NY Times "Who Owns a Song Created by A.I.?"

[dubbmann]

Hi all,

Debated whether to add this to the existing thread on AI IP or start a new one on the topic. Since the existing thread has gone in somewhat different direction, thought I'd go with Option 2. Anyway, here's the article. For those interested there are some interesting links in the original. Its URL is

https://www.nytimes.com/2023/04/15/busi ... right.html

Lawmakers are beginning to contemplate questions about authorship and ownership around creative machines. The stakes for creative businesses are high.

By Ephrat Livni, Lauren Hirsch and Sarah Kessler
April 15, 2023, 8:00 a.m. ET
7 MIN READ

Artificial intelligence tools that generate text, images and music are moving art into new territory — and that’s raising tricky questions for the business of creativity.

For early adopters like Insider, the publication that this week announced an experiment with A.I.-aided articles, the new tools promise more efficient content creation. But for many artists, and the businesses that own their work, generative A.I. is a double threat. These systems can produce copycats of human works that dilute the market, and they use artists’ production, without their permission, as training data.

Some see that as stealing intellectual property: Universal Music Group recently told music streaming platforms, including Spotify and Apple, to block A.I. systems from scraping its music. (The company is in early discussions to license its songs to generative A.I. companies, DealBook hears.)

Lawmakers have begun to contemplate new rules around authorship and ownership in connection with creative machines, and the stakes are huge for both the businesses that depend on creative work and the investors who poured billions into new A.I. tools. So far, there are three major debates.

What is owed to the creators of the original material? In January, a group of artists sued London-based Stability AI, a maker of image-generating software, arguing that it infringed on their copyrights by using their work in training data and creating derivative works. The cartoonist Sarah Anderson, who is part of the lawsuit, told The New York Times that she believed artists should opt in to having their work included in such data, and should be compensated for it. Getty Images is also suing Stability AI in Britain and the United States for what it calls “brazen infringement” of millions of photos. Getty argued that the theft is particularly offensive because it has agreements to license data for machine learning. Stability AI has not yet responded to the complaints.

Does “fair use” apply? Copyrighted works can be used without permission for commentary, criticism or other “transformative” purposes, and robots have traditionally been exempt from liability. But “courts in the future won’t be so sympathetic to machine copying,” wrote Mark Lemley, the director of a Stanford Law School program that focuses on science and technology, in the Texas Law Review with a former colleague, Bryan Casey. Lemley is calling for a new “fair learning” standard for using copyrighted material in machine learning. It would include the question: What is the purpose of the copying? If it’s to learn only, that may be permitted, but if the intent is to reproduce the work, it will not be. Not every machine learning data set would qualify for the protection. New tools also raise questions about who has liability for infringement — the user prompting the machine, the company that programmed the tool or both?

Who owns the output of generative A.I.? For now, only a human’s work can be copyrighted, but what about work that partly relies on generative A.I.? Some tool developers have said they won’t assert copyright over content generated by their machines. In February, the Copyright Office rejected a copyright for A.I.-generated images in a graphic novel, though the writer argued that she had made the images via “a creative, iterative process” that involved “composition, selection, arrangement, cropping and editing for each image.” The government compared use of the A.I. tool to hiring an artist. But the lines may blur as the use of such tools becomes more common. Like the tools, the intellectual property issues are a work in progress that will only get more complex. — Ephrat Livni

[Drew Stephenson]

I think this is one of those things where the best time to have had this debate is 5 years ago, but the next best time is now.
I think it's also right to ask the question, "does anyone need to own it?"
An AI doesn't need an income stream from a limited-term monopoly to keep being incentivised to create, so why would we put any restrictions on it at all?
Or if we decide restrictions are required, maybe copyright and ownership aren't the right tools for those restrictions.

I think the interesting stuff will be where there are human/AI creations. How much human element is required for a work to qualify for copyright?
It's a bit like the question of how much a work needs to differ from a previous piece to qualify as a new work.

What happens if you train your AI on work that's all in the public domain or has no copyrightable elements?

[Folderol]

The whole subject is as slippery as an eel. There was a time when people were paid to be musicians, full stop. What they played and sang was available to anyone.

[OneWorld]

The lawyers will create so much work for themselves that they’ll have the Lamborghini dealers on speed dial. I am sure artificial intelligence will give rise to artificial conflict requiring artificial solutions, but one road or the other someone will dream up a way of monetising the issue and right now there’ll be those claiming to be specialists in artificial solutions. But surely if copyright belongs to an AI botty then the bot itself takes a case a case to court. Seems much ado about nothing to me

[RichardT]

I think the training data issue is a red herring. It’s exactly what people who write music do, often without even realising it. That’s why most music sounds very much of its time.

