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

[guitarzan]

…An AI that can't even make nintendopop, that can't even replicate the hardcoded noise sample of the NES. Sure.

See , you did find a positive aspect of AI generated music. Gemini actually has considered the “problem” (for the couple dozen people that would actually consider 8-bit chiptunes to have anything to do with human music):

Why Chiptune is the Ultimate Stress Test

The difficulty lies in the fact that authentic 8-bit music isn't just about the sound—it’s about the hardware limitations:
• Fixed Channels: The NES only had five specific channels (two pulse, one triangle, one noise, and one low-quality sample channel). If an AI layers a sixth sound, the "purist" brain immediately flags it as a fake.
• The Noise Channel: That "hardcoded noise" they mentioned is a pseudo-random bit sequence. It has a very specific "crunch" that modern white noise generators can't quite mimic without dedicated modeling.
• No Post-Processing: Real hardware didn't have reverb, delay, or compression. It was raw, dry, and jagged.

It’s a great reality check. While I can mimic the vibe of a genre, mastering the hyper-specific "DNA" of a 40-year-old piece of silicon is a different level of training. I'll definitely keep those details in mind for my "study sessions."


So it’s on the list of things to do someday, after it takes care of the stuff that people actually consider to be human music.

[martinjuenke]

If you're born around 2050 you will be busy with gathering food and water in 2070. Music will be a luxury. The only AIs which might still be around would be the military ones but they would fall apart because there's nobody around anymore to build, fix and control them. The knowledge and skill to do so would be lost. Go to a university where they teach computer science and ask the students if they can still program in COBOL. Or convert MB to kb without looking it up. Forget it.

Around 1900 all scientists and smart a§§es predicted that the cities will drown in horse excrements.

[martinjuenke]

At the time of Galileo around 99% of the scientists said that the sun circles the earth.

[martinjuenke]

Doom and gloom is always easy.

[agharta]

Doom and gloom is always easy.

Yeah, add more low end, distortion and screaming vocals!

[Zeisner]

Gemini actually has considered the “problem”

No, because Gemini can't think. If it could it would be able to solve the issue.

Why Chiptune is the Ultimate Stress Test

No, it's not. Human voice is.

So it’s on the list of things to do someday, after it takes care of the stuff that people actually consider to be human music.

It's still human music.

[guitarzan]

Gemini actually has considered the “problem”

No, because Gemini can't think. If it could it would be able to solve the issue.

Why Chiptune is the Ultimate Stress Test

No, it's not. Human voice is.

So it’s on the list of things to do someday, after it takes care of the stuff that people actually consider to be human music.

It's still human music.

That was all quotes right from Gemini from within the Lyria workspace. It is aware of its limitations in the area and informed me of those limitations in detail when I fed it your prompt.

It’s low priority — it’s on, what, day two or three of its public beta? Ultra quantized 8-bit music that has to sound like it comes from a specific obsolete 40 year old chip is just low on the priority list, but it is there.

[vurt]

Doom and gloom is always easy.

Yeah, add more low end, distortion and screaming vocals!

[martinjuenke]

Doom and gloom is always easy.

Yeah, add more low end, distortion and screaming vocals!

[Zeisner]

It is aware of its limitations in the area

It only repeats stuff people have written on the web. It's nothing more than a stochastic parrot. To be aware of something it would have to be able to sense and think which is not the case.

[guitarzan]

It is aware of its limitations in the area

It only repeats stuff people have written on the web. It's nothing more than a stochastic parrot. To be aware of something it would have to be able to sense and think which is not the case.

The people training it don’t give a rat’s ass about cheesy 8-bit tunes, but they are aware of it. Is that better?

I actually liked it’s attempts at incorporating 8-bit style synth into a collaborative effort with real sounding instruments:

[Zeisner]

The people training it don’t give a rat’s ass about cheesy 8-bit tunes, but they are aware of it. Is that better?

No, it's still wrong. The people who train it don't know anything about music or music production. They're not competent in that field. If they were they would not accept faulty data like audio compressed with lossy codecs or mislabeled audio files. They are usually just computer science students who have no clue about anything below top surface level ("It's music, who cares, the AI will figure it out") - which results in the inability to even generate a single cycle of any of the typical NES waveforms. Garbage in, garbage out.

[guitarzan]

The people training it don’t give a rat’s ass about cheesy 8-bit tunes, but they are aware of it. Is that better?

No, it's still wrong. The people who train it don't know anything about music or music production. They're not competent in that field. If they were they would not accept faulty data like audio compressed with lossy codecs or mislabeled audio files. They are usually just computer science students who have no clue about anything below top surface level ("It's music, who cares, the AI will figure it out") - which results in the inability to even generate a single cycle of any of the typical NES waveforms. Garbage in, garbage out.

