After major cancellations, Fun Fun Fun Fest adds two big-name headliners

This June, The Atlantic published a searchable database of songs that may have been used to train artificial intelligence (AI) models to make new music. Artists everywhere, including here in Austin, started searching their names and speaking out. One local nonprofit is now working with attorneys on a possible lawsuit, and Austin musicians have a lot to say about seeing their own work turn up in the datasets.

For generative AI to “compose” a song, it needs to be trained on the work of human musicians. However, given the jealously guarded secrets and legal complexities of AI models, there isn’t much transparency about where this data comes from. Atlantic staff writer Alex Reisner started a hot conversation online when he uncovered four datasets “being shared within the AI-development community” in his words, a collection of more than 20 million songs combined.

The datasets contain songs by ubiquitous artists — from Taylor Swift and Bad Bunny to Miles Davis and the Beatles — as well as local ones who may not even be known outside of their respective scenes. The article notes that Google and UK-based AI company Stability have confirmed that they have used one of the sets, which was downloaded from the Free Music Archive, a free streaming service for personal use that charges a fee for commercial use.

Reisner gathered the other sets from research reports and AI data-sharing sites. A song's presence in the database doesn't prove a specific company actually used it for training, but it does show that the music was available to be used — regardless of the creator's intent or platform’s policy when uploading it.

Legal complexities

Nationally, according to the story in The Atlantic, musicians and labels have filed at least a dozen lawsuits against AI companies over training on copyrighted music without permission. In Austin, nonprofit Austin Texas Musicians CEO Pat Buchta posted on the organization's Instagram page to respond to the story and gauge interest in taking legal action. The post lets Austin musicians know their music may have been used without their consent and that "we may have a class action lawsuit on our hands."

Since posting, Buchta tells CultureMap dozens of musicians have reached out. Musicians who find their work in the database can reach him at pat@austintexasmusicians.org.

"Over the last weeks, Austin Texas Musicians has heard from dozens of musicians whose works were used to train generative AI models without consent, credit or compensation. Our organization is gathering this information and coordinating with attorneys, advocates and other national efforts to pursue potential legal action," Buchta says in an email.

CultureMap connected with some of the Austin musicians who appear in the database, and 11 of them answered three questions: Did they know their music was part of an AI training dataset? How do they feel about it? And, do they want to see legal guardrails? The answers submitted via email ranged from furious to philosophical, but everyone agreed nobody asked first.

Nobody asked us

Nearly every musician said they learned about the database through social media, not from any AI company or label. Ray Benson of Asleep at the Wheel noticed peers posting and "had an assumption that we might be in there." Carolyn Wonderland, on the other hand, said she was "caught off guard by the news." Several artists learned the news from CultureMap’s reaching out.

Kevin Russell, bandleader of Shinyribs, described the moment before he searched his own name: "I had a slight hesitation just before I hit enter, what if my music wasn't there?...[To be honest], I would’ve been disappointed had I not found my music there. Once I did, I just shook my head that these tech companies would deliberately use my music without my consent."

For others, surprise quickly turned to betrayal. Rob Glynn of Nuclear Daisies said he feels "overwhelming[ly] negative about AI and the commodification of human creativity," calling it part of "the bleak Blade Runner future dystopia." Thor Harris, frontperson of Thor and Friends, put it in historical terms. Musicians have always borrowed from each other, but "now AI can just poach from us," making the spread of ideas "less interesting because the ideas will not be filtered through a living being with a point of view."

Carolyn Wonderland added, “I thought AI would maybe do the taxes and laundry and we would get to create the art. Why are we willingly robbing ourselves of music and settling for approximations?”

Jammy Violet, frontperson of Pelvis Wrestley, echos these sentiments about emptying out the human element of music.

“The funny thing about it is that they call it ‘scraping’ the songs, like they're taking the guts out of a gourd, but that's not actually true,” says Violet. “What's inside the songs is unscrapable — it's personhood, and experience. Those are not AI's purview. What AI can do is make a [cast] of the surface and reproduce an aesthetic. The aesthetic is not me, but the songs themselves are. What the songs are is safe, but violated all the same."

