The current state of AI

[zerocrossing]

Evolution by natural selection in humans has ceased? I posted two examples earlier - the people of St Kilda (the island, not the hipsters) and the people in the Andes (and other places at high altitude)

People confuse evolution with progress. Evolution is just an organism using mutations to better fit within its environment. Humans would only get more intelligent if there was some environmental pressure that was causing dumb people to have fewer children than smarter people, and that's quite obviously not happening.

[zerocrossing]

Do people feel there's a solid ethical different between Agentic AI in music production for doing automated tasks, versus Generative AI - which is obviously just ripping of musicians.

This question is impossible to answer, because it relies on false premises. Generative AI isn't really "ripping off musicians," any more than I'm ripping off all the music that has inspired and influenced me. The difference is, I have intent and emotional connections with said music. Agentic AI isn't really any different than a very fancy macro.

I bring this up a lot, but everyone should listen to a podcast called Shell Game, where a guy attempts to create a software company that's just him as a "silent partner" and a few AI Agents. Hilarity ensues. The TLDL of it all is, the agents are terrible at a few basic things. First off, unless specifically asked, they do nothing. Unless asked to stop, they will continue to do something forever. They have no actual understanding of time. They can't reason, or make decisions. When asked to brainstorm a name for the company, one of the names that was suggested was Palantir. When asked to be more autonomous, they'd do strange things, like call applicants (they needed to hire a second human because social media campaigns kept blocking the agents) at night before a scheduled interview and launch into the interview, all while telling the applicant that it wasn't the interview and the interview was still scheduled for the next morning. They redid the interview the next morning and had no memory of the previous interaction. Getting agents to actually remember things was difficult to impossible.

So, unless they can figure out a way to give agents proper memory, and the ability to reason, I think they are doomed to doing tedious tasks, which is fine. Generative music AI is doomed to making slop that's just mimicking other music, but without any additional emotional input. It only sounds like music, because a lot of music isn't really very inspired in the first place.

[ksandvik]

Shell Game is a great podcast. Shows the 'tech bro' attitude to AI. Well, we are in the middle of of the AI bubble anyway. Waiting it to burst so I could get good prices on used SSD drives.

[stratology]

... I have intent and emotional connections with said music. Agentic AI isn't really any different than a very fancy macro.

You're making one of the most important points here, IMHO.

What are the easiest, and, at the same time, most enjoyable, things that musicians do?

- communicating emotion
- coming up with new ideas
- interacting spontaneously when playing together
- being able to identify immediately what sounds good and what doesn't - taste
- expressing a unique personality


AI does not even attempt to do any of those things.

Looking at music in the most cynical way - as a re-assembly of old ideas, devoid of personality and emotion - is the basis for generative AI music.
It does not reflect the attitudes and personalities of musicians, it reflects the attitudes and personalities of the people who create the generative AI technology.


There is an additional, more pragmatic consideration: is there actually an audience that prefers AI generated music to real music? ¨What's your favourite AI band?¨
AI music generation appears to be mostly targeted toward non-musicians who want to create their personal AI music, for an audience of one.

[ghettosynth]

This thread is hilarious.

[zerocrossing]

... I have intent and emotional connections with said music. Agentic AI isn't really any different than a very fancy macro.

You're making one of the most important points here, IMHO.

What are the easiest, and, at the same time, most enjoyable, things that musicians do?

- communicating emotion
- coming up with new ideas
- interacting spontaneously when playing together
- being able to identify immediately what sounds good and what doesn't - taste
- expressing a unique personality


AI does not even attempt to do any of those things.

Looking at music in the most cynical way - as a re-assembly of old ideas, devoid of personality and emotion - is the basis for generative AI music.
It does not reflect the attitudes and personalities of musicians, it reflects the attitudes and personalities of the people who create the generative AI technology.


There is an additional, more pragmatic consideration: is there actually an audience that prefers AI generated music to real music? ¨What's your favourite AI band?¨
AI music generation appears to be mostly targeted toward non-musicians who want to create their personal AI music, for an audience of one.

