Is Ethical Music AI Possible? Voice-Swap Thinks So.
a.k.a. An interview with Declan McGlynn and Michael Pelczynski about the platform they're building and why an AI ecosystem built on unauthorized copyrighted data doesn't have to be the norm.
When I was doing my book tour last year, each event would end with a Q&A session, during which someone would inevitably ask, “What do you think about AI?”
The truth is, I didn’t have much of an answer. I still don’t. Of course I’ve seen all the AI news that’s surfaced in the past couple of years, and have made the requisite “this technology is going to kill us all” jokes to my friends and colleagues, but I’m no expert on the matter. How do I feel about AI? Excited? Terrified? Both? Having had a front-row seat while streaming technology has completely upended the music ecosystem, I can’t say that I’m thrilled about the prospect of another disruptive force tearing through the industry, especially when that force is not only owned by giant corporations, but has been trained on copyrighted materials and content those corporations simply hoovered up from the internet, usually without asking permission or paying anyone.
At the same time, I’m not a luddite, and it’s pretty clear that AI technology has the potential to remake and streamline all sorts of things, including the creative process. Will that put some people—including music journalists—out of work? Probably, but when it comes to music-making, it’s something that will not only make the process far more accessible, but also open up new creative pathways, many of which we haven’t even imagined yet. If we look back at relatively recent history, things like synthesizers, drum machines and digital production software were all disruptive technologies too, and were all decried at one point or another for their supposed artificiality, but few would argue now that they haven’t thoroughly enriched the musical landscape. (They’ve also made it easier for more people to make a whole lot of thoroughly unremarkable music, but technology is often a double-edged sword.)
So yeah, I’m torn, and when asked what I think about AI technology now, I often say, “It doesn’t matter what I think, because it’s coming.” That’s the truth, and it is legitimately scary, particularly as we see the dizzying speed at which the technology seems to be progressing—and the cavalier attitude that its proponents, many of them driven by investment capital and dollar signs, seem to have about who might be hurt or exploited along the way. There are plenty of reasons to slide into full-blown AI doomerism, though I’m not there yet myself, possibly because I’ve found a small bit of solace in the messaging coming from companies like Voice-Swap during the past year.
Voice-Swap was launched by Daniel Stein (a.k.a. DJ Fresh, who many jungle heads may remember from his time as part of the group Bad Company) and Nico Pellerin, and its team includes, among others, artist / researcher / YouTube educator Benn Jordan, former SoundCloud exec Michael Pelczynski—who I interviewed in 2021 about his efforts to bring user-centric streaming royalties to that platform—and veteran music tech journalist Declan McGlynn. All of these people have been longtime public advocates for openness and fairness in the music industry, and that has carried over into Voice-Swap, which is one of the few music AI firms actively marketing the fact that none of its models were trained on copyrighted data. And while many AI tools are leaning into the novelty of “type in a prompt and see what crazy music comes out,” Voice-Swap has directly teamed up with a number of vocalists, some of them legitimate icons. Moreover, the company has made getting them paid—including when users generate an AI emulation of their voice—a core part of its business model.
What’s perhaps most interesting, however, is that Voice-Swap last month announced that it had teamed up with BMAT—an internationally recognized firm that already monitors and reports on music usage around the globe, helping rightholders get properly paid for their work—to develop a new certification for AI music models. This certification will verify that the training data used by any AI model contains no unauthorized recordings, and while Voice-Swap—who, again, helped to develop it—was the first company to be certified, the service is now available to anyone who wishes to take part and make clear that what they’re doing is legally above board.
Will it work? It’s too soon to tell, but it is intriguing that Voice-Swap—which is still basically a start-up—is devoting resources to making it easier for the entire AI industry to potentially do the right thing. Given the way things seem to be going, at least in the news, Voice-Swap appears to be battling an ethical headwind, and is maybe even putting itself at a competitive disadvantage. So why take this course of action? Do they think so-called “ethical AI” is even a semi-realistic goal? Looking to find out, I got on a call last week with Voice-Swap’s McGlynn and Michael Pelczynski, and we had a long conversation about not just the platform they’re building, but what kind of music AI ecosystem they’re hoping to see emerge in the years ahead.