Maybe there is a way to fix the AI problem!? (Let's talk about how we can handle this sh*t)

[Starbright]

Ok, so I've been having a hard time seeing what AI can do (especially the generative and transformative kind like SUNO).
And I got really frustrated and depressed over the outcome of IT being pretty decent till good over time, and having to compete against it, if not now, then in the future.
I started becoming depressed because it's becoming harder and harder to tell what's AI and what's not.
And so I started looking in the first major entertainment industry that fell victim to robots/AI: chess.
Every chess player in the world knows they can't beat Stockfisch. And so they use it as a training tool and learning resource.
And truth is, that people still watch tournaments because they root for people and the thrill of seeing their favourite player win or lose by a mistake.

And so I started thinking what is really bugging us all in the first place:
It's that companies make money from our work, using our music as training data for their business without our consent!

But that also gives us a solution to the problem: treating training data like samples/session musicians.
And what I mean by that is using "Band-in-a-Box" as an example.

For those that don't know, "Band-in-a-box" is an automatic, flexible backing track creator using pre-programmed and pre-recorded files by session musicians (they hired them).


But the short drift aside. Here's my idea:
Let's have generative AI companies flag the data used in a song and make prompters agree to be responsible for properly crediting them (as data songwriters/data session musicians ?) or pay the data originators a buyout sum (similar to ghostwriters or ghostproducers, etc.).
That way, everyone responsible gets paid more fairly and has more oversight over their art and what's done with it.
Plus, that way we get the rights to stop certain productions and songs that use our stuff for songs we don't agree with.

So what are your ideas and thoughts about it?
(I actually tried to send SUNO an email to suggest this to them, but I have no hope that it will reach them...)

PS. This is a repost since I accidentally posted it in "Getting Started"

[telecharge]

You might like this thread - viewtopic.php?t=622429

[zerocrossing]

My prediction is that AI is a problem that will fix itself. There's already some crazy percentage of AI content on the web, and it's only going to increase. Where does AI go to feed? The web. So it will be eating its own sh!t, so to speak. So, without us, who are now discouraged from posting, the quality in originality of anything AI can produce (which is already not great), will become much worse and boring. I became bored with it after a few weeks of seeing what it could do. My wife is already complaining about the content on TikTok.

But that's not the biggest factor. As we all know, it's super processor intensive and expensive to use. No companies are turning a profit with the current pricing, all hoping to rope in customers and then turn up the pricing. So cheap slop will become expensive slop that no one will really want to pay. The AI caused layoffs sure won't be good for the economy anywhere, so that will further stop people from paying for it.

AI can do a lot of amazing things that a human can't do. Look at millions of breast cancer scans and learn how to catch early breast cancer much faster and more accurately than any human. It's fantastic at diagnosing rare diseases. I bet it will eventually be able to trace money laundering in a very quick and accurate way. Distilling meaning from huge data sets is really its wheelhouse, but I think in a short time, we will see a rejection of it for use in music and art, at least when it comes to fully producing works. Being able to hum a melody and have it get turned into a guitar solo? We will definitely get a lot more of that sort of thing, which is fine, in a sense. No different than using a sample library.

[bermudagold]

Here's my idea:
Let's have generative AI companies flag the data used in a song and make prompters agree to be responsible for properly crediting them (as data songwriters/data session musicians ?) or pay the data originators a buyout sum (similar to ghostwriters or ghostproducers, etc.).

THIS exists and is already being done...see my threads
viewtopic.php?p=9130524#p9130524

[VOODOO U]

You guys don't have a problem. You can be rest assured your music isnt nowhere near the radar to be ripped by A.I.
Relax.

[funky lime]

start a band and play live gigs. human people still go to those.

[kritikon]

Scrolling YT, I already simply switch off from an awful lot of AI shite. There are a certain few voiceovers that are the obvious AI ones and I just scroll past them - as soon as the voice starts and it's the recognizable AI shit, I move on. Same with the vids that follow a pattern. For some silly reason recently I get those ones with a dog saving the baby from a leopard or tiger in a suburban garden, or a kitten outstaring a wolf (again, in suburbia ). I ignore it by habit now, and I'm sure plenty of others do. The music is the same - if it's AI then it's execrably derivative, that's how it works. It's not generating anything original so I ain't going to listen to it. Mind you, plenty of the pop shite is also so derivative I can't distinguish between the real thing and AI. Some teenager mumbling into a mic like her false teeth don't fit isn't trendy, it just sounds stupid like the previous teenager she mimicked because she thinks it's trendy.

So, yeah AI will be self-limiting simply because it's not doing anything original. It has it's uses. I toyed around with those image-generating things and for sure it's better than anything I can do because graphics is not my thing. But it still seems to come up with much the same thing again and again, no matter what commands I type differently.

[cryophonik]

The way to beat AI is simple. Fill the internet with lies, falsehoods, and bad music for it to learn from. I’m doing my part,..are YOU?!