"Musicians get told a million different ways that they don't deserve to make money, and every new technological advance give corporations and the public new ways to reinforce that message," said Violet. "Message heard! Guess I'll go eat worms!"Photo by Vera “Velma” Hernandez

What guardrails would look like

The artists largely agreed that consent, compensation, and disclosure were important factors in developing guardrails. Benson wants "a path to revenue sharing if consent is given," plus a requirement that "all of the DSPs [digital signal processors] and outlets identify to the consumers that this is an AI generated song/artist/band[,] so that there is full transparency to the general public.” Sarah Dossey agreed on labeling. "It should be law that music is marked as AI if it in fact is," she wrote in.

Thor Harris wants something more clear cut, writing an ultimatum that "Every time AI listens to music I made, I should be paid." Composer Graham Reynolds also started with the basics, proposing a royalty structure that starts with the initial use for training, which “would be easy” to calculate. The next steps would be more complex, and none of the musicians who responded had a detailed answer for uses after the initial training.

It's not just about royalties. Adding AI music to the commercial landscape means increasing competition that can dramatically undercut musicians. Paul Kresowik of Tomar and the FCs is concerned about a specific, often overlooked part of a musician's income.

"Many people don't realize that we make decent money when companies use our music in commercials, movies, or video games,” he said. “I can see a company using an AI version of Tomar and the FCs for a sync placement to save money, rather than paying us for one of our songs."

Some of the respondents pointed out that artists are currently held to a higher, more organized standard than tech companies when it comes to using other artists’ intellectual property. Topaz McGarrigle, both a solo artist and the frontperson for Golden Dawn Arkestra, feels hopeful that like sampling — a process that in its early days was unregulated and often contentious — compensation for training materials will one day be “sorted out.” McGarrigle isn’t letting brass tacks get in the way of his optimism, saying, “if AI is as smart as they say it is[,] it wouldn’t be that hard to track.”

Ann McNair, the managing director of choral group Conspirare, took more time to itemize her critique, listing some of the contributors the organization is responsible to and noting that it is “infuriating” that tech companies do not seek informed consent before moving forward.

“As a presenting organization, we are required pay licensing fees and obtain permission to publicly perform and record the music we sing so that composers and authors are compensated,” said McNair. “We also pay our artists, our crew, our engineers — all of whom have contributed valuable intellectual property to the music our audience loves. The financial burden is on the organization, and the use of our music for training is yet another example of how musicians bear the costs for other entities to profit.”

A large organization like Conspirare is responsible to many parties, raising even more questions about fairness in data use.Photo by Daniel Cavazos

On the even more defiant side, Violet stopped short of the idea of consenting at all. “AI is a tool for exploitation,” Violet said, “so I'm not interested in systems and protocols that legitimize finding the right amount to exploit artists.”

A few hopeful notes

A few respondents saw silver linings or took an ambivalent stance, like Reynolds, who “[doesn't] actually have strong feelings pro or con” his music being used. Russell, while irritated that his data was taken without consent, sees potential in AI as a studio tool. However, he predicted that just like AutoTune, "within a couple years we'll all be sick to death of it."

Dossey went further, arguing music shouldn’t be treated as a commodity in the first place. In her estimation, pushing these boundaries could remind us of what music is really for.

“Music, art, cannot be made as a tool to serve a capitalistic end. It then leaves the definition of art and becomes a commodity; a product. Sure, it would be nice to feel I could support myself making art, but also, relying on that absolutely steals my own joy in the process. If anything, the loss of any value at all for music … forces us to create from a completely altruistic perspective, which cannot be a bad thing,” Dossey said, adding that live shows are "inevitably going to become much, much more valuable."

For now, the legal battle is still taking shape. Austin Texas Musicians says it's continuing to gather names of local artists whose music is in the database, and it will coordinate with attorneys on potential litigation. Whether that effort results in a lawsuit likely won’t be clear for months or years. The type of industry-wide guardrails artists are asking for are even farther off. In the meantime, Austin artists will continue doing what they do: making music and hoping the community shows up.

Non-musicians may feel helpless in the matter, but they exert their own influence, Benson reminded them. “I understand that AI is not going away, and potentially the cat is already out of the bag,” he said. “I would encourage the general public/music fans to push back against AI generated music and art creations.” Finally, Russell sees it as bigger than music. “Artists, writers, labels, producers, and for that matter all Humans should and must control their own data,” he insisted.

--

Brianna Caleri contributed to this report.