music, then something else will come along, it always does. Electric instruments were worried over as a threat to acoustic instruments, then synthesizers were going to replace all other instruments, then auto tune and corporate manufactured music was.... Well, you get the picture. I'm sure AI generated music will be just the thing for churning out even more cost effective pop, doing TV show and movie soundtracks and a host of other things, but the stage show, or the arena concerts might be a tough ask.

Technology keeps advancing, but people are infinitely variable and for some reason we still cling to some of the old. Folk, traditional, baroque, classical, opera, blues, jazz, country (and western), rock and roll, rock, prog, punk, metal, soul, funk, techno hip hop and so on still exist and people love to see real people sing, play, emote and interact. Humans have lived and played music since they formed tribes and communities and probably always will.

I agree 100% with this.

 

[wintremute]

That's not what's happening here.

A huge amount of I.P. is being selected by humans for the purpose of ingestion, analysis, and reverse engineering, of a product they are manufacturing for profit.

It's a massively complex legal minefield, and if the music publishers haven't figured this out, they're asleep at the wheel.

That is an interesting angle. Some of what you say reminds me of years ago there were some musicians who were complaining that their records and CDs were being sold at used record stores, because they were not seeing any of that money.

At what point when I purchase a product am I not allowed to take it apart and see how it works? Does that need to be part of an EULA?

How is it different if I as a human repeatedly listen to a Pink Floyd album, and then try to recreate something similar to it, but with my style mixed in -VS- a computer doing the same thing?

 

[wintremute]

I think ^^^that is the key to any of the autonomous/self-generating/AI "creative" [sic] arts applications producing content that is compelling and engaging, rather than simply a curiosity: It still requires some aesthetic decision-making, both in terms of selecting what materials the app will use as its fodder, as well as culling the output to identify/filter the most rewarding product from the least rewarding. Right now we still require a human with vision to perform those two curating tasks.

Yeah. That goes with what I said elsewhere. You may generate 100 hours of "music", but you have to listen to all of those hours to find the 2 minutes of music that doesn't suck.

 

[wintremute]

Nice image, and an even better catch, I always assumed there might be something like that happening, given a finite library of imagery to draw on. This is the first time I've seen that confirmed.

The style is also in some ways singular, as if the same mind were judging the results at the time of programming the original system.

They're a very good judge, and people know how to select the best of its output well, but a view of twenty artistic book covers will usually show an astounding variation even though twenty minds or less contributed to the collection.

Artificial Intelligence (AI) generated music

 

[wintremute]

AI-generated music is way further along than you might suspect. For example, there was an AI project last year that set out to complete Beethoven's last symphony. He died before completing Number Ten. Actually, it was fairly incomplete, but the AI project picked up where he stopped and the results are amazing.

Read about it: Invalid Link Removed

Hear the Tenth Symphony played by the Beethoven Orchestra of Bonn.

I listened to that whole thing, and I thought: Yeah, that does sound like Beethoven. Yeah, that does sound like symphonic music. Wait, was that an organ?

It sounds like it was a very hands on project by a team of musicians and scientists. It also sounds like they had to do the "listen to hours of music and then pick out the seconds of it that sound right" method.

Invalid Link Removed

Overeall, pretty cool IMO.

 

[swarfrat]

The difference between you listening to Pink Floyd for decades and then generating something in the style of, is that you are a creative being. And more importantly, you can't provably examine your source code and determine that your album was created after listening to 10,000 hours of pink Floyd and nothing else. Where AI can only create what it knows. Is it practically the same thing? Perhaps. But I think it can and should matter from a legal standpoint.

Besides. Who owns the AI generated 'work'?

 

[swarfrat]

Another practical aspect of who owns AI generated work is that it will be the literal death of new human generated work. If I turn some bots loose I could copyright every single possible permutation of notes that will fit in every scale inside 4 bars down to 64th notes, and that should be enough to make me able to copyright claim everything that could ever possibly be created in the future.

 

[Lostgallifreyan]

Artificial Intelligence (AI) generated music

The links on the page didn't work for me but I think I get the idea, that there may be more examples of divergence than convergence. The thing is, if that is so, then it may just prove that artists are developing ways to use a new tool, and that may be the strongest use the AI will be put to.

About the copyrighting of all permutations, as described by swarfrat, I'm fairly sure that won't happen, partly because it would be unenforcible, and the cost would destroy those who wielded that power, but mostly because mood is something AI can't do, unless someone programs it to, and the moods would be very formulaic!

