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

[buffalo roam]

I think EDM is most at risk from AI, given its repetitive nature. Simple scoring for movies and other shows is also at risk when it is simply setting a mood without real creative expression. The state of a lot of pop music, lacking beautiful and creative tunes, is also at risk. But I don't think real creative expression is at risk, because AI seems to be about copycatting whatever has already been done over and over.

[ghettosynth]

Test: Ask your precious LLM for ideas about how to create fully mono compatible (!) equivalence stereophony. You know, ILD+ITD+ISD.

Mate, I don't have the first f**king idea what that even is, so it's not going to be possible for me to prompt an AI about it in any useful, meaningful way.

Well, as it turns out, it's pretty easy to do so. However, first we need some definitions. To meet the requirements of fully mono compatible equivalence stereophony, let's say pleonasm together, we have the property that when we sum L+R, i.e., as in a mono system playing a stereo source, then any stereo effect must cancel for mono compatibility.

So, if:

L = x + B(x)
R = x - B(x)

Then L + R = 2x, yay, we've lost the B. (Scale to taste).

Now, since all of this was worked out by Lauridsen-Schroeder in the fifties. There's not really a lot of room for ideas that are actually new, so, what we can do is focus on some, um, practical and interesting implementations. Here are some LLM suggestions for the operator B(x). Now, we did not work through the proper management of the delay times here, so you might have to tweak things by ear. You got this!

1. The "Aquatic Refraction" Matrix

The Operator ($B$): Audio is routed to a waterproof transducer submerged in a 50-gallon tank.

The Process: A hydrophone captures the signal after it travels through the water.

The Variable: A single goldfish acts as a dynamic modulator; its position between the transducer and hydrophone creates frequency-dependent ILD/ITD shifts.

Result: A "wet" stereo image with organic, fish-modulated "zipper" effects.

2. The "Relativistic Tape Loop"

The Operator ($B$): $B(x) = x(t - \tau_{room})$.

The Process: A reel-to-reel tape loop is physically strung across the room, around motorized pulleys, and out a window.

The Variable: Atmospheric humidity and wind resistance introduce physical wow and flutter (pitch-shifting) that cannot be modeled digitally.

Result: A stereo field that literally reacts to the weather.

3. The "Biological Buffer" (Synaptic Decorrelator)

The Operator ($B$): The side signal is converted into low-voltage pulses and sent through a cluster of lab-grown synthetic neurons.

The Process: The "audio" travels across biological synapses before being re-digitized.

The Variable: Synaptic latency provides the delay $\tau$. If the neurons are "tired" or chemically stimulated, the stereo width fluctuates.

Result: The only M/S matrix with Interaural Synaptic Difference (ISD).

4. The "Doppler Ceiling Fan"

The Operator ($B$): A continuous, high-speed circular Doppler shift.

The Process: A small speaker is mounted on one blade of an industrial fan; a microphone is mounted on the opposite blade.

The Variable: The fan's RPM determines the speed of the "zipper" rotation.

Result: A stereo field that sounds like it was recorded inside a helicopter, yet collapses perfectly to a clean mono signal.

5. The "Acoustic Mail-In" Delay

The Operator ($B$): $B(x) = x(t - 3\text{ days})$.

The Process: The side signal is recorded to a microSD card and mailed via USPS to a collaborator in another state.

The Variable: The delay $\tau$ is determined by the speed of the postal service.

Result: A stereo image so wide it transcends the Haas effect, featuring a unique tracking number for the Side channel.

Now, that's what I call power prompting.

[procrastive]

Nobody told me there'd be math!

Keep 'em coming ghettosynth. You seem to be the least clueless poster in this thread.

[Grizzellda]

Hey, KVR folk's this is heavy mathematical shit...


Who amongst KVR civilians could reckon this???

Could BONES do the mathematical procedures, in this f**k**g video???

[BONES]

There's a video? I failed second year pure maths at uni, but only because I didn't hand in any of my assignments. I actually topped the final exam. But that was 48 years ago, I don't think there's much of it left intact in my atrophied old brain. But I don't even know what the TLAs mean and I think it's unlikely an AI would, either.