[VOODOO U]

The way to beat AI is simple. Fill the internet with lies

KVR has that covered.

falsehoods,

KVR has that covered too.

and bad music for it to learn from

Holy shit to hell KVR *definitely* has that covered.

[BONES]

start a band and play live gigs. human people still go to those.

Do they? Not where I live. We get the same 10 people to every gig. 30 years ago, when we were terrible, we had no problem getting 50 to 100 punters to a show. We have to travel half-way around the world to get a decent crowd, which would be great if it didn't cost us $5,000+ to earn $1000 (after tax).

[BONES]

The thought that AI companies are ripping us off is complete bullshit. Humans do exactly what AI does - they hear music from many different sources and it inevitably informs their own creative output. e.g. When we first heard Leather Strip's Solitary Confinement, it completely changed our perceptions of what our own music could/should be and changed what we do forever.

When we go to a club or turn on the radio, we aren't paying for the IP of the music we hear, so why should AI companies pay for publicly available content they use to train their AIs? It's a double-standard that defies any logic.

Anyway, I think AI is f**king awesome and it's already making way better music that we can. My bandmate has been mucking around with Tunee for a week or two and some of the stuff he's come up with is unbelievable. As in, it's so good that if I heard it at a club, I'd have to go and ask the DJ who it was so I could go home and buy everything they've ever released.

As it is, we have a two-pronged plan. First is to deconstruct the AI generated songs, using AI powered stem separation, cutting the tracks into phrases so we can rebuild them into something we can use as a bed, before replacing the AI vocals (which have a very obvious American accent) with our own. If that doesn't work, or doesn't work well enough, we'll instead create a whole album of AI generated music to release as a side-project because what my bandmate has come up with is well worth sharing with the world and has the potential to be far more popular than anything we've done as two stupid humans chasing what makes us want to jump around like lunatics.

To be clear, the AI stuff isn't perfect out of the box but we've been able to improve it considerably by reinforcing the drums with Battery, using Ozone's Rebalancer to make the vocals sit in the mix better (they are always too loud), followed by running it through my standard mastering chain.

The other thing is that you can say that it's not our music or that we didn't create it, the AI did, but I think that is also bullshit. I would defy anyone else to come up with music like ours, it is completely and utterly a product of our input and is as much ours as anything we've ever done in Cubase or Studio One or even in our hardware from the old days. It's just a different set of tools that you have to use in a different way but there is as much creativity involved in writing the AI's prompts as there is in crafting a song in a DAW. Yes, it feels less musical but most of the production process is completely unmusical anyway, so it's really just an incremental change, not the massive paradigm shift people seem to see it as.

[funky lime]

start a band and play live gigs. human people still go to those.

Do they? Not where I live. We get the same 10 people to every gig. 30 years ago, when we were terrible, we had no problem getting 50 to 100 punters to a show. We have to travel half-way around the world to get a decent crowd, which would be great if it didn't cost us $5,000+ to earn $1000 (after tax).

Where I'm at, rock, punk, jazz, metal, soul, blues, jam bands, folk, funk, country, are all alive and well. Indies and big names.

From my observations, dudes with laptops and backing tracks aren't really in demand here unless it's karaoke night. Maybe nightclubs too, but that's not really my thing.

[BONES]

Public Image Ltd has a laptop on stage, it's not any kind of impediment to live performance or attracting a crowd. It's just another tool people can use to get the sound they want and/or that their audience expects.

Interestingly, in our city of 8 million inhabitants, I don't think jazz, soul, blues, jam bands, folk, funk, country have ever had much of a live presence. I assume they exist but they are anything but thriving. The one venue I know of that catered to some of those genres closed years ago.

AI will be self-limiting simply because it's not doing anything original.

Yes, it is, in the right hands. The stuff my bandmate is making with AI is extremely good and, vocals aside, it sounds more like us than any band/artist I can think of. (The vocals are great, they are just a world away from the way I'd do it.) But it has done things we'd never think to do, or done things in a way that would never occur to us, so it's really not too much like us, either. It's very much it's own thing, as original and as good as any of my favourite artists.

Where it gets a bad rep is likely from amateurs with low standards. Tunee gives you 150 credits a day and it costs 4 credits to render out a version of what you're working on. Sik has three accounts, so he is churning through 450 credits a day. That's more than 100 iterations per day, yet after two or three weeks - well over 1,000 rendered songs - we only have three songs that we'd consider good enough to do anything with. I doubt the idiots flooding the internet with AI music are anywhere near as discerning, so I wouldn't be judging AI's capabilities by what you hear online.

[BONES]

Nobody has anything to say now? A dose of reality is a bit overwhelming, is it?

[BertKoor]

> deconstruct the AI generated songs, using AI powered stem separation

Interesting... I've heard rumours this is the way to detect whether it's AI generated or not. Myth or truth?