The other point recent posts go into is the selection by humans from a large amount of computer generated output. That's almost like a reduction to human-as-quantum-computer. I don't know where that could lead, but I think it's safe bet that if people did find parallels between brains and quantum machines, it wouldn't be long before brains were found to exhibit behaviour that those machines can't do, and that like Turning machines, quantum machines may be found to be similarly limited in scope, restricted largely by the constraints imposed by making any design real.

The universe expresses itself in mass, energy, and information, all the time, but the moment we look at it selectively, with any judgement, it's a finite form we see, so we'll never experience it without limit.

How a machine may inherit what we inherited, and become 'alive' as we see life, I can't answer. The fact that every age has its automata suggests that the question may never be resolved, let alone in a universally accepted way. Suppose it did? It might then all stop, with no purpose.

Maybe Arthur Clarke had a point, about naming nine billion names of God. Maybe we shouldn't try too hard.

 

[swarfrat]

Melody is copyrightable, not feel.

 

[MonetBass]

One thing that humanizes a song are the flubs, whether intentional or not. That would be lacking in AI music.

 

[Bob_Ross]

If I turn some bots loose I could copyright every single possible permutation of notes that will fit in every scale inside 4 bars down to 64th notes, and that should be enough to make me able to copyright claim everything that could ever possibly be created in the future.

Have you read Jorge Luis Borges' fable "The Library Of Babel"?
He essentially posits an enormous library where every single book that could possibly be written (using any/every possible combination of the letters of the alphabet) exists.

Think about how large that library is. How many volumes it contains. Every. Single. Possible. Combination. Of Words.

In many ways his fable is a thought experiment, an intuition-pump. One thing that comes out of thinking about that library is the realization that the number of books it must contain is nearly infinite. "Vast" in the literal sense. But here's the rub: Suppose you were to try to find one specific book in that library? Try to imagine what a daunting task that would be. Hint: "Daunting" isn't the right word..."impossible" is more accurate.

And that's the problem with your premise: Presuming it even is possible to have the bots churn out "every single possible permutation of notes" -- and given the near-infinite vastness of the Borges fable, it's my contention that the melodic equivalent would be even vaster -- how could you search the resulting database (library) to find the melodic example that corresponds to the alleged plagiarism whose copyright you want to challenge?

 

[swarfrat]

You don't need every single posysing. Only enough to claim a phrase. That's why I limited it to 4 bars. Really two is probably enough.

We know there are already copyright trolls claiming works on YouTube as their own, and claiming ownership of white noise, don't deny it can't happen. It already is.

 

[wintremute]

Another practical aspect of who owns AI generated work is that it will be the literal death of new human generated work. If I turn some bots loose I could copyright every single possible permutation of notes that will fit in every scale inside 4 bars down to 64th notes, and that should be enough to make me able to copyright claim everything that could ever possibly be created in the future.

That is something I thought about years ago. There is finite amount of possible note and chord arrangements. What is stopping someone from generating them and copyrighting them?

 

[Lostgallifreyan]

You don't need every single posysing. Only enough to claim a phrase. That's why I limited it to 4 bars. Really two is probably enough.

We know there are already copyright trolls claiming works on YouTube as their own, and claiming ownership of white noise, don't deny it can't happen. It already is.

This is true, and it's why I haven't released my synth software. It's harder to find a way to get it out there so it survives, that it was to spend well over a decade on my own writing it.

That said, in the end I will release it, to a bunch of pirates if need be. I will not let fear of predation destroy it before it can be born into the world.

Now, if I can be ready to lose my right to work that took me longer than many murderers spend in prison, it seems to me that fearing the consequences of any predation are pointless. Better to gang up on them and mob them the way birds mob a cat that is trying to kill them.

People can and do resist. This is why those who wield legal power can destroy themselves. It's what 'class action lawsuit' means. There ARE defences available. All it takes is to make predators lose the thing they think they'll gain.

One nice thing about small sequences of notes is that the smaller they get, the more prior art there is. This logic is useful to protect a creator in patent as well as copyright law. One reason the Stranglers music is so original is that it is built from so many smaller sources. More riffs in one bassline than most artists use in a whole CD. Stravinsky, and many more recent composers, are blatant self-confessed magpies, and their originality with what they do with the bits they 'steal' is very effective at preventing predation, and I bet they're happy if someone steals bits from them. If someone plagiarised Nice And Sleazy, or the Rite Of Spring, people would notice, which is a very good protection in itself, it makes people reluctant to do it. Most recent music biz lawsuits are happening over very bland, very generic sequences with very detectable parallels, and several of the cases are thrown out of court.

Judges usually like justice, they don't like rich predators trying to eat the poor. The world isn't 'fair', but the most egregious (I love that word) examples tend to get corrected once they are big enough to be visible.