Why you shouldn't use AI to 'compose' music

[PatrickJazzGuitar]

(I apologize if this is the wrong forum),

Copyright law doesn't protect AI. That's why. So, that's important if you intend on selling your music and want copyright protection of your original compositions.

I've copyrighted (meaning, 'registered with the US Copyright Office') maybe some 40 or so compositions, and I've studied copyright law. (yes, I know a song is technically copyrighted the moment it's put into the real world, but it means squat if you are sued and you didn't get it registered [because no lawyer will take the case, the law won't back you up] ).

I'm not a lawyer, so watch this video by a copyright lawyer regarding AI. the long and short of it is that AI music cannot
be protected by copyright law. And if you are seeking sync licenses for a TV show, or film, contracts always have you assure
them that you have full authorship, or co authorship (with another human) which equal 100%. Yes, there are grey areas, like drum tracks, AI contributions, but overall, the composition has to be substantially yours. don't ask me about rap songs, I have no idea how they work in copyright law. but this lawyer gets into the weeds of the subject, and it's good stuff to know.

If you don't have time to watch it, here's quick (AI created, ironically) summary:

In the YouTube video titled "Udio AI Terms Of Service Exposed: Do You Legally Own AI Music?", a music attorney discusses the copyright implications of using AI music platforms, specifically focusing on Udio. Under US copyright law, an original work of authorship assumes human creation, and the Copyright Office has refused to register works made from AI. Udio's terms of service identify the company as Uncharted Labs Inc. and include a waiver of class action rights and an arbitration clause. The platform generates various types of content, including music, and users retain no copyright over the output they generate using the platform. The input provided by users is considered minimal and disregarded under copyright law. The speaker also discusses the trademark situation between Udio and another company with the same name in the same class. The user grants Udio and its associates the rights to reproduce, transmit, publish, display, and modify their content to operate, improve, and promote the services. The speaker raises concerns about copyright ownership for music created with AI and the potential impact on various industries, such as music production and mastering. The speaker invites viewers to share their thoughts on the issue.

[RJVB]

I'm fully expecting that one day the fine print in the TOS of those AI gadgets will state that the company behind them gets to own the copyright of everything they generate at your request.

FWIW Facebook (and so possibly Instagram too) are about to launch some kind of AI that they will train off our content. Maybe they already did elsewhere but here in the EU it's supposed to go live the 26th of this month ... and we have a right to object to the (direct) use of our content. You get a notification about the AI which has the relevant link(s); I filed an objection as soon as I got that alert. The form is presented as a "request that will be studied" but for the 2 accounts where I went through the motions I got almost immediate confirmation that they will respect my demand.

[Kirk Garrett]

I generally think the most promising creative use of AI is in coordination with human effort. I understand that in the chess world, the strongest performer isn’t a computer, but a human using a computer. I can’t imagine how such a combination would be seen in copyright law.

I’ve yet to see AI do something in visual art or in music that indicates some kind of step forward, but perhaps there will be a way to increase the human contribution enough to do something more eye opening when they work in combination.

[Bop Head]

I'm fully expecting that one day the fine print in the TOS of those AI gadgets will state that the company behind them gets to own the copyright of everything they generate at your request.

FWIW Facebook (and so possibly Instagram too) are about to launch some kind of AI that they will train off our content. Maybe they already did elsewhere but here in the EU it's supposed to go live the 26th of this month ... and we have a right to object to the (direct) use of our content. You get a notification about the AI which has the relevant link(s); I filed an objection as soon as I got that alert. The form is presented as a "request that will be studied" but for the 2 accounts where I went through the motions I got almost immediate confirmation that they will respect my demand.

Change to Adobe terms & conditions outrages many professionals - 9to5Mac

reuters.com

reuters.com

[Ukena]

[...] and we have a right to object to the (direct) use of our content. You get a notification about the AI which has the relevant link(s); I filed an objection as soon as I got that alert. The form is presented as a "request that will be studied" but for the 2 accounts where I went through the motions I got almost immediate confirmation that they will respect my demand.

