Skip to content
Lex Fridman PodcastLex Fridman Podcast

Gustav Soderstrom: Spotify | Lex Fridman Podcast #29

Lex Fridman and Gustav Soderstrom on spotify’s Gustav Soderström on Music, AI, Creation, and the Future.

Lex FridmanhostGustav Soderstromguest
Jul 29, 20191h 47mWatch on YouTube ↗

EVERY SPOKEN WORD

  1. 0:00 – 15:00

    The following is a…

    1. LF

      The following is a conversation with Gustav Soderström. He's the chief research and development officer at Spotify, leading their product, design, data, technology, and engineering teams. As I've said before in my research and in life in general, I love music, listening to it and creating it, and using technology, especially personalization through machine learning to enrich the music discovery and listening experience. That is what Spotify has been doing for years, continually innovating, defining how we experience music as a society in the digital age. That's what Gustav and I talk about among many other topics, including our shared appreciation of the movie True Romance, in my view, one of the great movies of all time. This is the Artificial Intelligence podcast. If you enjoy it, subscribe on YouTube, give it five stars on iTunes, support it on Patreon, or simply connect with me on Twitter @lexfridman, spelled F-R-I-D-M-A-N. And now here's my conversation with Gustav Soderström. Spotify has over 50 million songs in its catalog, so let me ask the all-important question, I feel like you're the right person to ask, what is the definitive greatest song of all time?

    2. GS

      (laughs) It varies for me, personally.

    3. LF

      So you can't speak definitively for everyone?

    4. GS

      (laughs) I wouldn't believe very much in machine learning if I did, right?

    5. LF

      Okay.

    6. GS

      Because it meant everyone had the same taste.

    7. LF

      So for you, what is... you have to pick, what is the song?

    8. GS

      All right, so it's, it's pretty easy for me. There is this song called "You're So Cool" by Hans Zimmer, uh, soundtrack to True Romance.

    9. LF

      Ah.

    10. GS

      It was a movie that made a big impression on me and it's kind of been following me through my life. Actually had it play at my wedding. I sat with the organist and helped him play it on an organ, which was a pretty, pretty interesting experience. It's, uh...

    11. LF

      That is probably my, uh, I would say top three movie of all time. Yeah, it's just an incredible movie.

    12. GS

      Yeah, and, and it came out during my formative years and, uh, as I've discovered in music, you, you shape your music taste during those years, so it definitely affected me quite a bit.

    13. LF

      Did it affect you in any other kind of way?

    14. GS

      Well, the movie itself affected me back then, it was a big part of culture. I didn't really adopt any characters from the movie-

    15. LF

      (laughs)

    16. GS

      ... but it was a, it was a great story of love, some fantastic actors and, and, you know, really I didn't even know who Hans Zimmer was at the time, but fantastic music. And so, um, that song has followed me, and the movie actually has followed me throughout my life.

    17. LF

      That was Quentin Tarantino actually, I think, uh, director, directed or produced that, or... So it's not Stairway to Heaven or Bohemian Rhapsody, it's, uh...

    18. GS

      Tho- those are, those are great. They're not my personal favorites, but, uh-

    19. LF

      But they're out there.

    20. GS

      I've realized that people have different tastes and that's, uh, that's a big part of, of what we do.

    21. LF

      Well, for me, I would have to stick with Stairway to Heaven. So 35,000 years ago, I looked this up on Wikipedia, flute-like instruments started being used in caves as part of hunting rituals, in primitive cultural gatherings, things like that. This is the birth of music. Since then, we had a, a few folks, Beethoven, Elvis, Beatles, Justin Bieber of course, Drake. So in your view, let's start like high-level philosophical, what is the purpose of music on this planet of ours?

    22. GS

      I think music has many different purposes. I think there is, um, there is certainly a big purpose which is the same as much of entertainment, which is escapism and to be able to live in some sort of other mental state for a while. But I also think you have the, the opposite of escaping, which is to help you focus on something you are actually doing. And so I think people use music as a tool to, to tune the brain to the activities that they are actually doing and, um, it's kind of like, in one sense maybe it's the roost signal. If you, if you think about the brain as neural networks, it's maybe the most efficient hack we can do to actually actively tune it into some state that you want to be. And you can do it in other ways, you can tell stories to put people in a certain mood, but music is probably very effective to get you through a certain mood very fast, I think.

