Non-stop AI-generated Music: Is this Heaven or Hell?
Welcome to the maiden voyage of The Music Futurist Newsletter!
A weekly publication all about AI in the Music Industry.
You can look forward to finding these exciting features, all packed into one email, in your inbox each week:
A short opinion piece (from me and guests) on the latest developments and trends regarding AI in the music industry
Tools of the week, a roundup of interesting, innovative, and useful AI tools for the music industry
A curated list of links to the best social media conversations, videos, and articles from the past week, all focused on AI in music
Have fun! Be good.
Non-stop AI-generated Music: Is this Heaven or Hell?
Time to read: 9 minutes, 15 seconds. You've got time.
A YouTube channel for the “Dadabots” has been playing AI-generated funk music on a continuous livestream 24 hours a day since April 2021. The livestream, titled “No Soul,” is one of many generative AI music projects by two former Berklee College students who describe themselves as “a cross between a band, a hackathon team, and an ephemeral research lab.”
They’ve got another livestream with AI-generated death metal music titled “Relentless Doppelganger” that’s been playing non-stop since September 2019 and one called “Infinite bass solo”, which is precisely what it sounds like.
I’ll clarify that they are not playing a loop of the same music over and over. The AI is constantly generating new music and playing it on the livestream.
You’ll find the following generative AI music projects on their website:
”Deep Beatles” an AI-generated album trained solely on The Beatles' “One” album
4 different AI versions of “Clint Eastwood” by Gorillaz being covered by Nirvana
“Toxic” by Britney Spears being covered by Frank Sinatra (this one is scary good and was generated two years ago)
(Links to Dadabots projects including their research papers and technical details are in the Link Curation portion of this newsletter)
Dadabots are CJ Carr and Zack Zukowski, two musicians and engineers who originally formed as a team for an MIT hackathon event in 2012, Music Hack Day, where the duo created an “army of remix bots, spidering SoundCloud for music to remix, posting hundreds of songs an hour.” Soundcloud supposedly banned them numerous times.
Probably a little tongue-in-cheek, the Dadabots “about” section on Youtube includes this line:
“Read our scientific research on eliminating humans from music.”
Dadabots aren’t profiting off of this work. The AI albums they’ve generated can be listened to on their Bandcamp page and include this disclaimer:
“This site may contain copyrighted material the use of which has not always been specifically authorized by the copyright owner. We are making such material available non-commercially in an effort to educate and advance research in machine learning, generative music, music information retrieval, computational creativity, etc. We believe this constitutes a ‘fair use’ of any such copyrighted material as provided for in section 107 of the US Copyright Law”
That makes sense to me and I’m all for this kind of experimentation!
I’m sure Dadabots have been pacing in agony awaiting my approval.
But some open-source projects don’t take the same approach, which leaves artists open to exploitation.
Some organizations appear to be taking a “shoot first and ask questions later” approach to generative AI. AI text-to-image generators like Midjourney, Dall-E, and Stable Diffusion have quickly become enormously popular and highly controversial at the same time.
Most of these text-to-image AI tools are trained on millions of images (if not billions) and can generate images in the style of almost any popular artist with stunning accuracy. These systems are trained on datasets that scrape the web and pair images with text descriptions. The images include the full gamut of copyright licenses. Music artists are worried the same thing is going to happen to music.
It already did and it has been going on for a while.
OpenAI’s Jukebox was released in April 2020. You remember 2020, right? I feel like there was something else dominating the news that year… anyway…
Jukebox is “a neural net that generates music, including rudimentary singing, as raw audio in a variety of genres and artist styles.”
The dataset the model is trained on includes 1.2 million songs they “scraped” from the internet.
Remember CJ and Zack from Dadabots? They joined forces with the team at Stability AI, creators of Stable Diffusion, to work on Harmonai which produces open-source generative audio tools. Harmonai released Dance Diffusion in October 2022, which is trained on datasets comprised solely of copyright-free music and audio samples.
That sounds like Stable Diffusion is taking music artists’ ownership rights into account, but visual artists rightfully ask, “why the double standard?” and they are happy to answer their own question (please see screenshot below):
In December 2022, Stability AI announced that they’ll honor requests from visual artists who want to remove their images from the Stable Diffusion dataset for V3.
Artists must go to https://haveibeentrained.com/, create an account, search for and manually request each image to be removed. It looks a little tedious. But some artists are happy.
Other artists have said they shouldn’t have to opt out, suggesting an opt-in model would be the fairest. Of course, it’s easy to see how an opt-in model would make it nearly impossible to create datasets large enough for these systems to work as well as they do now.
Frankly, things appear to be all over the place. But that’s to be expected, right? Progress in the field of AI is creating a new frontier in many ways.
Setting copyright law aside, what is the right thing for these AI companies to do?
Are new AI tools leaving artists open to exploitation?
Are the datasets in most AI systems so large that their output should be treated the same as something created by a human who has been inspired by a lifetime of experiences?