You only have their word that they will respect your demand. We have a pretty good idea how well corporations honor commitments like that.

As for the OP, if I have AI write a composition, how can anyone tell it is generated by AI? If it is generated as a score, I can copy/paste or transcribe that score into MuseScore, export it as a PDF, and claim I composed it as I register/copyright it.

[RJVB]

I understand that in the chess world, the strongest performer isn’t a computer, but a human using a compute

Is that true? Chess is in fact a purely deterministic game from what I understood, one where every move leaves the opponent with a limited number of choices and where it's thus possible to predict all possible outcomes at any given time. Guess who's better at that particular game...
Human intellect may play in coercing the opponent to react in a certain way and that can include reading the other's reactions but I'd be surprised if self-learning algorithms couldn't be written that glean the relevant information from the moves the opponent makes, and how.

<buncha urls>

In the future, could you at least summarise the gist or why I'd want to go clicking?

You only have their word that they will respect your demand. We have a pretty good idea how well corporations honor commitments like that.

True, but it's better than nothing and there are watchdogs that will be watching.

As for the OP, if I have AI write a composition, how can anyone tell it is generated by AI?

No one, as long as you also wrote the AI yourself and take care that every trace of you using it.

This is self-learning software. It'll also learn from anything you make it do, which means that something very similar to what it made for you can end up in the results for someone else. And then what?

[Kirk Garrett]

Is that true? Chess is in fact a purely deterministic game from what I understood, one where every move leaves the opponent with a limited number of choices and where it's thus possible to predict all possible outcomes at any given time. Guess who's better at that particular game...
Human intellect may play in coercing the opponent to react in a certain way and that can include reading the other's reactions but I'd be surprised if self-learning algorithms couldn't be written that glean the relevant information from the moves the opponent makes, and how.



In the future, could you at least summarise the gist or why I'd want to go clicking?



True, but it's better than nothing and there are watchdogs that will be watching.



No one, as long as you also wrote the AI yourself and take care that every trace of you using it.

This is self-learning software. It'll also learn from anything you make it do, which means that something very similar to what it made for you can end up in the results for someone else. And then what?

Is that true? Chess is in fact a purely deterministic game from what I understood, one where every move leaves the opponent with a limited number of choices and where it's thus possible to predict all possible outcomes at any given time.

Yes, while chess (unlike life) is definitely a closed universe, it is not fully deterministic simply because there is an opponent who can make choices.

Open up that universe and increase the number of variables, and the human element becomes even more important.

Why Computer-Assisted Humans Are The Best Chess Players And What That Means For Technology Operations

[RJVB]

Why Computer-Assisted Humans Are The Best Chess Players And What That Means For Technology Operations

You'll forgive me if I don't take Forbes for a definitive source here, nor the fact that Kasparov could beat a dedicated supercomputer in 1997 as the end-all conclusion that a human is required. In fact, I seem to recall that he lost in a later rematch.
Chess being deterministic means formal proofs are possible, and if memory serves me well you won't be able to beat a player who is capable to assess all possible outcomes of all possible reactions to each of his/her/its currently possible moves within the allowed response time. Not unless you're able to do the same yourself. If you are, it's simply the opening move that will determine who wins.

There are many systems where predictions become untenable as soon as you put a human in the loop (robotics is a good example as many control theorists discover when they start working on driving or flight simulators) but AFAIK chess isn't one of them because humans are bound to a narrow set of rules.

[fep]

I play chess.

Humans beating machines at chess are a thing of the past. From this article: Can A GM And Rybka Beat Stockfish? - Chess.com

(Magnus Carlsen is considered the best player in the world, many think he's the best of all time)

It is well established that the days of human-computer rivalry are long gone. In a four- or five-game match, even Magnus Carlsen will stand no chance against Houdini or Stockfish running on decent hardware.

[Kirk Garrett]

I play chess.

Humans beating machines at chess is a thing of the past. From this article: Can A GM And Rybka Beat Stockfish? - Chess.com

(Magnus Carlsen is considered the best player in the world, many think he's the best of all time)

The contention isn’t that humans play chess better than computers, but that the strongest player is a human or a team of humans in combination with a computer.

