Aerosmith removed from roller coaster preshow at Disney's Hollywood Studios

Lord Gonchar:

It just sort of turned into "What you do sucks" on this last page and at that point I have to figure there's no other retort.

I know what that feels like and I am sorry. I took it too far.

LostKause:
Everything will come from what has already been created, just changed.

You could say it's no different than what's happening now. Every pop country song sounds the same right? But truly great music is inspired by other people, it's not copied. I have a lot of conflicting ideas about A.I. in music it's a bit like sampling in 90s hip-hop. The truly great producers took other people's work and made it their own to the point that you couldn't even recognize where it came from. It takes a lot of skill and musical ability to do. Then Puff Daddy came along and just stole hooks from other songs wholesale.

I use AI at work. I see the use case for prototyping ideas, being more productive, filling in for instruments I don't have at my disposal, etc, but using prompts to create a song bristles me. We would never accept someone submitting a photograph created with prompts to the coaster db. "In the style of Ansel Adams take a photo of Millennium Force at Cedar Point from the main stage midway." I know it's not the same but I think the comparison makes some amount of sense?

Also, I love the artist I linked to above. They're using A.I. tools to do a lot of what they do and I think it's really creative and funny.

LostKause:

AI mimics real music so well that it's difficult to tell the difference, but it's still counterfeit.

This post has SO much to unpack. I literally can't give you all of my thoughts (that novel I mentioned) except to say, I'm not sure you fully understand how to use AI creatively and not lazily.

eightdotthree:
I took it too far.

Not at all. I'm a big boy. (and it really wasn't you) It was just a terrible argument. AI doesn't suck because I do and the thread didn't need to go there.

The rest of your post has so many great thoughts that I'm a little overwhelmed with it as well. I agree with it all right up until:

but using prompts to create a song bristles me.

Clearly, you're not the only one. I'd love to know why. And that's not just a question for you.

Is the song less valid? I don't think so.

A good song it a good song. I literally don't care how the sausage is made. (Most people don't - it's the earlier discussion that the average person doesn't even understand what they're listening to anymore)

Is it the idea that someone takes 'credit' for 'writing' a song?

This I get a little more. But it still feels a little like (and I'm repeating myself here) "I had to go uphill in the snow to school - both ways!"

The tools and technology have advanced to the point where spitting out a perfectly serviceable song requires little more than a rudimentary ability to describe it.

Beyond that? I don't know why we'd care - we already listen to highly processed, transformed, produced 'music' that is so far from the pure "instrument to ear" experience that we have to have banks of processors and computers help fake it live to even get close.

We would never accept someone submitting a photograph created with prompts to the coaster db.

I'm going to be very 'CoasterBuzz' on you here. I apologize in advance.

No, but I'd absolutely enjoy it as an interesting image. And I think it'd fit on a site full of interesting roller coaster imagery/art. The db is documentation. Generation is not.

That's why that comparion falls apart for me.

An AI song generated by a toddler accidentally mashing the keyboard in the prompt box on Suno is still a song in the end. That AI generated photo is not documentation of real life, but it's likely a great image that is worthy of view.

It does make some amount of sense as an argument...and I only partially disagree.

LostKause:
The urge to write a song is sometimes unignorable. Sometimes I wake up in the morning humming a song that does not exist.

You do understand you can hum that to AI and create a song? Or parts. Or whatever.

In fact, now we're touching on what my latest stuff does. (again, not self-promotion but I've been sitting on whole set of 10 more of those 'southern hard rock' tunes and the following is in reference to those)

You can absolutely put audio into these generators.

I can't play guitar or bass or anything at an effective level beyond drums. I can't even really sing well. But I understand how to write a song and put music together. (oh my god, I'm Lars!)

So I have lyrics written. And in my head the guitar should go "Dunnnn dun dadadaa dunn dunn dunnnn dada" (imagine me looking like an idiot air guitaring and singing that guitar part)

At this point, I give that to Suno with some prompt info. It'll spit back...something. Sometimes it's what I expect. Often it's not. Sometimes it doesn't work. Sometimes it's better than my original idea. Once I form, transform, remix, regenerate based on existing, playing with generation settings and stuff, I can rip the guitar stem and put it in the DAW.

At this point what I do is no different than if I (or a guitar player) and recorded that track. Now I can do that for every instrument.

I also have an electronic kit. I haven't gone for it and full replaced AI drum tracks with my own playing, but I have punched in fills and percussion things.

And you guys both mentioned rap music. Rappers are benefitting most from this, I think. Know how easy it is for these guys to pump out top-notch backing tracks to work over? Stupid easy.

