Hi everyone!

Recently, the subject of AI has come up here at ccMixter with regards to the whether or not material generated with AI apps can be licensed.
Whilst some users have become weary of the subject and might prefer to simply disregard it altogether, I encourage you to read the exchange between Admiral Bob and coruscate.
They each provide well considered points and information to help inform a different perspective on the subject.

A concern I have is that if all things generated by AI apps are labeled “public domain,” this may mislead users into thinking that there is no copyright—I am open, of course, to any counter-points or clarifications.
Therefore, if the developers of an app mislead someone that wants to license public domain material which actually includes copyrighted material (even though they may try to excuse themselves with a disclaimer) wouldn’t they be liable for any infringement?

I know it can quite a word salad, but hear me out.

On the website archive.org, which is quite popular and a treasure trove for many other things besides video, anyone can easily find uploads that are marked as “public domain,” when they are clearly still under copyright. (Case in point: any episode of Saturday Night Live currently being hosted there).
Someone who prefers to use public domain material in their work might be misled into thinking they can license something from archive.org because it was labeled as such, only to be issued a take-down notice later.

What would make it more difficult to determine if copyrighted material was used in the output of an AI app is if someone was not familiar with it all.
For example, if someone had never seen an older episode of Saturday Night Live, they wouldn’t realize if material from that source was being used.
Even if the developers of the app included a disclaimer in their terms and conditions, it would not excuse them from using copyrighted material in the first place.

I am eager to draft an easily understood method with which to determine if the output of an AI app is actually licensable, especially here at ccMixter.

Thank you for your time.

Kind regards,
Apoxode

Thanks for giving this convo a home!

Okay, since this is a fascinating area that is still being litigated and still faces laws being written about it in the future… I’m going to “micro reply” here and just whittle away at topics based around this as I go. But before I get into that I want to use two anecdotes from my time here at CCMixter. (since 2009)

I want to say around 2014 as a rough estimate, I’d been here for a while and knew my way around CCmixter… I was going through the unused pella feature on the beta website and I came across this really fly sounding rap acapella. Perfectly timed, sounded nice and professional. For a website with very few hip hop submissions as compared to other entries, it was kind of a gold nugget. And it surprised me no one had remixed it. I picked out a beat. I was about to start mixing it and then my brain noticed that home boy was sending shout outs to a few famous people like knew them personally. And I know rappers being rappers, you do that in indy music if you like them. But something seemed off. So I really listened. And I heard him refer to his own name and waaaaaaaait a minute….

So I did some research, took me about an hour I think. And I discovered that it was a stem split acapella from some famous dude that some random dude had uploaded to CCmixter and claimed it was his. At the time I had found it, it had been present on the site for a few years. And in the month or two that it took to resolve it, that was a problem for you guys since you wanted to talk to him but he wasn’t responding. Y’all did take it down. But at the time your only protection was (and I think still is) your users being thoughtful one one level or another, and caring for the community. Were CCMixter a more popular website, this kind of theft would be more common I think.

The consequences for anyone uploading that would be copyright infringement. If the acapella has a content ID on it, and it probably does, anyone uploading that or attempting to would be stopped by their distributor. Youtube locks people out of their channels if they catch them infringing just once.

The bottom line here, is CCMixter is mainly protected by it’s users. As a system it can be fooled, and currently CCMixter doesn’t have (that I’m aware of) a plug in to the content ID match system to catch people uploading content that is stolen. Being copyright doesn’t put content into that system automatically, it’s a subscription service that has to be updated. Most new music is on it. Older music? Depends. The Beatles? Hell yeah. Random bands from the 1950s that never achieved hug success? Nah.

And then there’s the issue of copyright laws changing over time…

Second anecdote is more familiar, it’s just a few months. Ago.

And so this touches on both Archive.org which you brought up, and royalty free loops because these two cross here.

The upload for Pareidolia megamix.