Right. It succeeds as so far as it mimics music, and unfortunately, it is mimicking a lot of music that is pretty uninspired. We can't be critical of AI and not also be critical of most pop music. Not just modern pop music, but from all eras. We think old music is better, because we blank out all the crap, and remember the good. Time heals, as they say. It's praying on an already broken system, so it's easy pickings. Think about one of the best eras for pop music. The 80s are almost universally admired. Why? I think it was a confluence of two things, both technology. Synthesizers were becoming cheap enough to afford by many, due to the DX line and synths like the Junos or Polys. Add to that, MTV hit the world and there was not enough content to fuel a 24/7 music video station, so they played everything. In an hour, you could hear Duran Duran, Jethro Tull and Joan Jett. I remember walking out of the room while my sister had "Joan Jett and Friends" on MTV, and the opening band was... King Crimson. That kind of variety of music being funneled down one outlet was unheard of, and hasn't really happened again. In the 90s it happened a bit via Napster. I remember seeing something on someone's hard dive that I liked, so I'd download everything to see what it was. I discovered a lot of great music that way, and every time I did, I went out and purchased the CD.

I have been having some success with TikTok. Tell it what you like, and it's pretty good at finding it, and other things that you might like based on what other people with your tastes also like.

Anyway, I'm going off topic, but the point is, even if I hate something, I can tell it was made by someone who liked it. Why would I listen to your AI song, when I could just have my own made for me? Someone did some research and found that people who were using AI to generate music were only consuming their own renders. That's sort of sad.

Alternately, I can work on a song for a long time, produce it over the course of weeks or even a month, and release it into the world, including here, and not get more than a dozen or so listens. It was like that when I used to play in bands, as well. The same 10-20 people would show up to all our gigs, and except for a period of time when a new member brought in her own audience, it never really changed much. The people who liked us, seemed to really like us, but it never snowballed into anything sustainable. So... I don't really care about AI music, because it's not really replacing my success. I've never really had success. (don't pity me, I think my sh!t's genius. )

[BONES]

Do people feel there's a solid ethical different between Agentic AI in music production for doing automated tasks, versus Generative AI - which is obviously just ripping of musicians.

How, exactly is generative AI "ripping off musicians"? It has no agency, it does what you tell it to do. It can certainly be used to rip off musicians but it's people doing the ripping off, not AI. It's completely and utterly up to how you choose to use it.

Evolution by natural selection in humans has ceased? I posted two examples earlier - the people of St Kilda (the island, not the hipsters) and the people in the Andes (and other places at high altitude)

Yes. Those people didn't adapt last week, they've adapted over millennia.

Generative music AI is doomed to making slop that's just mimicking other music, but without any additional emotional input.

That would only apply if it was generating music of its own accord but as soon as you put a human at the keyboard, writing prompts and guiding the process, none of that applies. The "additional emotional input" all comes from the human element overseeing the work.

It only sounds like music, because a lot of music isn't really very inspired in the first place.

Given that music is sound, surely all it needs to do is "sound like music" to actually be music? Maybe it won't connect with you, personally, but it is completely obvious that to the vast majority of listeners it's easily good enough to get the job done. For genres like those I work in, where the human element tends to be actively suppressed, AI is pretty much the perfect tool. But it's still just a tool, it's only doing what we tell it to do (most of the time, anyway).

Well, we are in the middle of the AI bubble anyway. Waiting it to burst so I could get good prices on used SSD drives.

Why do you think it will burst? It is saving businesses all over the world countless billions already and it hasn't got anywhere near its potential yet.

[Shabdahbriah]

That ^^^

[ksandvik]

The reason is overinvestment, I do think Claude and Gemini will survive but many of ther other investments are circular investments that do not make business-wise sense. Apple is also playing the right cards with a wait-and-see strategy.

[BONES]

What are the easiest, and, at the same time, most enjoyable, things that musicians do?
- communicating emotion

AI can do this extremely well, at least as well as you or I

- coming up with new ideas

The vast majority of musicians do nothing of the sort, they mostly play songs they already know. They strum a few chords on their acoustic guitar and think "wait a sec, that sounds a little bit like this song I like" and then spend the rest of the evening working out how to play the whole song. Or they play in a symphony orchestra, reading from sheet music someone else has prepared, of music written centuries ago. They probably aren't even interested in coming up with new ideas, they just like to show off at parties.

- interacting spontaneously when playing together

Not something I've ever experienced in my 40+ years in the game. In every jam session I've ever been involved in, all we did was work out how to play songs we already knew. Not an enjoyable time, really.

- being able to identify immediately what sounds good and what doesn't - taste

AI is much better equipped to do this than any musician.

- expressing a unique personality

Except, again, the vast majority of musicians do nothing of the sort, they just do the same things as everyone else.

AI does not even attempt to do any of those things.

