Is artificial intelligence in music really intelligence? What is the difference between intelligence and a pre-written algorithm, script or simply a recorded macro? Is artificial intelligence the new version of batch processing? Will artificial intelligence products be artificial till the end?

 

Last edited: Dec 8, 2024 at 11:54 AM

This term artificial intelligence is a made-up marketing term to sell the products better.
We call artificial intelligence = artificial stupidity. Don't fall for the industry, AI is the new hype to fill the companies' coffers again.

The only one who is intelligent is you, who just used your brain to ask this question.

 

When I hear something like this, I still believe in mankind (instead AI):

https://www.youtube.com/shorts/8xglW8p_EbY
and this one just for fun:
https://www.youtube.com/shorts/X-js7sOX5ik

PS. close to week ago I posted on audiosex a topic kinda AI or Human, who win? Then that topic had 4 pages of discussion (I added there 2 tracks above)...then suddenly the topic disappeared...I thought:
"AI intruded to prevent the victory of mankind"

 

I have no idea how this works, but it seems to me that you need to specify exactly what the AI should do, and it just does it with different results

 

Music is already a way oversubscribed space.

Most people in it have no inherent creative ability, although they can copy or attempt to replicate what already exists with varying degrees of success. So can AI.

Like with any new technology ever, there will be a very very small handful of geniuses come along who will find a way to use it in new and pleasing ways.

It depends on your outlook on life whether AI is a good or a bad thing.

It cannot replace humans, in the sense that it isn't and never will be, human.

It can however make them redundant.

 

The reason for this question that I asked "how does it work?" is because if the functioning of these platforms is known, many doubts will be resolved and people will understand what they are facing.

So far, I have not been surprised by the specific results of artificial intelligence, but the way these platforms work is questionable, why companies that have a very strong history in making audio tools have not developed these platforms but some sturtup guys in partnership with Microsoft which do not have even a known artist in their teams have been able to develop these platforms.

https://en.wikipedia.org/wiki/Suno_AI
...Suno was founded by four people: Michael Shulman, Georg Kucsko, Martin Camacho, and Keenan Freyberg. They all worked for Kensho, an AI startup, before starting their own company in Cambridge, Massachusetts...

 

The differance is AI learns from the outcome.

So a Process (Algorithm/scrip/macro) will do something it's instructed to, to the data you feed it. In some cases will do differant things depending on what you feed it.

Ai will like any process will do something it's instructed to, to the data you feed it.
Where it differs is a AI will learn from the outcome or another way to put it, process the outcome and use it to amend the process.

So say the process you have is to fade the end of the song over a set amount of time depending on the average volume of the track and the peak volume in the segment being faded.

A process will do just that, exacterly the same way forever.

An AI will review whats been done and if will change what it dose. If the peak volume in segment being faded has no impact on the fade being applied it may stop scanning for the peak to improve on speed. Another thing that may happen is if you feeded it every song with the same average volume, it may learn all songes are the same so don't scan and just apply new predefined fade.

They are great for tasks but every once in a while they'll learn something bad and need reset.

If we're going to go onto the type that are ment to create music on their own. You're just looking at something like the old Dance e-jay from the early 2000 that sets the speed and throughs a bunch of sound banks into a project at the push of a button.
https://www.ejayshop.com/product/ejay-virtual-music-studio/
It's all down to what sound banks it has and how lucky the person who pushed the button.

 

true Ai doesnt exist yet , i guess not even in military research .
---
enjoy the last days with the slight aftertaste from 1960-2000 human life and dont contribute in foster threads
look at me ...look at .... , you dont want that dont you....eheh

---
fuck all this actually know nothing about it AI threads , which are mostly even about pseudo marketing AI threads
---
man foster your notes ,industry and AI themes are so plain and amateurish in compare to your pre 2020 threads
i could open 10 threads with 100 comment each in shortest time if i wanted

 

Haha, Dance e-jay and acid pro made me like an David Guetta of old...

 

personally I feel a bit guilty that in order to download anything from the sister site I have to identify the pics with (insert whatever) in it, or whatever the capture is looking for .... knowing fine well, what I'm really doing is helping train AI. Oh the irony!

 

del

 

Last edited: Dec 9, 2024 at 5:22 PM

I trained ai models of my voice and now my ai model is training me. I stopped singing almost completely for a few months, and it's true what they say; "use it or lose it". I've been recording my voice since I was a kid so I put together some lead vocal stems from before I stopped and made an AI model of my voice (a few actually) and they turned out so realistic I kept almost believing I'm listening to myself. To be fair, I tend to use songs with singers that share similarities to my singing and that I could sing in real life. It's really helped with training my real voice again and has helped expand it as well. Plus when I'm producing a song I can put my vocals through someone else's ai model so I don't get annoyed hearing myself all the time. It's not perfect, but it's been helpful and fun to play around with. Still can't come close to actually singing and still needs a human to input a vocal track to alter.

