[don-code]

A few months ago I spoke with the frontman of a local Boston band from the 1980s, who recently re-released a single with the help of AI. The source material was a compact cassette tape from a demo, found in a drawer. He used AI to isolate what would've been individual tracks from the recording, then cleaned them up individually, without AI's help.

Does that constitute "wholly or in substantial part"? Would the track have existed were it not for having that easy route into re-mastering?

I understand what Bandcamp's trying to do here, and I generally am in support of removing what we'd recognize as "fully AI-generated music", but there are legitimate creative uses of AI that might come to wholly or substantially encompass the output. It's difficult to draw any lines line on a creative work, by just by nature of the work being creative.

(For those interested - check out O Positive's "With You" on the WERS Live at 75 album!)

[anigbrowl]

No I don't think that really qualifies because it's solving an engineering problem. I hang out on an electronic music creators' forum which is stringently anti-AI, but nobody objects to things like stem separation. People are skeptical about AI 'mastering' but don't really object for similar reasons.

What people get mad about is the use of AI to generate whole tracks. Generating rhythms, melodies, harmonies etc via AI isn't greeted warmly either, but electronic musicians generally like experimenting with things like setting up 'wrong' modulation destinations in search of interesting results. I don't think anyone seriously objects to AI-produced elements being selected and repurposed as musical raw material. But this is obviously not happening with complete track generation. It's like playing slot machines but calling yourself a business person.

[crtasm]

That's not AI generated at all. Using acoustic models to stem out individual sections from a recording is not creating new material (and I wouldn't even describe that as "AI" despite what I'm sure a lot of the tools offering it want us to believe).

[tensor]

Sorry, that's AI. So is OCR, so is voice recognition, and many other things you probably use and take for granted. I'd suggest you focus on use cases not trying to redefine definitions for an entire area of science and technology based on your own preferences.

Saying "I'm against fully AI generated music" is at least precise, and doesn't throw out detecting cancer along with the AI bandwagon term.

[marcianx]

I think it makes some sense to allow leeway for intelligent "signal processing" using AI (separating out individual tracks, clean-up, etc) vs generating new content with AI.

Similarly, say, for video editors, using AI to more intelligently rotoscope (especially with alpha blending in the presence of motion blur - practically impossible to do it manually), would be a great use of AI, removing the non-creative tedium of the process.

It's not clear where the line is though. I was quite impressed with Corridor Crew's (albeit NVidia+Puget-sponsored) video [1] where they photographed dolls, motion-captured human actors moving like the dolls, and transferred the skeletal animation and facial expressions to those dolls using GenAI. Some of it required nontrivial transformative code to accommodate a skeleton to a toy's body type. There's a massive amount of tedium being removed from the creative process by GenAI without sacrificing the core human creative contribution. This feels like it should be allowed -- I think we should attempt to draw clearer lines where there are clearly efficiency gains to be had to have less "creative" uses be more socially acceptable.

[1]: https://youtu.be/DSRrSO7QhXY

[lossyalgo]

It may be a tedious job to spend days rotoscoping but I personally know people who get paid to do that, and as soon as AI can do it, they will have to go find other work (which they already do, on the side, because the writing is on the wall, but there's a ton of people worldwide who do this kind of work, and that's not the only process being delegated to AI).

[amanaplanacanal]

That feels legit to me. We have been using software to isolate individual instruments from a recording for a while.

[kazinator]

We've been using software to fix grammar for a long time, and AI does it also. The question is valid: if I get an LLM to fix a few grammar errors in my own writing, am I ripping anyone off? We can't just dismiss the question just because grammar fixing is something we did without machine-learning AI trained on vast numbers of other people's texts.

The output does depend on training works, even if you are just fixing grammar errors. But the document is obviously a derivative of your own writing and almost nothing else. A grammatic concept learned from vast numbers of worsk is probably not a copyright infringment.

Similarly, a part extraction concept learned from training sets such as pairs of mixed and unmixed music, and then applied to someone's own music to do accurate part extraction, does not seem like an infringing use. All features of the result are identifiable as coming from the original mixed audio; you cannot identify infringing passages in it added by the AI --- and if such a thing happened, it would be an unwanted artifact leading us to re-do the part extraction in some other way to avoid it.

[nottorp]

The question doesn't feel legit to me though. The OP somehow found the one justifiable example among a sea of AI slop.

Justifiable because there were some filters. That may not even have been "AI". They may have been some deterministic algorithms that the software maker has to label "AI" because they otherwise think it won't sell...

[strangecasts]

I think any line is necessarily going to be arbitrary, a blanket ban on any ML model being used in production would be plainly impossible -- using Ozone's EQ assistant or having a Markov chain generate your chord progressions could also count towards "in substantial part", but are equally hard to object to.

But we also live with arbitrary lines elsewhere, as with spam filters? People generally don't want ads for free Viagra, and spam filters remain the default without making "no marketing emails" a hard rule.

The problem isn't that music Transformers can't be used artfully [1] but that they allow a kind of spam which distribution services aren't really equipped to handle. In 2009, nobody would have stopped you from producing albums en masse with the generative tech of the day, Microsoft's Songsmith [2], but you would have had a hard time selling them - but hands-off distribution services like DistroKid and improved models makes music spam much more viable now than it was previously.