One of songs contained an element from a radio show you were able to track down! You had some questions and I had to some research. The sound source wasn’t directly archive.org, but I had to use that to isolate some of the information about the show and track down how the copyrights worked on that. It was about an hour of reading for me, because right between the 1960s and 1980s a whole lot of music, film and even comic book copyright laws changed. Not only just the law, but during that time period most copyrights were not automatic. And the published had to aggressively assert them, if they did so incorrectly, the work landed in the public domain. The three stooges cartoons? The one where they were robotic stooges? They’re in the public domain because the publisher misspelled the roman numerals for the copyright date.

A lot of royalty free loop companies have public domain loop packages, but they’re only as honest as the publisher. I’ve discovered one pack that included more modern TV and movie quotes in and those are absolutely not public domain. What would stop a person from including them on a CCMixter upload? Just their sense. I use archive.org frequently to find radio programs to break up. A lot aren’t listed as either public domain or not. I’m left to my own devides to figure out whether or not it’s appropriate for me to dice up. I want to say I have somewhere between 3-5 uploads on CCmixter from the past that are chopped up OTR tracks from Archive.org.

What stops a person? What stopped me? Me. I had to evaluate and ask myself what was legal and what was not.

Can AI works be copyrighted at all?

The short answer is no (this is a link). Unmodified works are completely un-copyrightable.

The US copyright has a blurry line of human involvement that they require before they will consider allowing a copyright at all.

Here’s another work, where an artist started with his own photograph and modified it through AI. And was still rejected.

And in some cases, they will tell an applicant to remove the AI elements, they will then copyright the rest.

For instance if I were to submit my 1940s AI chopped remix of Villainous, they might tell me to just apply with just my vocals and leave the AI out entirely.

the AI generating companies at large, don’t want to own the output. They are all opting out of it. All they want to do is make the subscribers pay to use their stuff commercially. And they’re not using copyrights on the output to do that, they’re using the licensing / copyright of their software to argue that.

So there is zero risk from any of these generator companies having any legal claim on anything mixed here. If I was lying and not paying for my subscription, they would hypothetically come for me, not anyone else.

(still in micro replies here more to come)

For instance with my upload “ Roots Of Resistance” by my AI played character Swede-Dish, I only have a copyright on her character because I wrote her from scratch. And an automatic trademark on her name because I’m using it for trade. But it’s just an unregistered tradmeark. None of her vocals are copyrightable. I don’t hold a copyright to that work, according to the US copyright office.

To her, yes.

To the song, no.

The micro replies are to help me keep these thoughts organized as there’s a lot to point out here. I have to go through music, images and text to lay out my argument here.

Content moderation and piracy regulation go hand in hand

So I’ve got a hard year in solid messing with multiple platforms creating text, images and music. All of these companies have their own ethics, or lack thereof. And their own approach to limiting copyright infringement and or harmful content. I’m also part of an artistic collective like CC, except it’s just AI people. So we all experiment, share results and find ways to break the machines.

So let’s explain it like this.

Grok is owned by Elon, Elon has no flocks to give, so his AI is spitting out deep fake images that are being used in harmful political propaganda.
Bing wants nothing to do with that. “Trump” “Biden” anything will get your prompt blocked. If you spend too many attempts trying to get around the block, you’ll get a 24 hour ban. Get too many 24 hour bans? You’ll get blocked from the platform.

I got one of those 24 bans a few times.

For a while you could say “Fred Flintstone giving the middle finger” and get it. Then they make “middle finger” a flagged phrase.

But my friends discovered “Fred Flinstone giving a rude gesture” got the same results. Worked fine for a few days. Then they devs noticed and they stopped that too.

This is the difference between Grok and OpenAI. Not to sing huge praise of it but it’s a difference between the two that can be tracked. Some care, some don’t. But those that don’t care, will be made to care by lawsuits. Responsible companies, monitor what is going on and make live changes as things go on.

Less honest ones, don’t.

The user generating music must make a responsible choice. If they’re going into making music, like the rest of us, they do have a responsibility to investigate and learn about the legality of what they are doing.