    23. LF

      You know, there's a, there's a social component historically to music, where people listen to music together. I was just thinking about this, that to me, and you mentioned machine learning, but to me personally, music is a really private thing. Like, I'm speaking for myself, I listen to music... like almost nobody knows the kind of things I have in my library except people who are really close to me, and they really only know a certain percentage. There's like some weird stuff that I'm almost probably embarrassed b- by, right?

    24. GS

      It's called the guilty pleasures, right? Everyone has them.

    25. LF

      (laughs) The guilty pleasures, yeah. Hopefully they're not too bad. But, uh, so it's a, I- for me it's personal. Do you think of music as something that's social or as something that's personal? Or does it vary?

    26. GS

      So I think it's the same, it's the same answer, that you use it for, for both. We- we've thought a lot about this during these 10 years at Spotify obviously. In one sense, as you said, music is incredibly social. You go to concerts and so forth. On the other hand, it is your, your escape, and everyone has these things that are very personal to them. So what we've found is that when it comes to, to, um... M- most people claim that they have a friend or two that they are heavily inspired by and that they listen to, so I actually think music is very social, but in a smaller group setting. It's an, it's an intimate form of...... of, um, it's an intimate relationship. It's not something that you necessarily share broadly. Now at concerts, you can argue you do. But then you've gathered a lot of people that you have something in common with. I think this broadcast sharing of music, um, is something we tried, uh, on social networks and so forth. But it turns out that people aren't super interested in just what their friends listen to.

    27. LF

      Mm-hmm.

    28. GS

      Um, they're interested in understanding if they have something in common, perhaps, with a friend, but not, you know, not just as, as information.

    29. LF

      Right. That's, that, that's really interesting. I, I was just thinking of it this morning listening to Spotify. I really have a pretty intimate relationship with Spotify, with my playlists, right? I've had them for many years now and they've grown with me together. There's, there's an intimate, uh, relationship you have with a library of music that you've developed. And we'll talk about different ways we can play with that. Can you, uh, do the impossible task and try to give a history of music listening from your perspective, from before the internet and after the internet. And just kinda everything leading up to streaming on Spotify and so on.

    30. GS

      I'll try. It could be a 100-year podcast, but... (laughs)

  2. 15:00 – 30:00

    Mm-hmm. …

    1. GS

      and then Pirate, in Sweden, Pirate Bay, which was one of the biggest. And it, you know, I think from a consumer point of view, which, which, which kind of leads up to the inception of, of Spotify, from a consuming point of view, consumers for the first time had this access model to music where they could, without kind of any marginal cost, they could, they could, um, try different tracks. You could use music in, in new ways. There was no marginal cost. And that was a fantastic consumer experience. To have access to all the music ever made, I think was fantastic, but it was also horrible for artists because there was no business model around it, so they didn't make any money. So, the, the user need, uh, almost drove the user interface bef- before there was a business model. And then there were these download stores, uh, that allowed you to download files, uh, which was a solution, but it didn't solve the access problem. There was still a marginal cost of 99 cents to try one more track.

    2. LF

      Mm-hmm.

    3. GS

      And I think that that heavily limits how you listen to music. The, the example I always give is, uh, you know, in Spotify, a huge amount of people listen to music while they sleep, while they go to sleep and while they sleep. If that costed you 99 cents per three minutes, you probably wouldn't do that. (laughs)

    4. LF

      (laughs)

    5. GS

      And you would be much less adventurous if there was a real dollar cost to exploring music. So, the access model is interesting in that it changes your music behavior. You can be... You can take much more risk 'cause there's no marginal cost to it.

    6. LF

      Maybe, let me linger on piracy for a second 'cause I, I find, uh, especially coming from Russia, piracy is something that's very interesting. To me, uh, not me of course ever, but, uh-

    7. GS

      (laughs) Of course.

    8. LF

      ... I have, I have friends who have partook in, uh, piracy of music, software, TV shows, sporting events. And usually, to me, what that shows is not that they're... They can actually pay the money and they're not trying to save money. They're choosing the best experience. So, what to me piracy shows is a business opportunity in all these domains, and, uh, that's where I, I think you're right Spotify stepped in, is basically piracy was, is an experience. You can explore, w- find music you like, and actually the interface of piracy isn't, is horrible because it's f- I mean, it's-

    9. GS

      Bad metadata.

    10. LF

      Yeah. Bad metadata, there's a lot-

    11. GS

      You know, long download times, all kinds of stuff.

    12. LF

      And, uh, what Spotify does is basically cre- (laughs) first, rewards artists, and second, makes the experience of exploring music much better. I mean, the same is true I think for movies and so on, that piracy reveals... In the software space, for example, I'm a huge user and fan of Adobe products.