[1] I personally find neural synthesis models like RAVE autoencoders nifty: https://youtu.be/HC0L5ZH21kw

[2] https://en.wikipedia.org/wiki/Microsoft_Research_Songsmith as ...demoed? in https://www.youtube.com/watch?v=mg0l7f25bhU

[Isamu]

>The source material was a compact cassette tape from a demo, found in a drawer.

Was this demo his, or someone else’s IP? If he is cleaning up or modifying his own property, not a lot of people have a problem with that.

If it is someone else’s work, then modifying with AI doesn’t change that.

I think they just don’t want AI generated works that only mash up the work of other artists, which is the default of AI generated stuff.

[xoxxala]

If the Beatles can use AI to restore a poorly recorded cassette tape of John Lennon playing the piano and singing at his dinner table, I think it's alright if other bands do it, too.

https://en.wikipedia.org/wiki/Now_and_Then_(Beatles_song)

[delusional]

> It's difficult to draw any lines line on a creative work, by just by nature of the work being creative.

If you want to be some neutral universal third party sure. If you're OK with taking a position, the arbitrariness actually makes it much easier. You just draw the line you want.

Creativity demands limitation, and those limitations don't have to be justified.

[kazinator]

Suppose we have two image-oriented AI's.

One is trained with a set of pairs which match words with images. Vast numbers of images tagged with words.

The other is trained on a set of photographs of exactly the same scene from the same vantage point, but one in daylight and the other at night. Suppose all these images are copyrighted and used without permissions.

With the one AI, we can do word-to-image to generate an image. Clearly, that is a derived work of the training set of images; it's just interpolating among them based on the word assocations.

With the other AI, we can take a photograph which we took ourselves in daylight, and generate a night version of the same one. This is not clearly infringing on the training set, even though that output depends on it. We used the set without permission to have the machine extract and learn the concept of diurnal vs. nocturnal appearance of scenes, based on which it is kind of "reimagining" our daytime image as a night time one.

The question of whether AI is stealing material depends exactly on what the training pathway is; what it is that it is learning from the data. Is it learning to just crib, and interpolate, or to glean some general concept that is not protected by copyright: like separating mixed audio into tracks, changing day to night, or whatever.

[kouteiheika]

> With the one AI, we can do word-to-image to generate an image. Clearly, that is a derived work of the training set of images

> The question of whether AI is stealing material depends exactly on what the training pathway is; what it is that it is learning from the data.

No it isn't. The question of whether AI is stealing material has little to do with the training pathway, but everything to do with scale.

To give a very simple example: is your model a trillion parameter model, but you're training it on 1000 images? It's going to memorize.

Is your model a 3 billion parameter model, but you're training it on trillions of images? It's going to generalize because it simply doesn't physically have the capacity to memorize its training data, and assuming you've deduplicated your training dataset it's not going to memorize any single image.

It literally makes no difference whether you'll use the "trained on the same scene but one in daylight and one at night" or "generate the image based on a description" training objective here. Depending on how you pick your hyperparameters you can trivially make either one memorize the training data (i.e. in your words "make it clearly a derived work of the training set of images").

[cess11]

You don't make it clear whether the music on that tape was "generated" "by AI", only that it was post-processed in such a way.

[qmmmur]

That's not generative AI, just source separation which has existed way before large language models and transformer architecture were big.

[rfw300]

The example you present seems fairly straightforward to my intuition, but I think your point is fair.

A harder set of hypotheticals might arise if music production goes the direction that software engineering is heading: “agentic work”, whereby a person is very much involved in the creation of a work, but more by directing an AI agent than by orchestrating a set of non-AI tools.

[kazinator]

This is very similar to, "am I ripping people off if I just get the LLM AI to make a few grammar fixes in my own writing?"

[sodapopcan]

Ya, "AI" is too broad a term. This was already possible without "AI" as we know it today, but of course it was still the same idea back then. I get what you're saying, though: would he have bothered if he'd have to have found the right filters/plugins on his own? idunno.

[cultofmetatron]

thats sounds more like unsupervised learning via one of the bread and butter clustering algorithms. I guess that is technically AI but its a far cry from the transformers tech thats actually got everyone's underwear in knots.

[pier25]

AI unmixing or denoising is not really generating music AI.

[VBprogrammer]

Where does it stop. My dad is a decent guitarist but a poor singer (sadly, I'm even worse). He has written some songs, his own words, some guitar licks or chords as input material and AI turning it into a surprisingly believable finished piece. To me it's basically AI slop but he's putting in a modest amount of effort for the output.

[interludead]

This is exactly where the policy starts to get murky

[mcpar-land]

That's not "generating" the music with AI - that's isolating the tracks of existing music. Probably not generative AI at all, and depending on who you ask, not even AI.

This is why it is to these generative AI companies' benefit that 'AI' becomes a catchall term for everything, from what enemies are programmed to do in video games to a spambot that creates and uploads slop facebook videos on the hour.