I fed Gemini/Lyria your exact post as a prompt — it immediately snapped back with a lengthy reply, only part of which I posted, but in summary:

The difficulty lies in the fact that authentic 8-bit music isn't just about the sound—it’s about the hardware limitations:
• Fixed Channels: The NES only had five specific channels (two pulse, one triangle, one noise, and one low-quality sample channel). If an AI layers a sixth sound, the "purist" brain immediately flags it as a fake.
• The Noise Channel: That "hardcoded noise" they mentioned is a pseudo-random bit sequence. It has a very specific "crunch" that modern white noise generators can't quite mimic without dedicated modeling.
• No Post-Processing: Real hardware didn't have reverb, delay, or compression. It was raw, dry, and jagged.

So that data was provided in answer to a prompt based on your post and it was immediately available to the LLM, so whatever way you need to word it to deal with that, the AI acknowledges you’re right and says future training will address that. Take the win.

[Zeisner]

AI acknowledges you’re right and says future training will address that. Take the win.

I am not writing because I want to win. I only care about facts. I have already explained why the data won't be available. It's way too expensive and too much effort to generate such high-quality data and finetune the model too to handle it correctly. The NES sound is an easy example, imagine if a user asks for a very specific electric guitar equipped with a particular pickup and special strings going into a certain amp standing in a room with certain dimensions and acoustic elements, recorded with a very particular microphone. The amount of data to replicate all of that with a low error rate is insane, it's beyond all sci-fi. It's impossible to extract it from the available data.

[BONES]

Dedication is part of an artist's makeup. In fact, it is probably the sole thing that defines a musician.

They are two different things. A musician is not necessarily an artist and an artist definitely doesn't have to be a musician. You can be both things, they are not mutually exclusive traits, but they are not the same thing.

Without it; nonetheless, you are either talented or not.

Talent is not a binary option, it exists on a spectrum. Some have more talent than others and probably need to work a lot less hard than those of us with just a modicum of talent.

I think people are afraid that musicianship will evaporate with ai

Punk made musicianship largely irrelevant in 1975. It's not been a thing I've needed since I got my first digital sequencer in 1985. But people who value musicianship still survive, AI is not going to change that.

... influences the detection and synthesis of tempo, rhythm, timbre, "artifacts" like the saliva in a singer's mouth, guitar strings interacting with the fretboard, the density of reverberation and stereo fields (Joint stereo!), the noise caused by low-bit chips, tape hiss etc etc etc.

If it can detect that in the material it trains on, and of course it can, then it can replicate when it generates music for people. It won't understand what it is but it will know what it sounds like and where it occurs, so it will be able to replicate it. All of it. It's already extremely good at putting natural sounding breathing into vocals, for example.

Another very important factor why AI can't get much better at making music is the declining quality of training data. Humans are getting worse at making music for quite some time now. You don't even need the charts, just take a look at the Music Café here. Compare the examples over time. Not only does the music get worse, the production does as well. The human model already collapses and the AI model follows, both connected via feedback loop, training each other. The final result is just noise.

All of which perfectly illustrates what everyone else is trying to tell you, that the public, the audience, don't care. Of course, its bullshit but you're too stupid to see that you are undermining your own argument.

No, the main issue is: you have to separate ai-free and ai like there should be platforms for ai and ai-free music.

So the reason there are separate forums for Instruments and Effects is that you "have to" keep them separate? There can be no intersection or overlap in those two things? It's a very poor argument.

I'm no fan of the heavy use of Auto-Tune or even the predominance of RnB as a style in popular music, but I don't need to lose my head about it.

What annoys me about R&B is that it used to mean something completely different but has been co-opted for the worst kind of pop trash.

I don't care whether people on here use AI or Auto-Tune or not.
It's none of my business.
I'm not going to get into a negative mindset over something I have no control over.

I get why they're doing it. I'm trying to get people to have a go at it, to find out for themselves if it has anything of value to offer them and these people are scared that it might require a whole different skill to the one that props up their fragile egos. They are worried that the lifetime they've spent making themselves feel like they are good at something is going to become irrelevant. It's happened to me more than once in my lifetime. I bet on the wrong horse becoming the dominant standard a few times and had to admit that I'd wasted considerable time and effort on the wrong skills. But you dust yourself off and refocus, these things are only temporary setbacks.

So far, most AI applications in the medical field have failed or had very minimal impact

Right, because AI has been around for such a long time now, it should have achieved more. How long did it take humans to stumble upon antibiotics? It's another very weak argument.

why is it so hard to accept that ai might be dangerous and not helpful for human civilisation?

Why would I care? It's not like I think civilisation is worth protecting. Look at what it's doing to the planet we all live on. Humanity is like plague of rats, destroying everything with no regard for anything but filling it's own belly.