[RJVB]

The contention isn’t that humans play chess better than computers, but that the strongest player is a human or a team of humans in combination with a computer.

Try to think of this as an equation where the chessboard is the equal sign.

If you can bar the computer on both sides because they are equals you're left with a "doh" situation (human vs. no one).

Still, I don't see what a human could bring in in this context; the relevant capacities of our brain are already limited w.r.t. those of computers and that situation isn't likely to improve in our favour.

[Kirk Garrett]

Sorry about the chess sidetrack. Evidently the chess players here say that a human contributes nothing to AI in a closed, solved system like chess. Which is not surprising, though I’ve heard and read differently.

I think in a creative field the two will work together most of the time.

[fep]

The contention isn’t that humans play chess better than computers, but that the strongest player is a human or a team of humans in combination with a computer.

Okay, I read the article. At best it seems the article presents an untested hypothesis and there was no evidence or actual games presented as examples.

I hypothesize that the human would just slow down the computer and given any normal chess tournament time constraints the alone computer would win.

[fep]

Interesting sidetrack.

Back on topic, I have been following "Top Music Attorney" - the woman in the YouTube video of the original post. She does a great job of explaining the law and what musicians should be on the lookout for. Definitely something useful for working musicians. And, it's not just the AI that she posts youtube videos about, also copyright law, should you register your copyrights, split sheets, setting up an LLC, setting up your own publishing company, etc.

[Kirk Garrett]

Okay, I read the article. At best it seems the article presents an untested hypothesis and there was no evidence or actual games presented as examples.

I hypothesize that the human would just slow down the computer and given any normal chess tournament time constraints the alone computer would win.

So the AI creative model is probably: AI enhances human effort. Not the other way around. In chess, AI is also much superior to humans, but it’s only a game and nobody wants to watch AI tournaments, so human participation survives.

In the arts, will AI enhance human effort or replace it? It’s easy to envision the former. When AI writes a classic song by itself, it will be easier for me to envision the former.

I think we’re going to be cyborgs of some kind….

[CliffR]

Chess being deterministic means formal proofs are possible, and if memory serves me well you won't be able to beat a player who is capable to assess all possible outcomes of all possible reactions to each of his/her/its currently possible moves within the allowed response time.

A formal proof of chess? What exactly would that look like? No player, nor computer, is capable of assessing possible outcomes to the end game of any given move in the early-mid stages of a chess game in a reasonable amount of time. The number of possible combinations are way too high.

"Try to think of this as an equation where the chessboard is the equal sign.

If you can bar the computer on both sides because they are equals you're left with a "doh" situation (human vs. no one)."

I have no idea whatsoever what you're attempting to say here.

[RJVB]

No player, nor computer, is capable of assessing possible outcomes to the end game of any given move in the early-mid stages of a chess game in a reasonable amount of time. The number of possible combinations are way too high.

That's the kind of assumption we should have learned not to make anymore in absence of formal proof. Actual ("real") chess is not what computer scientists and mathematicians call NP complete: see e.g. Is Chess NP complete or NP hard? - Math of Games - Quora


To get back on the other part of the topic: Dag van de Componist – New Music Now
(sorry Dirk ... Belgium used to be part of the Netherlands )

[Stevebol]

Recent advances in A.I. are going to be over the head of everyone. When you have to have a lawyer to even consider being a musician these days, something is off.

[Stevebol]

Lawyers know as much about AI as politicians know about climate change. I thought 2023 was going to be the end of the world as we know it because of drought in the western states.
Then it rained, and snowed like crazy.

The only I know for sure is, I know nothing.

[emanresu]

If I was owner of some AI system, my first concern would be how to get money from the stuff it produces.

[RJVB]

To get back on the other part of the topic ...

Or ... someone ask their favourite Mozai to compose a jazz version of this (with new music video, preferably)?

[Ukena]

It's already basically beyond should or shouldn't; it's a fait accompli: