10:37
If you want to understand what's going on in AI right now, not in the future, right now, then there is no better place to look than AI music. First, because the technical achievement in here is simply incredible. Deezer, who are kind of like French Spotify, a smaller streaming site, surveyed 9,000 people and did a blind test with three tracks. And only 3% could reliably tell the difference between AI and human music. 3%. This stuff is uncanny. But you know what? The technical achievement isn't actually to me the most interesting thing here. To me, what's really interesting is the economics of it all.
Because AI music is enabling the mass production of music on a scale we have simply never seen before. That's what this chart shows. This is a chart of monthly AI uploads to Deezer. Again, they provide most of the data here. Back when they first started tracking this in January 2025, there were around three 300,000 monthly AI uploads. Now, 2.2 million. 2.2 million fully AI tracks are uploaded to Deezer every month. And you can see the trend is only going up and up.
This chart shows the same thing, but as a percentage of uploads, 44% of all uploads to Deezer are now fully AI. And no wonder they call it a tsunami. But there's something really odd about this chart. Because if you see this, you would assume, well, supply is going up, demand, listenership, must be going up, too. It kind of only stands to reason, right? But no. This is a chart of share of listens. And you can see AI music is completely flat. This isn't the most granular data I've ever presented for you, but the message is very clear.
AI music is not going up. In fact, it's stuck steadily between 1 and 3% of all listens. Why? Well, the reason is this. Because Deezer believe that AI music is mostly fraudulent. Specifically, they think that 85% of all AI music is associated with fraud. And so they are taking steps to basically suppress it. They're taking it out of promotions and out of playlists, making it hard to discover. And that, along with possibly the fact that people aren't really connecting with AI music, is why, although demand is exploding, listenership is staying flat. But I want to pause here because this is quite an odd concept, odd concept, isn't it?
What does it really mean to be fraudulent in this area? You know, it's not the tracks that are fraudulent. It's actually the behavior around them. Because Deezer believe that people are making artificial tracks and streaming them again and again to steal royalties from their system. And I can show you that how they go about detecting this, because I went to visit Deezer at the end of last year, and they gave me a tour of their fraud detection system. And this is a screenshot from it. This is one particular artist's streams every hour of the day. And instantly, it stands out, doesn't it? This artist is apparently getting the same number of streams every hour, regardless whether it's day or night.
I mean, this is not a human pattern. This is a machine pattern. And this is exactly the kind of signal they're using to detect fraud. This is another example. Three apparently different artists, but look at the pattern. The pattern of streams is the same. This is the kind of their signal they are taking and using it to detect fraud. And if you want an analogy, what I think about when I see this is I think about a bank. Because you know when you go abroad or you take money out
or do something unexpected, and your bank sends you a text saying, "Hey, was that you?" Well, that's exactly the kind of fraud detection process that Deezer is having to do. They're in the music business, but they're also in the fraud detection business. That is partly what AI music has done to all this. And if you want to know how far this kind of thing can reach, take a look at this. I want to introduce you to this guy. This is Michael Smith, who recently this earlier this year pled guilty to music stream fraud aided by artificial intelligence in the United States.
He'd been doing the exact thing that Deezer described. He'd been making tracks artificially and then using bots to listen to them over and over again. And according to the prosecutors in this case, Michael Smith made $8 million doing this. I mean, that's a huge sum of money, especially if you know about music streaming. Because on streaming sites, you don't get a penny per stream. You get something like a fraction of a penny. So, think how many streams Michael Smith must have completed to make $8 out of the streaming sites. I mean, we're not talking millions of streams here. We're talking billions of streams.
Very difficult for the streaming sites. And also very difficult for musicians, because every time someone takes money out of the collective royalty pot, their income goes down, too. It's a big problem. But you know, the funny thing is that when it comes to the overall question, will AI replace musicians? I'm actually an optimist. That's partly because I'm a big fan of chess. And chess is something where computers have been better than humans for decades. But personally, I don't care.
I don't want to watch computers playing chess. I want to watch humans. I don't really enjoy playing against computers. I enjoy playing against humans, although I be it very badly, I must admit. And you can see from this chart of interest in chess on Google that a lot of people appear to feel the same way. Despite being essentially completed by computers, people still love chess. And I think that they love that human aspect of it. And I also think people will always love that about music. Because I mean, honestly, what could be more human than watching people play and sing and listening to music that's made by humans? But I think for this optimistic future to come about, I think you need one
thing. I think you've got to know what you're listening to, right? And that's not always the case on streaming sites. This is an AI artist, Holden Landry, and this is track Running Right Back on Deezer. You can see, very clearly, AI-generated content. And this is what it looks like on Spotify. No sign that it's AI. And with all due respect to Deezer, who are fantastic at providing all this data, Spotify is the biggest streaming site in the world. What they do has a much bigger impact. And right now, they're not labeling AI music. Feels kind of important. But it also means that I think this is important. Because very recently, Spotify took a step to address this.
They started a new program called Verified by Spotify. And you can see here, one of the criteria is, "This artist has grown an active fan base." What that means is that they want people, artists, who are connecting with people, who have real people who care about them. And another criteria they use is that these artists shouldn't be mostly AI or providing mostly AI music. Some notes can be AI, but not the whole thing. I think they're trying to get around the fact that in the future, maybe everything will have a little sprinkle of AI. But we still want to keep that essential human connection. And that is really the challenge here.
Because when I look at this whole thing of AI music, what I come back to is that AI has progressed so far that it's not really a question of whether or not AI can make convincing tracks anymore. It's kind of done. It's a question of whether, in the face of the AI tsunami, we can preserve that essential human element that I believe we all yearn for.
