1. Inside the Black Box
Light Box
Composer Mark Henry Phillips explores the profound existential and professional impact of generative AI music tools like Udio and Suno. The epis
Key takeaways
- Professional musicians are widely adopting AI tools in secret to maintain a competitive edge without facing public stigma.
- Generative AI models like Udio and Sun/o function as neural networks that predict audio patterns rather than simple memory banks of existing songs.
Main topics
- Generative AI in music production
- Existential crisis of creativity
Notable quotes
If a machine can replicate your creative instincts, were they ever really yours to begin with?
Conclusion
While AI threatens traditional musical identities and livelihoods, it
Transcript preview
Speaker 4 (0:00) From Windhill Studios, this is Lightbox. The first time I played around with an AI music generator, it caused an existential crisis. I literally lost sleep staring at my ceiling in the middle of the night, wondering, will this be my last year making money as a musician? So what happened? I was scoring a project that was Alfred Hitchcock-esque. Hitchcockian? I was supposed to write a dozen tracks that sounded like they could have been written by Bernard Herrmann, one of the greatest composers of the 20th century and a frequent collaborator of Alfred Hitchcock. Great, I thought. This'll be easy. Before I sat down at the piano for my impossible task, I thought I'd do a little experiment. So I logged on to an AI music generator called Udio. There was a little box that said, describe your song. I typed in Bernard Herrmann, theme, Alfred Hitchcock, film, mysterious, and then clicked create. About a minute later, I got this. My mind was blown. The composition, the orchestration, the recording quality, it was all spot on. A million times better than anything I could do. Speaker 4 (1:35) That's stunning. This is Michael Spitzer, author of The Musical Human, a renowned musicologist and Beethoven expert. I played him this track to see what he thought. So that's AI? That's AI. And it creates the Speaker 3 (1:49) actual sound? Yeah. Speaker 4 (1:51) All I typed in was words, and it produced that in about a minute. Speaker 3 (1:56) I'm still trying to process it. It's mind-blowing, isn't it? Speaker 4 (2:03) And yeah. My name's Mark Henry Phillips. For the past dozen years, I made a living as a composer. Music for commercials, for clients like Google, Nike, and Ford. Scoring indie films, making music for podcasts, podcasts like Serial, Startup, This American Life, and Homecoming. It isn't exactly what I dreamed of when I was a kid and I decided I wanted to be a musician. But on most days, I counted myself lucky. I was making a decent living by playing music. But now, it felt like the jig was up. Speaker 4 (2:47) This whole crisis started about a year and a half ago, when I logged on to Udio for the first time. I was there just to see what it could do, and I stumbled across a track by a user named Man or Monster. He was trying to recreate a Toots and the Mädels track, and I typed in some lyrics along with soulful, reggae, ska, 1969, Hammond organ, and out popped this. Speaker 3 (3:21) I Speaker 4 (3:22) didn't want to admit it, but it's good. Really good. It doesn't feel like it was created inside a data center. It feels like it was made in a makeshift studio in Jamaica on a hot summer night in 1969. It brings up all sorts of copyright issues because this sounds a lot like Toots, a.k.a. Frederick Toots Hibbard. aka a real live human being. And I would wager to bet that's because Toot's entire catalog was used in Udio's training data. But let's ignore that for now. We'll get to it in future episodes. For now, let's just focus on how good this sounds. Speaker 3 (4:09) It Speaker 4 (4:09) might not be perfect, but it feels like real music. And if it feels real, that has big implications. Like what, you ask? Let me give you a hypothetical. You're an ad exec making a beer commercial. You want a track that sounds like it was recorded in Jamaica in 1969, but you don't want to deal with the money and legal back and forth that comes with licensing a vintage track. You might turn to a composer, like me. You pay me, not as much as you'd pay Toots or Desmond Decker, but still enough that it'd be worth my while. Boom, you get a commercial, and I have a job. Years ago, that's exactly what happened to me. I was hired to make a song that sounded like an early rock steady tune from 1969 Jamaica. Essentially, I was given the same prompt as the AI track, but here's what I came up with on my own. Speaker 4 (5:11) It took me a couple days and it's pretty good, fine, mediocre. It's kind of embarrassing playing this. And I was going to take out the vocals because that part is super embarrassing, but it's kind of important to our story. Speaker 4 (5:42) Whatever you think of my track, the vocals on the AI version sound way better in that legally and morally dubious type of way. And don't get me wrong, there are many musicians out there who could have produced something way better than me and the AI. But the vocals? Here's the AI-generated track again. Speaker 3 (6:01) I know I need to stop holding... Speaker 4 (6:14) To get a vocal performance like this, you'd kind of need to be an amazing Jamaican legend with a group of amazing backup singers. And even