I think the issues with AI will come into focus only when AIs start producing human-quality music.

I think two main issues will really come to a head - the copyright questions, and the effect of replacing of human composers by AIs.

Unlike previous replacements of people by technology, such as real string players being ‘replaced’ by sample libraries, I’m guessing that this time the quality of AI music will soon be just as good as the real thing. I don’t doubt that AI will soon be able to improvise very good jazz solos, for example.

I think it will come down to whether people want AI generated music or not. In some contexts, I’m sure people will be happy - incidental music for example. But I’ve no idea about music that really matters to people.

[Drew Stephenson]

Unlike previous replacements of people by technology, such as real string players being ‘replaced’ by sample libraries, I’m guessing that this time the quality of AI music will soon be just as good as the real thing. I don’t doubt that AI will soon be able to improvise very good jazz solos, for example.

I think it will come down to whether people want AI generated music or not. In some contexts, I’m sure people will be happy - incidental music for example. But I’ve no idea about music that really matters to people.

I agree on this. I think AI-generated music (and other content) will become commonplace in background scenarios. And I for one won't miss the old elevator-muzak...
But I think there will still be demand for 'hand-made' music in the same way that there is demand for other hand-made goods now.
However I think that also suggests a considerably smaller consumer market (even as the barriers to entry come down), so not only will the pie get smaller but more people will be taking bites.
This could amplify the two-tier nature of 'hand-made' music where, for a very few at the top, it is a profitable career choice with sponsors, patrons, and very expensive exclusive access, and for the rest it becomes a hobby. The journeyman role is the area that will decrease or disappear I think.
My initial thoughts were that live music would still remain as a 'hand-made' avenue but I can easily envisage a scenario where a virtual pop band perform a holographic show of AI music (complete with new improvisations and variations of previous material) to a live crowd. In time I'd expect the tech to be able to interact with that live crowd as well.

Music technologist is going to be an interesting career for some.

[DC-Choppah]

Music is a human activity not a product. Lucky for me human made music is very much alive and well, in demand, and growing.

Can you imagine showing up at a place like this (https://www.keystonekornerbaltimore.com) trying to pull of an AI generated music show or sell AI generated music?

People pay for the human connection and value the human experience. Human connection is rare and highly valued.

The food and drinks are great too!

Only humans can be authors. Case closed.

[ef37a]

Just FYI chaps? There is a programme about AI and music on Radio 3 as I type.

Dave.

[Drew Stephenson]

Music is a human activity not a product.

Some music.

Lucky for me human made music is very much alive and well, in demand, and growing.

Lucky for me too.

Can you imagine showing up at a place like this (https://www.keystonekornerbaltimore.com) trying to pull of an AI generated music show or sell AI generated music?

Nope. But in a huge stadium where half the audience can't actually see the stage properly and rely on big screens and megawatts of lighting and sound reinforcement. Absolutely. In k-pop and j-pop you've already virtual bands giving live gigs.

People pay for the human connection and value the human experience. Human connection is rare and highly valued.

Again, some people. Which is something I'm very happy about.

Only humans can be authors. Case closed.

For now. There are already lobbying efforts going on to get patents and copyrights assigned to AI creators. As I've said elsewhere, I think that would be a travesty and probably the final nail in the coffin of any kind of wider respect for copyright law from the general public, but there's huge amounts of lobbyist cash washing around this at the moment. The DABUS / Thaler case will only be the first of many I reckon.
On a related note, I see that Stable Diffusion have filed anti-slapp suit in their case and, frankly, it looks a lot stronger than the original complaint.

[OneWorld]

"Who Owns a Song Created by A.I.?"

Whomsoever can afford the best lawyer

[Drew Stephenson]

Or lobbyist...

[Drew Stephenson]

On a lighter side: http://www.sheldoncomics.com/archive/230424.html

[DC-Choppah]

Folks give too much credit to AI when it is simple optimization. It is literally just an Oracle data base of stuff and some algorithms to maximize some objective function. It cannot create anything new and is literally as interesting as talking to an automatic phone answering machine.

Ya'll are being scammed if you think there is anything remotely like intelligence here.

[Drew Stephenson]

For what it's worth I think you are oversimplifying a very complex and varied set of approaches to machine learning.

[ManFromGlass]

To me it looks like you both said almost the same thing.