AI does all of those thing very, very well. I hated jamming with other people, I love jamming with AI - throwing around ideas and seeing what works, letting it contribute to the results where it's unique perspective gives the project a huge boost, rebuilding the bits that we think we can do better. It's a far more enjoyable, less awkward process, than working with people. You don't have to be polite, you don't have to worry about upsetting it if you tell it it's idea is garbage. You can get shit done.

Looking at music in the most cynical way - as a re-assembly of old ideas, devoid of personality and emotion - is the basis for generative AI music. It does not reflect the attitudes and personalities of musicians, it reflects the attitudes and personalities of the people who create the generative AI technology.

That's absolute bullshit from someone who has clearly never even bothered to try it out. Our experience with AI is that it allows us to absolutely nail those aspects of our music, far better than we've ever been able to do in the past, for the simple reason that you have to tell it what you want, which makes you think a bit more about what it is you want to achieve before you start. That doesn't stop you from taking all manner of diversions, of trying out a lot of different things, way more than you might otherwise because it's so quick and easy. You're not wasting half-a-day on an idea that doesn't pan out, you can put it together in minutes and if it doesn't work, you move on.

There is an additional, more pragmatic consideration: is there actually an audience that prefers AI generated music to real music?

Why is that even a question when nobody can tell the difference? I could play you a dozen of our songs and you wouldn't have clue one as to which were written by AI and which by us.

AI music generation appears to be mostly targeted toward non-musicians who want to create their personal AI music, for an audience of one.

Which is a great thing, don't you think? Democratising music production, bringing it to the masses. Better than anything mainstream music has foisted upon the world in the last 20 years.

It succeeds as so far as it mimics music, and unfortunately, it is mimicking a lot of music that is pretty uninspired.

How do you not, by this point, understand that's only because the people guiding it are uninspired? It's just a tool, it's output is a direct reflection of the input made by the human directing it.

It's praying on an already broken system, so it's easy pickings.

It's not preying on anything, that's the people who use it. As always, it comes down to people.

Synthesizers were becoming cheap enough to afford by many, due to the DX line and synths like the Junos or Polys. Add to that, MTV hit the world and there was not enough content to fuel a 24/7 music video station, so they played everything.

I don't think that was it at all. It was all well underway before the DX7 or MTV. It goes back to Punk in 1976, which spawned amazing pop bands like Buzzcocks, the Police, Generation X (Billy Idol), The Stranglers, The Cars, Blondie, Talking Heads, Devo and countless others. By 1981, when MTV launched, it was well underway and synths were already a huge part of it. If anything, I'd suggest the DX7's arrival in 1983 probably heralded the downturn of great 80s popular music. For me the Halcyon Days were 1977-82 but no matter how you look at it, it was happening before either of those things affected anything.

In an hour, you could hear Duran Duran, Jethro Tull and Joan Jett. I remember walking out of the room while my sister had "Joan Jett and Friends" on MTV, and the opening band was... King Crimson. That kind of variety of music being funneled down one outlet was unheard of, and hasn't really happened again.

We had that here in Sydney from the mid-1970s, courtesy of a government-run radio station, 2JJ, later 2JJJ FM. And I think the UK probably had similar radio available, given the enormous impact of radio DJs like John Peel. The variety of music you heard/saw on MTV was nothing more than a reflection of the times, rather than a driving force.

Anyway, I'm going off topic, but the point is, even if I hate something, I can tell it was made by someone who liked it.

Why would that not include AI generated songs?

Why would I listen to your AI song, when I could just have my own made for me?

Because I'm better at it than you are. Because the fact you made it yourself doesn't mean you won't also enjoy what I've created. I mean, we all make our own music here but the vast majority of us still appreciate music made by other people. Why would it be any different with AI generated music? It's a nonsense argument.

Someone did some research and found that people who were using AI to generate music were only consuming their own renders. That's sort of sad.

It also seems highly unlikely, don't you think?

Alternately, I can work on a song for a long time, produce it over the course of weeks or even a month, and release it into the world, including here, and not get more than a dozen or so listens. It was like that when I used to play in bands, as well. The same 10-20 people would show up to all our gigs, and except for a period of time when a new member brought in her own audience, it never really changed much. The people who liked us, seemed to really like us, but it never snowballed into anything sustainable. So... I don't really care about AI music, because it's not really replacing my success. I've never really had success. (don't pity me, I think my sh!t's genius. )

I hear ya, it's the same for us. We can get to no. 2 in Germany but it's the same 20 enthusiastic people who show up to gigs at home. And chart position does not mean anyone actually buys our music (it's an "industry" chart, like the Oscars, not a sales chart). But we know how good what we do is, we're proud of what we've achieved with it, and that's always been enough. Don't get me wrong, it would be great if we could fill a venue, because we'd get to play a lot more often, but it is what it is and that's more than enough to keep us going and to keep our label supporting us.

