SoundCloud's Hazel Savage on the Creative Opportunities of AI

For the first few years of its existence, SoundCloud had a rep as a dynamic, mutable creator ecosystem, but one that could be disorganized and alienating for the uninitiated. That’s changed dramatically over the years, and the service has become a more welcoming place to its millions of listeners, and has harnessed real value for its deep bench of talented artists. A not insignificant degree of credit belongs to Hazel Savage, her co-founder Aron Pettersson and their team. Savage began her career as a record store clerk in the UK, before quickly ingraining herself in the digital music world through stints at Shazam, Universal, Pandora, and Bandlab. In 2018, she co-founded the AI tagging company Musiio and served as CEO. Through its ingenious technology, the company was able to use AI to process and understand the relationships and attributes of millions of tracks. This was useful in deciphering things such as genre or key, and developing throughlines between tracks that powered discovery. It was particularly advantageous for SoundCloud, who acquired the company in 2022. That service focused Musiio’s tech on the millions of tracks by its creators and provided clear pathways to discovery for those tracks.

Third Bridge Creative co-founder Sam Chennault recently caught up with Savage to have an open and clear dialogue about the possibilities and limitations of AI, and how these new technologies work with human workforces.

As someone who’s spent two decades working in the digital music space, I've always heard terms like algorithmic curation, audio fingerprinting, big data, etc. As it’s been presented to us in this past year, AI seems to be something novel, but is there a throughline between all these concepts?

To me, the throughline has always been looking at the problem and then looking at the solution. I feel like those of us who've worked in music and curation for a long time went from a phase where scarcity was our biggest challenge to massive abundance, where all music is available online at your fingertips at any given second. There’s 100,000-plus tracks uploaded every day. The challenge then becomes not what can you access, but what should you access? How do you actually find the stuff that you want to listen to either from a curation or search perspective? 

Back when I started in a record store 18 years ago, we didn't talk about things like algorithmic playlisting, collaborative filtering, or even AI fingerprinting because we didn't need these things. You could walk into the HV on Oxford Street in London where I worked and you say, “I like classical cello music, what do you recommend?” And I'm walking over and picking you out two or three things. We've gone from that very manual process to an automated process, so the technology has risen to suit the occasion.

It's an interesting concept. The technology, and the tools, were built in order to engage with a certain market dynamic where we went from scarcity to suddenly 100,000 tracks being released every day.

I’ve always tried to view it through the lens of positivity. I came up in a time where releasing music was insanely expensive and very limited to people with the financial means, and that's no longer the case. I would never want to go backwards. I would never want to limit creativity or say that we need less music. I'm always for more music and more creation.

Makes a lot of sense. Do you think people like promoters, A&R, music supervisors, DJs, curators, radio programmers, critics – people who were traditionally thought of as tastemakers and gatekeepers – are going to exist, or will AI displace them?

It’s an interesting question. I think that there's something very human about finding someone who likes what you like, and then wanting to know what else they like. And I think that's true whether it's recommendations for clothing and footwear, right through to restaurants and music. We're social beasts. So I don't ever think AI truly replaces that need for the tastemaker and the fascination humans have with the tastemaker. But you can't hire a tastemaker for every single person on your platform. Zane Lowe is a great rock curator, but we're not going to send his playlist to everyone. I'm a Bon Jovi fan, and another person might be a Coldplay fan, and he might not playlist either of those, so it doesn't necessarily scale the way that AI does. I think where AI is powerful is it can put very, very personalized recommendations in front of millions of people for a very, very low cost.

If you just played me one song right now, Sam, and you asked me to list the genres, the key, and the BPM, and say whether there's a vocal, I'll do it, I'll do it a hundred percent flawlessly, and I'll do it better than an AI could, with more level of detail. But what I can't do is five million of those a day, and that's what AI can do. Like I said, I don't want to stop people creating music just because we can't tag it quick enough, that's not where we should put a limit on creativity, but understanding where the limits of AI are is really powerful.

As someone who’s worked at tech companies, I’ve also been told that I should never suggest that a human can do anything better than a machine.

I feel like there's a real benefit and a real power in talking about AI’s limitations, because that's when you start to see that there really isn't anything to be scared of. Talking about the limitations makes people comfortable. To the man on the street, they might just think AI playlists are taking jobs from human curators, and when you start to understand the reasons why you would choose a human curator or why you would choose an AI curator, it starts to make sense and you start to really understand.

Developers may feel very strongly that technology is the only way to do things, and people that have got 20 years' history in the music industry at the major labels may feel very strongly that only humans can do things. I just strongly feel that the reality is somewhere in the middle and that people like me, or maybe people like yourself, we understand the power of technology. But we also understand the benefits of humans, and if you can figure out how to harness the best of both, that's when you figure it out. The minute you just blindly say, "there's only one way of doing things and we cannot change and this is absolutely a hundred percent the best way of doing things," that's when you're wrong, and I think that's true whether you're on the tech side or the human side. So, for me, I’m trying to figure out how you blend the best of both worlds.

What are some of the limitations of AI? Can you expound on that?

I'm not sure if this is what you’re looking for, but I’m fascinated and I love to really dig in on where the ethical limitations are and where they should be. AI is good at doing something that's super tedious, the super boring [work] that humans don't want to do—for example, listening to hundreds of songs a day and writing down the key that they're all in. That's not super exciting or creative work, and humans don't really want to do it. So AI is very helpful in that category. The second category where AI is really helpful is when it's something that humans aren't physically capable of doing, or where there are chances a human might miss or miscalculate. If I've listened to 200 songs in C sharp, I'm going to make more errors.

We have a great use case for AI. Now we can say this application is for good more than for bad. And I'm really fascinated by companies that want to look at these ethical frameworks and want to apply them because I think it reassures a lot of the creative people in our industries that we're not just going ahead as fast as possible and doing anything Jurassic Park-style. We stopped to think if we should. It's really about the ethical and the integral integration of these technologies...I'm willing to step up and be one of the custodians of this technology because I think it's really powerful and can bring a lot of benefits.

‍