[Zeisner]

There's not really a lot of room for ideas that are actually new, so, what we can do is focus on some, um, practical and interesting implementations.

"Let's change the test until LLM can solve it successfully!"

Now, that's what I call power prompting.

I call it a spectacular fail. Neither does you nor your precious LLM understand in the slightest what my approach does. Nor what the Haas effect is (Hint: The Haas effect is based on ITD but not ITD itself). You could have sat down and replicated my approach in your DAW and then looked at a spectrum analyzer and goniometer while listening and changing variables. And then worked on your prompts to at least come close to HRTF and techniques like high-instance parallel ITD (if you're lucky). But of course you couldn't do that. Another point for all those studies showing that using LLMs reduces brain activity.

[Zeisner]

But I don't think real creative expression is at risk, because AI seems to be about copycatting whatever has already been done over and over.

It's not even good at copying. Those artifacts are atrocious. But it doesn't matter for the target audience because they can't tell the difference between a Minimoog and a guitar anyway. You can feed them anything as long as it has enough bass.

[GaryG]

But it doesn't matter for the target audience because they can't tell the difference between a Minimoog and a guitar anyway. You can feed them anything as long as it has enough bass.

True... it's like people here fretting over using real hardware, if the general public is your target audience then they dont give a hoot if it's a vst, behringer moog clone or the real thing...

[Zeisner]

I had fun with psychoacoustic tests in the past. Turned out that most people hear shit. Part of this comes from hearing damage but their brains don't do much with the signals either. Some of that comes from a lack of training, most is inherent though. So if you finetune that 16-18 kHz region of your cymbals or use 5 instead of 6 dB for your tremolo, keep in mind that not even one percent of the entire population can hear it. And it's getting worse too.

[ghettosynth]

There's not really a lot of room for ideas that are actually new, so, what we can do is focus on some, um, practical and interesting implementations.

"Let's change the test until LLM can solve it successfully!"

Now, that's what I call power prompting.

I call it a spectacular fail. Neither does you nor your precious LLM understand in the slightest what my approach does. Nor what the Haas effect is (Hint: The Haas effect is based on ITD but not ITD itself). You could have sat down and replicated my approach in your DAW and then looked at a spectrum analyzer and goniometer while listening and changing variables. And then worked on your prompts to at least come close to HRTF and techniques like high-instance parallel ITD (if you're lucky). But of course you couldn't do that. Another point for all those studies showing that using LLMs reduces brain activity.

joke /jōk/
noun
Something said or done to evoke laughter or amusement, especially an amusing story with a punch line.

A mischievous trick; a prank.

"played a joke on his roommate."

Something that is of ludicrously poor quality.

"Their delivery service is a joke."

[agharta]

As for AI created music, I'm not sure how that will tour? 
Isn't all the money in touring these days? 

I'm not sure that whether Spotify has either one or two hundred million songs makes any difference to me. 
I'm not getting recommendations from there. 
If local supermarkets add extra aisles full of processed food, meat products and confectionary, it won't impact my diet. 

I can't see AI developing artists with the artistry of:
Al Green 
Jimi Hendrix 
Miles Davis 
John Coltrane 
David Bowie
The Beatles

Maybe mimic some of their songs, but plenty of people mimic the greats anyway and that is not new. 
Creating a body of work that is innovative, exciting, fresh and captivating, that moves you, touches your heart and soul is rare. 
Not many musicians consistently do that, so training AI on the majority of music, which is relatively mundane, can at best lead to more mundane music. 
Think, more aisles of processed cheese and bread, to go with the processed music and processed Bud lite. 
Use your discrimination. 

[agharta]

Nowadays you are allowed not to pay the licence fee as long as you don't access the BBC services TV, but it's a bit of a honesty box situation, I think.

I believe that you require a TV license if you do any of these:
Use iPlayer. 
Watch any live TV channel. 
Record a live TV channel and play it back later. 

So as long as you only watch on demand streaming TV, that isn't the BBC, you are fine. 
That's annoying for those that only want to watch one or two live sports event per week, as you are paying about £13 per month just for the privilige, on top of what you pay for the sport channels.

BBC Radio is great and I use that more than the TV channels. 
I mainly watch comedy stuff on their TV channels but they do have a history of music programmes where multiple bands play live in the studio. 
The Old Grey Whistle Test and Later with Jools Holland are two. 

