Decoder with Nilay Patel
01:01:04 · Mitch Glazier
[01:01:04] S4: Right. I think they're walking down the road slowly, probably a little too slowly. I'm trying to see if they can do like just enough to keep those two things in balance. Like right now, for example, Spotify is saying, well, we'll only we will only disclose or label a recording as AI in the credits. How many people look at the credits? Yeah. I wish more people looked at the credits. They discover producers and songwriters and other awesome folks who have contributed to the recording, but they don't. They need to do it at the track level. They're not there yet, but they are disclosing AI personas that are made because that's sort of like the highest on the dupe on on the dupe scale. Right? And then on under that, they're looking at the fact that already 90% of the recordings available on Spotify get like 10% of the revenue and 10% of the content gets 90% of the revenue because most of what's on there gets less than 10 listens or less than a 100 listens because all of this slop is just going on. Which means storage costs and all sorts of other things that they have to think about. So I think that they will go as far down the road as fans will demand, as their users will demand. And I think when we get to a point if people get recommendations in their algorithm, Like, if you like this song, you'll also like this one. And when you listen to it, it turns out it was AI and you didn't know it was AI and you feel deceived, that's when they've got a giant problem. Because they're responsible for that recommendation. They're the one that fed that to you. And so we do have lots of consumer research out there that shows that subscribers just like all consumers wanna feel like nobody's pulling the wool over
01:02:49 · Nilay Patel, Mitch Glazier
[01:02:49] S3: their mouth. But is there an actual market pressure? I can think of all kinds of times fans have revolted against Spotify, right? Like I love the band Stars and every now and again they'll tweet like we didn't make any money on this song on Spotify and everyone tweets at Spotify and they're mad. Spotify is not paying artists any more money. Right? Like is there an actual market pressure that will make this labeling happen or is it or is it policy pressure? [01:03:12] S4: It's both, it's combination, but because there are already laws in other countries that are mandating labels and once a market like The EU or The US that's such a big market or The UK or China, Once they demand labeling on a global system, it's actually cheaper to do it all around the world than it is to try to just sort of geo block it. [01:03:33] S3: This is the California car emissions model of regulation. Right?
01:03:33 · Nilay Patel, Mitch Glazier, S10
[01:03:33] S3: This is the California car emissions model of regulation. Right? [01:03:36] S4: The... So the cal... It's amazing what what California can do because these companies are headquartered there. Yeah. I mean, not Spotify, but Apple and Google and others. And so like the transparency law in California is the reason why Suno has to now as of August put a watermark in there, you know, saying this is AI generated that's machine readable. One state but they're doing it all over the world. Right? Because that's where that grows. So it's a combination of policy which obviously we are very directly engaged in and competition. I predict that what Apple decides to do that is more fan friendly will drive Spotify and vice versa. There's not that much churn between them, but if there's some sort of perception that you're being more subscriber friendly than somebody else, that tends to drive results. Just like it does on the privacy issue for example. Yeah. You know, where there's no law on it, but now companies are competing to say, we're the private browser. Apple, we protect your privacy. One [01:04:43] S3: one company says that. Let me ask [01:04:44] S10: you this.
01:04:44 · S10, Nilay Patel, Mitch Glazier
[01:04:44] S10: you this. [01:04:45] S3: I I only got a few minutes left and I I wanna stick one turn at the end to talk about how you measure for platinum certification because that's also up in the air. Sure. But let me just ask you one more question here on labeling. We spent a lot of time at The Verge covering labels for images and media, right? There's all these standards that exist. I have one poor reporter, her name is Jess Weatherbed and she just, she has been forced to learn everything about the C2PA standard and content, synth ID It's and [01:05:13] S4: all these important though. [01:05:15] S3: They're very important and we spend a lot of time covering them because they have to work and they have to work on a number of different places in the chain, right? They got to work in the camera, they got to work in Photoshop and then they got to work on Twitter and everyone has to agree. And if Elon Musk does not want SynthID to show up on his platform, it doesn't matter if it worked in the camera. This is a big chain. It seems harder in music, like much harder because I don't know a bunch of bedroom DJs are probably not going to participate in your technical scheme for labeling when they upload to DistroKid. An artist getting a beat from a producer over iMessage, who knows if that metadata will survive. You might have to watermark the sound files themselves in some deep intrinsic way and that can get you into all kinds of trouble because now you're screwing with the audio files at the software level and artists don't like that. How do you make labeling work at all?
01:06:10 · Mitch Glazier, Nilay Patel
[01:06:10] S4: Yeah. Like, you're you're like hitting on our on our daily meetings. So I mean, there are sort of the international organizations like DDEX for example that can help align fields and stuff like that. C2PA which can give you sort of like a model for providence. But at the end of the day, it has to be the platforms and the distributors that are willing to do it. Because when it comes to generative AI, right, the record labels are not tagging and putting generative AI up on Spotify. Right? If that's out of the DIY world that somebody going to a music generator, downloading it, putting it up through DistroKid. Right? So your gateway is DistroKid TuneCore. It's the distributors. You know, are they using AI detection technology to determine that it's fully generated? You might not be able to use AI detection technology to say your song is 5% of this or use it for attribution yet, but you can definitely use it to say this is fully generated AI. And if so, are they willing to tag it so that it works with Apple and Spotify's platform? And then on the platforms themselves, this is sort of like the parental advisory program. It might be voluntary and everybody might have a little bit of different way of implementing it. But fans are pretty clear. Like when something's labeled as parental advisory, which probably just draws more young people to listen [01:07:32] S3: to it. Did to me.
01:07:32 · Nilay Patel, Mitch Glazier
[01:07:32] S3: to it. Did to me. [01:07:33] S4: When it's label it's labeled as parental advisory, like people kind of get it. They know it's become sort of a culturally known thing. I think labeling for AI for fully generated AI can definitely get there, but the distributors have to be willing to play. It can't just be the platforms because if it doesn't go downstream, they're gonna go in different directions. And already Deezer is going in one direction, Apple is going in another direction, Spotify is going in another direction. And to me, the fully generated problem, which is a DIY problem is a distributor problem. And we have to provide incentives to the distributors to get them to take responsibility and play ball. [01:08:13] S3: Do you think that technology exists to support the policy there? Right, you need a technology that can detect AI media. I'm not sure that any of that exists or works well.