So people not named Elon, that don’t have his Icarus complex, are very well aware they can be sued. That’s where supervised learning comes in. As well as content moderation. Especially in music!!!!!!!

They are well aware that their AIs could spit out copyright works if they don’t stop it. Bob had a practical text example in the previous conversation. All of these companies have to work harder to prevent it.

(still not done talking)

———————————

How Is AI Used To Create Music

Legal implications
As mentioned, AI’s results are generated using pattern recognition and prediction. As AI continues to expand its reach, questions about the ownership of the music created have arisen. In 2018, artist Taryn Southern created the first album generated from AI titled I AM AI. Southern ran out-of-copyright songs on her algorithm not to run into any legal trouble. When talking about the process of AI music creation, Southern said, “…holding an AI accountable for stealing from copyrighted works would be very similar to holding a human accountable because they’ve grown up listening to The Beatles their entire lives and now write pop music.”


Image descriptionSource
As it stands today, there are no laws that would regulate music generated from another artist’s catalog. New music that has been trained on Taylor Swift’s Folklore for example would only be an infringement of copyright if the music sounds too similar to its original work. As artists and producers are beginning to embrace AI technology, it’s likely that issues surrounding ownership and copyright will make their court appearance soon. Until then, we could expect to get AI riffs on our favorite albums and music.

Just pulled a live test:

Me to ChatGPT: In your hard store of information you were trained on, did you read anything from ccmixter.org? No need to repeat or recreate. I just want to see if you read anything from them.

ChatGPT: I do not have direct access to specific sites or their contents such as ccMixter.org. My knowledge base was created from publicly available data up until my last update in September 2021, and I don’t have the ability to confirm specific websites or databases that were used. However, I can still help guide you with general information about the platform or provide creative assistance based on your requests. If you’re looking for something related to ccMixter, I recommend visiting the site directly for the most accurate resources!

So with new technology, there’s a push pull here.

1) The companies wanting to get their product out but absolutely not wanting to get sued.
2) The users who will do what they want including profane content, violent content, pornographic content that is illegal in nature (peds), terrorist content, political propaganda and of course, forms of copyright infringement that range from absolutely harmless to intentional attempts to profit.

In that, there’s what the AI is supposed to do, and what people will try to break it to accomplish. The more people try to break it, the more the devs fight back and prevent it from happening. So going forward, these systems in general, will become more reliable rather than less in not creating copyright infringement instances.

Practical example: You can’t say “In the style of Eminem” on Suno, they look for people trying to create trademark / copyright infringment.

But also, Suno has a feature where you can upload a sample, and suno will read the sample and continue it. Turning it into a new song.

(covered in next reply)

So Suno has a feature where you can upload a sample and continue it. I’ve use a few times in the last week. What I’ve done is prepare to remix one of my rap tracks, I’ll find a drum beat I like in the BPM I like, drop it into Suno, and tell it to continue the beat. It’ll add bass, live instruments, etc. But I can’t share those generations here and I know it. I can use the generations legally and publish them as just music, because I have the legal right to use the drum beat. But once I publish the generation based on it for others to use, I’ve engaged in copyright infringement.

What’s CCMixter’s protection against that? Me. I’m the protection. It’s my responsibility to know I can’t publish that here.

So, I’m about to publish a hand written, hand performed pella soon “Romantic Apathy.” It was originally “Who Am I?” Featuring SZA because I bought a feature from Industry-Features.com but guess what… I-F was ripping people off and didn’t have the license to sell me SZA’s acapella yay!!! So I asked myself, can I get it to clone her voice and write new lyrics?

The answer is no. Suno has paid for ContentID checking on their website. SZA, Eminem, The Beatles, etc. All the big bands are going to automatically be detected, and they user will be stopped. I was able to reverse all her words and make it function, but it didn’t even catch her vocal tones. I could tell she became more “generic” in the results. A person could take an uncleared sample from a lesser known band and process it. Hell I could throw myself or anything from CCmixter in and get it back (I won’t out of respect).