 

I think the line between art and entertainment will become relevant again in public discourse soon enough. It should be obvious by now that most uninspired, lazy and generic music has always been disposable. Yes there are gray areas, there are weaker genres, but it's pretty much across the board, from goa to bebop. Feeling sorry for these human replacements should be like feeling sorry for the invention of typewriters, in the sense that we're talking about jobs, families, work relations, very concrete situations and individuals. It's Neil, the good neighbor and father of four, who has had to abandon his dream of becoming a pop star and will have to settle for being a local entertainer, like in the old days. Yes, it might hurt the communities, but that has always been the case with new technologies. With cheaper energy sources, depopulation and UBI, communities will hurt less over time. Now it's a very different subject when you think of Frank Zappa or Lili Boulanger. These artists can be copied alright, but when A.I. can reach and surpass their level of artistic innovation, then it'll come the time for a deeper sorrow, for our kind and our limits.

Could you share your tools and method with us?

 

Do you want a technical answer or a "philosophical" one?
If you're looking for the first, here's roughly how it works:

Machine learning techniques are ways to learn statistical patterns in data when you have no idea where to even start if you wanted to define a mathematical equation (a physical model) for that thing. I'm going to touch on supervised learning, which is the paradigm mostly used for these generative models. (It's actually "self" supervised, but that's being technically pedantic and not worth differentiating for this discussion)

You basically define a structure that is sufficiently complex and can adapt in order to describe whatever you want it to (in our case, the neural network architecture) and then you give it an input X and an output Y and you check how far off from the output Y your model's effect with X is, so a distance (error) between Y and model(X). The learning process is how you algorithmically take steps that change the values of the parameters in the model so that the distance between Y and model(X) becomes smaller.

The generative models, including the ones for music, are mostly trained with excerpts of the thing you want to generate and the corresponding text description of that thing, so they learn the statistical connection between a description like "upbeat funky track with slap bass and horn section" and a sound excerpt that matches that description.

The models are noisy and have a random sampling component to it, so when you try to use the already trained model with some text description, that random element will make it so it's not a total recall like listening to an mp3 from your hard drive. The model just learns a statistical mapping between words and musical excerpts, simulates a random part and picks based on the probability distribution that it learned. It's likely that if the model if good and your description includes "heavy metal" it will produce something resembling distorted guitars, for example, since it's statistically likely.

We can go philosophical and say that the random aspect of these models is somehow akin to creativity. I don't think like that because whenever I hear these generated tracks they all have this distinctive shit tone, just like the generated images all seem to have the same glossy and shiny art style that is unmistakable.

If there really was a robot that was consciously creating more interesting music than what I could find from humans, I would listen to it, but I haven't heard it yet. I don't know why an hypothetical being whose mind's physical support are silicon circuits instead of organic neurons like ours would be less deserving of our respect than any other human. But that's in the realm of science fiction; my hot take is that the grifters from silicon valley really want you to buy into the hype that it's actually close by but they won't tell you that for the past year and a half they haven't gotten any progress on making their models significantly better even though they're doubling down on the computation power and data used. Just my opinion.

 

Interesting, but I wouldn't say that about graphic arts (Midjourney, specifically), text or video. Would you say it's different with music?

 

if you do anything online you train algorithms. google products, apple products, etc..

 

This is a sad statement. I wish you can sing again someday.

 

Take one song, tag it's style and other characteristics (the more granular the better) and then proceed to add white noise in little increments until the song become only that: white noise. Do this for all the music at your reach, tag, add noise, and repeat. Then when you ask the model to give you some music based on certain tags you are requesting (prompting) the transformers (fancy name for an algotrithm tech that we have at the moment) will "denoise" from pure white noise and give you a representation of all the tags its has been trained on. That's how all the diffusion models work, video, text, and image as well (the media hype call them AI, but they are not).

Then when you have a pretty decent model based on human knolodeage, you start to produce an curate billions of syntetic material to retrain it so it will be better with each iteration.

For me is the biggest theft of knwolodeage on the history of humanity.

 

AI music is relatively unintelligent. The generator cannot does not hear or appreciate the sound. It simply structures a huge amount of data and blends it according to a pre-programmed schema (developed by a human) to produce "new and similar" results. There is no soul in the recognition of music, and that is why it needs a ton of data.

Meanwhile, a child can learn decent piano skills at the age of 5, without listening to terabytes of music first. I have taught people to play piano decent piano in two years from scratch, with the ability to play by ear. And that's the key intelligence in my opinion - the ear. It triggers reasoning, judgement, music taste. Appreciation of music is what makes us amplify one style or blend another, not some algorithm that tries to blend parameters you ask it to.

So, no, AI music is not really intelligent. It's just performant when tuned extensively and laboriously by human intelligence...and it is not yet "great", but continues to improve.

With that said, Ai music generators will rob it's users of the skill, experience, and brain development that comes with learning an instrument and composing music naturally.