I call first-right-dibs on posting this Jurassic Park Meme on an AI post in 2023. There will inevitably be hundreds+ more posts that use this meme to talk about AI throughout 2023 and they are all now derivative of me. happy new year!
And what about accessibility?
You gotta be pretty evil to argue that accessibility in art and music is a negative thing. This New York Times article talks about some of the great work going on in that field: https://www.nytimes.com/2022/11/03/arts/music/technology-disability-music.html
It's obviously a good thing to help people with disabilities create music and express themselves fully.
But AI can also unintentionally discriminate. Infamous cases with AI in criminal justice, corporate hiring, healthcare, and other industries have shown the potential for AI systems to be biased and prejudiced. To make matters worse, the methods used for avoiding those problems, such as removing data that indicates a person’s race, aren’t always effective and can sometimes result in the AI acting even more prejudiced than before AND making it harder to identify.
Ahhh!! Is there no right move?
But wait, there’s more!
I think another concern from music artists is that AI music generation tools will make it too easy for dispassionate creators to flood an already saturated market with new music.
Let’s get mushy.
Music belongs to all people; those who want to dedicate their lives to it, those who use it to relax, those who see it as a way to make money, and those who just think it's fun to tinker with.
But is it art?
Art is creative human expression.
If you’re saying to yourself, “who cares?” That’s the artists’ point. They care and some artists look at the AI community like a weird group of strangers that breaks into their house to redecorate.
The argument is that it takes more than human involvement in the creation of something for it to be considered art. A sidewalk can be art, but most aren’t.
Many would argue, myself included, that we are already deep into this very scenario. Digital music production tools have allowed the uninspired to “whip together a song” in seconds without the help of AI for years.
And so the concern is that we are making matters worse for artists by allowing even more “non-artists creators” to be “music artists.”
I know, I know. Keep up or get left behind.
At this point in the conversation, some eyes glaze over and comments about talent, dedication, skill, and passion bubble up.
Then someone rolls their eyes, an artist gets called egotistical, and nothing productive happens.
But artists can ask important questions from a unique perspective on these topics that others cannot. They should have more than one seat at the table and a voice while these tools are being created. I think that’s the way to proceed forward with integrity.
The guys in Dadabots are musicians, but they're builders too. It sounds like they already had a foot firmly in the computer science camp. I think they're an important segment of people in AI. But I'd love to see AI companies engaging more with traditional musicians who aren't already in this world and showing that they are considering the wide ranging impact it will have on an industry where there are already a lot of bad faith actors.
This article may make it sound like I think the ethics in AI are being ignored. They’re not. It's a vast field and there are numerous organizations dedicated to them. But with this much excitement, it's easy for important things to slip by.
In many cases, it may come down to the responsible use of powerful tools. Oh god!
“With great power comes great responsibility.” (I couldn’t help myself).
That’s the scenario we have in many domains: the internet, cars, guns, nuclear technology, agriculture… a thousand others.
There are rules, laws, licenses, and regulations, but it's people who use and misuse powerful tools.
That doesn’t mean the creators of these tools bare no responsibility. They most certainly do.
But will they be good stewards as we continue to make progress?
So, how can we mitigate the misuse of AI tools in art and music?
Not Sure in the film Idiocracy
But I’m going to start by trying to ask the right questions so I don’t let my excitement get in the way of thinking clearly.
I am a music artist myself and I believe that the human element in music creation is special. AI-generated music may dominate in the future or it may just become another instrument we can play — though it will be an instrument that is even more impactful to music than the electric guitar!
Art and music are inherent parts of human expression. But AI tools will give people the option to approach music creation and human expression in ways we’ve never imagined. That’s good, right?
As for music marketing, music performance, and music discovery, AI is going to bulldoze those hills too. But in their place will rise amazing new creations that I am excited to experience.
Let’s all just agree to never create another 24/7 Infinite Bass Solo.
🔸 Tool of the week
Boomy is an AI music generator who has been around for a bit but I see that they regularly make some big claims. So, I'm putting them on display here just so someone can either verify their claims or point me back to reality because this just can't be correct...
"Boomy users have created 11,000,559 songs, around 10.89% of the world's recorded music." (Link)
Boomy uses music automation technology powered by artificial intelligence, which you can use to create and save original songs in seconds for free. (Link)
🔸 Random Curated Links from around the web
📍Two different perspectives on the impact AI Music will have on the industry
📍Art Created By Artificial Intelligence Can’t Be Copyrighted, US Agency Rules (link)
· We're a cross between a band, a hackathon team, and an ephemeral research lab. (Link)
· Dadabots research (Link)
· Dadabots on Youtube (Link)
· We are a community-driven organization releasing open-source generative audio tools to make music production more accessible and fun for everyone (Link)
If you enjoyed this newsletter, please share it with at least one person. Just one. Of course, you're allowed to share it with more, but I am only asking for you to share it with one.
Copy and paste the link below and share it willy-nilly:
I'll leave you with a Midjourney image of the devil forcing people in hell to listen to his bass solos for all of eternity.