But I’m not so sure the point about creating something new is accurate.
What is the difference between Ai scraping some chords together and me ending up with the same? Did we both create something new? I’m leaning towards yes as the answer. Leaving out any judgement of quality or legal issues like copyright. Even leaving out the idea of machine learning.

I could assign chords to numbers on dice and then roll them to get my chord progression. Did I create something new? Is Ai just a sophisticated pair of dice?

[BigRedX]

Also from what I've seen all AIs require some kind of user input in order to generate whatever they are designed to do. Therefore whoever provides the user input (inspiration) should be credited with the composition.

[Drew Stephenson]

Also from what I've seen all AIs require some kind of user input in order to generate whatever they are designed to do. Therefore whoever provides the user input (inspiration) should be credited with the composition.

This is one of the big legal questions that will doubtless get kicked around the courts soon enough. How much input does it take for it to count as a material contribution?

[amanise]

The training of the AI seems no different to me than the process a person goes through listening to records and the like as they grow - admittedly the AI training happens in a much shorter time frame. As the musician develops a style - so would the AI. Except the AI would be able to develop many more styles. (using myself as a gauge here - I hasten to add).

Who owns the song would seem to be about the licensing arrangements around the software producing it - and there are many models for that as we all know. All the copyright arguments would be established just as they are now, by precedent in test cases with the most likely winners being those who can afford the best lawyers. Just as it is now. I find it hard to imagine an AI playing its guitar in court to defend itself though.

I don't see AI making much difference to the global music biz such as it is - apart from enabling the people who already have all the money - to concentrate even more of it in their own hands. As for the quality of the music - hasn't that horse already bolted some time ago? We may even see some improvement!

Now medicine... if you want an AI can of worms to unpick....

[ManFromGlass]

Ai is going to totally replace the use of composers for almost all low and medium budget film, tv, radio, elevator, shopping mall, dentist office and game scores. Producers of those projects are already squeezed budget-wise to complete their projects with less and less money every year.

I think the person who said there will be a new skillset required to tell the Ai what music flavours the show needs is correct. I can’t remember the name suggested, was it Music Technologist instead of composer?

[The Red Bladder]

I think the person who said there will be a new skillset required to tell the Ai what music flavours the show needs is correct. I can’t remember the name suggested, was it Music Technologist instead of composer?

I could suggest the most appropriate word - but I seem to remember that, that word is not allowed on this forum.

[Drew Stephenson]

Ai is going to totally replace the use of composers for almost all low and medium budget film, tv, radio, elevator, shopping mall, dentist office and game scores.

Low, yes. Medium, I'm not so sure. Or at least, not yet.
The reason I query this is because of the late and rapid requirements change that comes with film and TV. If you're just looking for some background music for a youtube video then yes, stick half-a-dozen words into an AI generator and let it do its thing in the background.
But where there's slightly more attention to detail you might get a request of "Like cue 1 but with the strings a bit louder," or "Like cue 1 but with a slower build," (or more likely "That bit's good but can you make it a bit bluer," or something helpful like that ).
I don't think the tools around at the moment can work to that kind of prompt (yet) as they're generating the whole piece as a single stereo waveform, not a bunch of stems that can be adjusted after the fashion.
But I'm sure that will come in due course.

[sonics]

Some suggestions:

Game Audio ReBalance And Generation Engineer

Sonic Human Interface Techologist

Computer Rendering Audio Producer

?

[Drew Stephenson]

Some suggestions:

Game Audio ReBalance And Generation Engineer

Sonic Human Interface Techologist

Computer Rendering Audio Producer

?

Who wants to start a band called Sonic Human Interface?
(I think synths may be involved...)

[RichardT]

For medium and even high budget productions I think there will be a partnership between composer and AI. The composer might come up with a theme and basic harmonic structure and ask the AI to orchestrate it or arrange it for string quartet, or blues band, or whatever.

So composers may remain, but their assistants might not.

[DC-Choppah]

I could assign chords to numbers on dice and then roll them to get my chord progression. Did I create something new? Is Ai just a sophisticated pair of dice?

That is literally what the theory of generative AI is based on. Its a simplistic model that reduces content to tokens. For music that is for example MIDI messages. It scans zillions of examples of MIDI files. It tries to predict the next token from a new random starting point, or some other cue. This is the same as a spell checker, or automatic text completion. It just predicts most likely next token from the training data. By using subsets of the training data (in the style of Bach) you get a different result. There are different ways to do the predicting of the next symbol, but these just amount to computational efficiencies in the scanning of such a large space.

Humans do not make music this way since they also use functional intelligence and other forms of intelligence instead. Functional intelligence means that you interact with the environment and try to achieve some goal.