if you were, you know, Toots himself, you couldn't write, produce, record, and mix a track in 10 seconds. That's the crazy thing. Our hypothetical ad exec can now make this track, hell, 10 tracks in under a minute. And it's basically free. See why I was having an existential crisis? This was my thing. You could give me a prompt, make a song for a beer commercial, whatever it was, and I could make you a song that fits. Not a lot of people can do that. So it gave me a job. And it also gave me an identity. And now a machine could do it in 10 seconds. Speaker 4 (7:10) And yet, when it comes to AI music, all I hear is that it's slop. It's taken as read that AI couldn't possibly create good music. So what was happening? Maybe I just stumbled across a few standout examples? I kept digging. And I kept coming back to soul music because, well, the title of the genre kind of explains it perfectly. Soul music is supposed to have soul. Surely AI couldn't do anything that resembled the holy grail of soul for me, Booker T and the MGs, the legendary house band of Stax Records. But without really trying, I created this. Speaker 4 (8:03) I've listened to Stax regularly for decades, and if someone played me this track and told me it was an outtake on a Booker T box set, I wouldn't have flinched. You might be less convinced than me, but would you really call this slop? Speaker 4 (8:29) So why this huge disconnect between the slop narrative and my experience? Do I just have horrible taste? It's definitely possible, but I've made a career as a commercial musician with very modest musical ability. In some ways, all I have is my taste. So how could my taste be so out of whack with this general narrative? Something was going on. Speaker 4 (8:57) Unsure what to do with all these anxieties and questions, I made a radio piece for the NPR show I used to work for on the media. And the response I got was huge. Over 100 people tracked down my contact info on the internet and wrote me. Speaker 4 (9:14) Most didn't even have a specific question or comment. They just seemed to want to talk about what was happening, about what it meant for them. A lot of them were musicians, but not all of them. Programmers, designers, writers, people in completely different fields who heard the piece and thought, that's my situation too. And the more of these conversations I had, the more I realized something. Music might be the perfect way. to understand what AI is doing to everything. It's older than every profession. It's older than every language. Everyone has a relationship to it. And yet, almost nobody was really looking at what AI was doing to music. Speaker 4 (9:59) So I set out to discover what was going on. Over the past year and a half, I talked to over 40 people. Musicians, CEOs of the AI music companies, industry insiders, lawyers, machine learning scientists. What I found? This transformation is simultaneously much scarier and much more alluring than I imagined. And also, I quickly realized this isn't just a story about music. To get a little grandiose, it's about what happens when machines start doing the thing that you thought made you human. So where to start? Though you might not realize it, the AI music revolution is already deeply underway. I believe that everybody's using it. Speaker 2 (10:51) You have Grammy award-winning producers, artists, songwriters using these tools in their professional workflow. Speaker 3 (10:58) People are using it. It's very used in the industry. Speaker 4 (11:04) This was the overwhelming takeaway from the conversations I had. AI music isn't just coming. It's already here. Speaker 2 (11:12) But... I probably shouldn't even say this, but when I talk to my artist friends who are serious in music, 100 % of those people are adopting AI tools, and 100 % of those people also don't talk about it publicly. They're really all using it, but they don't talk about it. Speaker 6 (11:30) In reality, AI-generated stuff is everywhere now, and a lot of people like it without knowing that it's AI-generated. Speaker 4 (11:39) Those clips were from people we'll hear in future episodes, and there's nothing like a montage to make a point seem more convincing. But ultimately, all of that was just a handful of anecdotes. It's not scientific. Of course, there's no data to base your judgment on. This is Helmets Bemz, the CEO of Sonarworks, a music software company. A couple years ago, Helmets and his team realized a generative AI revolution was coming to music. And they wanted to get a slightly more scientific grasp on what was happening. Speaker 5 (12:12) So we said, OK, let's go out and talk to many people in many fields in the music industry. We ended up doing like over 100 interviews. Speaker 4 (12:22) To be clear, they were interested in generative AI. These more recent AI tools that generate entirely new sounds, whole songs. And what they found was surprising. There is definitely a debate out Speaker 5 (12:37) there in terms of is it good or bad. But of the people we interviewed, there was nobody who said, this is crap and this is not impactful. I mean, everybody have tried it. They have realized that it actually works. It brings value and they have fully leaned into using it. Were Speaker 4 (12:55) you able to get a sense of where in the industry it's being adopted? I mean, I would assume it would be