[Vortifex]

radio 3 is indeed classical, night tracks not entirely so! it still has the serious music vibe, but you will find modular or laptop tracks, alongside new symphonic stuff.
cool show.
Sundays they do some good stuff too, i heard one show about tape loops through history.

Yep. Radio 3 is where I first heard William Basinki's 'Cascade'.

[Tubeman]

You know what's funny? We're all arguing about AI and AI don't give a crap.

LMAO.

Carry on.

Not all. Just a few peeps getting into a catfight over it.

[BONES]

I think EDM is most at risk from AI, given its repetitive nature. Simple scoring for movies and other shows is also at risk when it is simply setting a mood without real creative expression. The state of a lot of pop music, lacking beautiful and creative tunes, is also at risk. But I don't think real creative expression is at risk, because AI seems to be about copycatting whatever has already been done over and over.

I think that's a very prejudiced view. If anything, I'd suggest AI is more suited to less repetitive, more melodic styles because of the way it seems to do things.

"Real creative expression" is just as strong in repetitive musical styes as in any other for the simple reason that all songs have structure. In rock, Pop and even most Jazz, there is the tried and true formula of intro-verse-chorus-middle 8-verse-chorus-break-chorus-chorus-outro and variations thereof. It's as old as the hills and for good reason - it works. Id' also suggest that working within those and other kinds of constraints often requires/inspires far more creativity than a rambling, structureless piece.

I call it a spectacular fail. Neither does you nor your precious LLM understand in the slightest what my approach does. Nor what the Haas effect is (Hint: The Haas effect is based on ITD but not ITD itself).

Mate, I don't even know what ITD is so if the LLM knows at least that much, it's already miles ahead of me and I've got a (lapsed) MENSA card. (Yes, you get a membership card.)

You could have sat down and replicated my approach in your DAW and then looked at a spectrum analyzer and goniometer while listening and changing variables.

Or you could just get on with your life using the time-honoured method of "sounds good, is good" and be done with it. I can't say I've put more than about one second's thought into mono compatible mixes in 45 years and 10 albums. If people have so little respect for the music that they'll listen to it in mono, f**k 'em. Where would go to listen to music in mono in 2026? I can't even remember the last time we played over a mono PA and AM radio is all talk-back now, with little or no music. Why would you even bother?

It's not even good at copying. Those artifacts are atrocious.

What artifacts? I don't hear anything half as bad as Synth 1 or even Atomika, and we use the f**k out of those things all the time. If you can't deal with artifacts, you're not much of a f**king engineer.

But it doesn't matter for the target audience because they can't tell the difference between a Minimoog and a guitar anyway.

... and neither can you in a mix, guaranteed. Again, you have to resort to prioritising the process over the result, which is the stupidest f**king thing imaginable. People don't care because it has nothing to do with what makes a song good. It's very sad that you don't understand this but I suppose when someone gives you a hammer, everything looks like a nail.

So if you finetune that 16-18 kHz region of your cymbals or use 5 instead of 6 dB for your tremolo, keep in mind that not even one percent of the entire population can hear it.

And of that group, I'll guarantee that not even 1% of them would give a flying f**k.

I can't see AI developing artists with the artistry of:
Al Green 
Jimi Hendrix 
Miles Davis 
John Coltrane 
David Bowie
The Beatles

Who would want them to, that's the past. Do you honestly not think those guys wouldn't be using AI if they were starting out today? These are artists who pushed boundaries and today AI is where the boundaries lie. The Beatles would be all over it.

Creating a body of work that is innovative, exciting, fresh and captivating, that moves you, touches your heart and soul is rare.

 Yes, it is and none of the artists you've listed have ever done that for me. I like a lot of Beatles songs but they don't connect with me in the way that Killing Joke, Nitzer Ebb or Die Krupps do.

training AI on the majority of music, which is relatively mundane, can at best lead to more mundane music.

It is if that's what you ask for but, as you've noted, it's not the only stuff they train on, so it's not the only thing they learn how to do. What you're saying is like suggesting that a classically trained pianist can only play Classical music. It's nonsense.