[Danno]

Weird that these AI systems simultaneously have no agency but are also able to communicate emotion.

[ghettosynth]

[ghettosynth]

Weird that these AI systems simultaneously have no agency but are also able to communicate emotion.

So, in other words, probabilistic systems can mimic emotions just fine, that is, in effect, what you're saying. This doesn't mean that they have agency, it just means that your mind can't tell the difference between human words and machine words when interpreted through your own emotional register.

This is, of course, true. It's why people get angry at LLM conversations. It's behavior appears as intent, but, it's your mind that is doing all of the work there. To the extent that there is intent encoded in the output, it's, broadly, in the preferences of the RLHF raters and the organization, that manifests in the alignment layers. This is why ChatGPT 4o appeared much more emotional to users who wanted to treat LLMs as a romantic simulation.

So to reject emotion in AI music, you will have to identify what conveys that emotion in human music and then argue why that isn't expressed in the outputs of a generative system trained on that emotional input. In particular, you will have to argue why your mind is immune to it. Keep in mind that you have already admitted that generative systems are "able to communicate emotion."

[zerocrossing]

Do people feel there's a solid ethical different between Agentic AI in music production for doing automated tasks, versus Generative AI - which is obviously just ripping of musicians.

How, exactly is generative AI "ripping off musicians"? It has no agency, it does what you tell it to do. It can certainly be used to rip off musicians but it's people doing the ripping off, not AI. It's completely and utterly up to how you choose to use it.

Evolution by natural selection in humans has ceased? I posted two examples earlier - the people of St Kilda (the island, not the hipsters) and the people in the Andes (and other places at high altitude)

Yes. Those people didn't adapt last week, they've adapted over millennia.

Generative music AI is doomed to making slop that's just mimicking other music, but without any additional emotional input.

That would only apply if it was generating music of its own accord but as soon as you put a human at the keyboard, writing prompts and guiding the process, none of that applies. The "additional emotional input" all comes from the human element overseeing the work.

It only sounds like music, because a lot of music isn't really very inspired in the first place.

Given that music is sound, surely all it needs to do is "sound like music" to actually be music? Maybe it won't connect with you, personally, but it is completely obvious that to the vast majority of listeners it's easily good enough to get the job done. For genres like those I work in, where the human element tends to be actively suppressed, AI is pretty much the perfect tool. But it's still just a tool, it's only doing what we tell it to do (most of the time, anyway).

Well, we are in the middle of the AI bubble anyway. Waiting it to burst so I could get good prices on used SSD drives.

Why do you think it will burst? It is saving businesses all over the world countless billions already and it hasn't got anywhere near its potential yet.

In all honesty, Bones, I put your music in the "uninspired derivative crap" bucket, so anything you say is really just making my point to me. You should definitely use AI, because you have failed at achieving anything better.

[zerocrossing]

The reason is overinvestment, I do think Claude and Gemini will survive but many of ther other investments are circular investments that do not make business-wise sense. Apple is also playing the right cards with a wait-and-see strategy.

As predicted by Sam Altman, there will be a huge bubble burst and many people will lose tremendous amounts of money. I can smell it, just like I could smell the dot-com bubble bursting from the inside of a dot-com that seemed to be running on hype. Too much desperate hype about AI, that frankly isn't really that useful or cool. Sure, AI can generate a tune, but the money it cost you to get that tune generated by prompts is going to be more money than just buying a song on iTunes, or you could probably buy an entire album on CD, and it's going to be a hell of a lot better. And that's what they're charging now, when they're operating at a loss. Just wait until they try and generate a profit. That equation will look even worse. Soon, we'll have seen cats doing all the funny stuff that cats can't really do. We'll see Donald Trump as Jesus and Hitler. We'll see all the wacky stuff that is funny, but not worth the effort of generating it conventionally, and then what? My wife already complains that it's ruined her TikTok feed and she's bored of it. Since more and more of online content is AI generated, it's training on its own output, like a dog eating its vomit.

Will some AI stuff survive? Sure. I've said before that I'm very impressed with Synthesizer V, but that's a lot different than generative AI. More like a synthesizer that's been trained by listening to humans sing. I like it, because I have like 15 parameters to mess with, not including phoneme, and timing. I imagine it's only a matter of time before Kontakt orchestral instruments are replaced by something similar. I guess the SWAM instruments are already this, in a sense.