Useful innovation usually increases accessibility and reduces work required to complete a task.

AI music generation does both.

Lord Gonchar:

I'm not sure you fully understand how to use AI creatively and not lazily.

Those who use it as a tool, and not simply write a prompt and post a completed song, are not what I feel is threatening to me.

Is it the idea that someone takes 'credit' for 'writing' a song?

That might be a part of my concern. "Look what I wrote!!!" then you realize that they didn't really write it. They told a computer to write it for them.

You do understand you can hum that to AI and create a song? Or parts. Or whatever.

That's very interesting to me. Because like you, I'm not the greatest guitar player, and although I have a bass, no matter how hard I try, I'm not good at playing it.

Lord Gonchar:
So I have lyrics written. And in my head the guitar should go "Dunnnn dun dadadaa dunn dunn dunnnn dada"

That's how I collaborated with past band mates. "No, no! I think the drums should go Da-dat-DUUDut-duuuu!" LOL

Lord Gonchar:
...and put it in the DAW.

I do recall you saying sometime, somewhere, that you do more than just write a prompt and tell it to spit out a completed song. You just won me over in the argument. I totally understand now. I still feel icky about it, but I understand.

I mean, it's to the point that I might even try it out myself sometime. Damn.

Sorry for the double post.

I see those metal songs covered as soul songs by AI on Facebook sometimes. They are awesome! I wish all Soul songs were that cool. But I never re-post them because it might offend my musician friends. Listening to them makes me feel icky too, because AI.

Those song were mentioned in this conversation, so I felt I had to comment about them.

Lord Gonchar:

Know how easy it is for these guys to pump out top-notch backing tracks to work over?

LostKause:
Those who use it as a tool, and not simply write a prompt and post a completed song, are not what I feel is threatening to me.

I just don't think it sounds good and the artists that may be using it are just not what I am into. So maybe it's just down to taste.

I don't think Kanye West is writing The College Dropout with AI. We’re not getting Low End Theory, Return to the 36 Chambers, The Chronic, The Marshall Mathers LP, etc, with AI especially if you take away all of the source material the model would have been trained on.

I also find autotune and pop country to be an abomination so again... taste.

I wrote a whole lot more but ended up deleting it all. One record I was going to mention is Soul Blind's Red Sky Mourning. They are obviously inspired by Alice in Chains, Hum, Helmet, all of that late 90s rock and shoegaze but they have their own modern take. That whole album has a quality to it that AI can't replace.

eightdotthree:

...with AI especially if you take away all of the source material the model would have been trained on.

Two things:

1. All music ever (and certainly the commercial music of the last 50-100 years) is influenced or derivative - no one plays an instrument or writes a song totally in a bubble. That's how music has been made forever - you watch others do it and emulate it in your own way.

2. You mentioned 90's rap as a great example of transformative work. They literally took someone's work and re-worked it. I mean, that can't exist without the original work. It's not just based on the original work, the core IS the original work.

What AI does is actually closer to #1 than #2.

LostKause:
That's very interesting to me. Because like you, I'm not the greatest guitar player, and although I have a bass, no matter how hard I try, I'm not good at playing it.

Here's some weirder grey arera to help twist up your grey matter a little more on this fine December afternoon.

We've talked about the idea of letting AI generate a song then recording/performing it as your own. But...

What if I take an entire song I wrote and recorded myself - but it didn't sound real great because...well, realistic limitations in skills, equipment and the recording, mastering process - and I run the whole damn thing through AI?

I can give an example.

A buddy's song from 2015:

I dug it. Though it had a real "Corrosion Of Conformity" vibe. The raw recording lends itself to the song, but I couldn't help but wonder what it would be like if it were a little more 'studio'. So I made this (and he knows and thought it was cool):

I think it's incredible technology. A little like having that mythical "good" button. But at what point do you guys find it disingenuous?

Can someone upload the AI "fixed" version and call it their own work?

Is it really doing much more than autotune or other studio "tricks" do to make a performance more than it is? (I mean, we all know Lars has timing issues, so those studio tracks are 'creations' to some degree)

The moment you add any edit/effect to a performace, it's been modified.

You should have made that a blind test. His drums aren't great but the guitar tone and his vocals are really good. Your version just doesn't sound right to me, similar to how you said your narration experiments didn't. It doesn't sound real. But again. My taste is different so maybe it's just me.

Lord Gonchar:

They literally took someone's work and re-worked it

I know it seems contradictory. I don't really know how to articulate it. Someone had to experience that work in their life then isolate it and create something out of nothing. It's gritty and real. It's the opposite of what I've heard AI models produce.