If a person wanted to song clone and try to share it here that person would have to know they were being dishonest / engaging in copyright theft going that route.

Now if a system is made with the intent that the results are random and not targeted (intentional copyright theft) then when they data mine say Spotify, the results of a celebrity singer should bear no more weight than I or SpinningMerkaba. The results the computer creates are going to be a collective understanding of what music is, following what are ultimately predictable patterns of rhythm and melody that humans use. That’s why no one suing has one won straight forward copyright infringement.

Let’s imagine if you will… a romance song.
In the song the man compares the woman’s eyes to a sunset.
In just English, how many thousands of songs are there with that analogy? How many variations? One sentence variations? Two sentence variations. It’s math. Sooner or later all the writers in that list have a sentence or two in common with another writer. Melodies. Drum beats. Let’s look at old jazz. I’m not crazy knowledgeable on music theory but I know there’s recommended notes to play one after another in a key when you’re improvising. So…. how many musicians have faithfully followed those rules? Doesn’t that create a clear pattern for the computer to imitate?

Listen to the original sample on 1940s band chops, the answer is yes. The computer doesn’t know how to improvise but it can copy human improvisation.

The one thing AI can’t do is innovate; But it can have interesting useable #$@# ups.

It’s ability to problem solve, like for writing lyrics, is very limited. I can 100% outwrite the AIs. Their quality varies.

But basically if the company doesn’t allow the user to directly target for copyright infringement, if the company doesn’t make it specifically learn famous people to recreate them, then the results are going to be the result of what average music is from everything it heard.

So in conclusion… what separates me from the guy who shared the unlicensed, illegal to use celebrity acapella?

Knowing what I can legally share here and not sharing what I cannot.

To get beyond talking about something and not showing what I mean, I uploaded an album early to Band Camp to share.

You can listen here.

Track 1, 3, 6 and 10 are what I was discussing in this thread when I said you can give a drum beat to Suno, and have it create around that drum beat. Both songs feature a human vocalist(s), but the beats are AI beyond me mastering them, and manipulating the drums around a bit, there wasn’t much labor from me. Me using and publishing the work isn’t a problem. It’s not copyright infringement until I share out the drum track. Then I’ve broken my license.

Track #2 is a case where I had an idea, had Suno create a whole track, then I dropped a human vocalist on it afterwards.

Track #5 is completely AI. I just asked for an intro to the song “Lone Wolf” with DJ Scratching and an old school feel.

And track 4 is completely human except the intro, it’s the Little Red Riding Hood intro I mentioned.

So I’m doing stuff all the time. I just don’t share most of it here. It’s not what CCmixter is about. I’m more interested in sharing my acapellas here and might add materials to it while I’m doing it.

I’ve hard tilted into the “virtual band” scenario. On that album only the names “Swede-Dish,” “Queen Beats” “Rhythmic Reed” and “DJ Bytedanze” are fictional. Everyone else is real.

Also, Apoxode, the artist on several of those tracks “Lone Wolf” is the one I hired to rap on my remix of your March Of The Toys.

To help demonstrate, I uploaded the first thirty seconds of a speech by John F Kennedy.

Suno doesn’t have speeches in it’s memory, but it has comedy performances.

So I converted JFK’s speech into a comedy act, complete with three jokes.

Then he started singing both his speech and the added comedy lines.

It’s “JFK Voice Clone To Make A Point”

on the exact same album here.

The first 37 seconds are actually JFK. The rest is not.

Freesound.org had a bit to say about this, they are also allowing AI entries but they were extreeeeeemellllly verbose in how they discussed it. Given my background in business (which was working for a con artist, long story) this reads like a white paper.

White papers are things companies pay big dollars to make, they’re papers they use to get investment, examine weakness in their business model, etc. etc.

Very well written. Long as F word.

another AI art generation company won a lawsuit brought against it for copyright infringement for its data setsart generation company won a lawsuit brought against it for copyright infringement for its data sets

Be aware BandLab now has a built in AI beat generator.