    13. GS

      Mm-hmm.

    14. LF

      And, uh, the, there was much more incentive to pirate Adobe products before they went to a monthly subscription plan, and now all of the said friends that used, used to pirate Adobe products that I know, uh, now actually pay gladly for the monthly subscription.

    15. GS

      Yeah. I think you're right. I think it's an, it's a sign of an opportunity for product development, and that, um...Sometimes the, there's a product market fit b- before there's a business model fit.

    16. LF

      Right.

    17. GS

      Uh, in product development, I think that, that's a, that's a sign of it. In, in Sweden, I think it was a bit of both. There was, there was, um, a culture where we even had a political party called The Pirate Party. And this was during the time when, when people said that, you know, information should be free. It's s- somehow wrong to charge for ones and zeros. So I think people felt that artists should probably make s- money somehow else in, you know, concerts or something. So at least in Sweden, it was part really social acceptance, even at the political level and that... But that also forced Spotify to compete with, with free, which, uh, which I don't think would actually, could've happened anywhere else in the world. The, the music industry needed to be doing bad enough to take that risk. And Sweden was like the perfect testing ground. It had government-funded high bandwidth, low latency broadband, which meant that the product would work and it was also, there was no music revenue anyway. So they were kind of like, "I don't think this is gonna work, but why not?" Uh, so this product is one that I don't think could've happened in America. The world's largest music market, for example.

    18. LF

      So how do you compete with free? 'Cause that's an interesting world of the internet, where most people don't like to pay for things. So Spotify steps in and tries to, yes, compete with free. How do you do it?

    19. GS

      So I think two things. One is people are starting to pay for things on the internet. I think one way to think about it was that advertising was the first business model, because no one would put their credit card on internet. Transactional with Amazon was the second, and maybe subscription is the third. And if you look offline, subscription is the biggest of those. So that may still happen. I think people are starting to pay, but definitely back then, we needed to compete with free, and the first thing you need to do is obviously to lower the price to free. And then you need to be better somehow, and the way that Spotify was better was on the user experience, on the, on the actual performance, the latency of, uh, you know, even if, even if you had high bandwidth broadband, it would still take you 30 seconds to a minute to download one of these tracks. So the Spotify experience of starting within the perceptual limit of immediacy, about 250 milliseconds, meant that the, the, the whole trick was it felt as if you had downloaded all of Pirate Bay. It was on your hard drive. It was that fast, even though it wasn't, and it was still free. But somehow, you were actually still being a legal citizen. That, that was the trick that Spotify managed to, to pull off.

    20. LF

      So yeah, I've actually heard you, uh, say this or write this, and I was surprised that I wasn't aware of it because I just took it for granted. You know, whenever an awesome thing comes along, you're just like, "Oh, of course it has to be this way." That- that's exactly right, that it felt like the entire world's library is at my fingertips because of that, of that latency being reduced. What, what was the technical challenge in reducing the latency?

    21. GS

      So there was a, a group of really, really talented engineers, uh, one of them called Ludvig Strigeus. He wrote the... Actually from Gothenburg.

    22. LF

      Mm-hmm.

    23. GS

      He wrote the initial, um, uh, the uTorrent client, which is kind of an interesting backstory to Spotify, you know, that, uh, we have one of the top developers from, uh, from BitTorrent clients as well. So he wrote uTorrent, the world's smallest Bit Torrent client, and then, um, he, um, he was acquired very early by Daniel and Martin, who founded Spotify. And they actually sold the uTorrent client to Bit Torrent, but kept Ludvig. So Spotify had a lot of, uh, experience within peer-to-peer networking. So the original innovation was an, was a distribution innovation, where Spotify built an end-to-end media distribution system. Up until only a few years ago, we actually hosted all the music ourselves. So we had both the server side and the client, and that meant that we could do things such as having a peer-to-peer solution to use local caching, uh, on the client side, because back then, the world was mostly desktop.

    24. LF

      Mm-hmm.

    25. GS

      But we could also do things like, um, hack the TCP protocols, things like Nagle's algorithm for kind of exponential back-off or ramp-up, and just go full throttle and optimize for latency at the cost of bandwidth. And, uh, all, all of this end-to-end control meant that we could do an experience that felt like a, a step change. These days, we actually are on, on, um, GCP. We don't host our own stuff and, and everyone is really fast these days. So that was the initial competitive advantage, but then obviously you have to move on over time.

    26. LF

      And that was, uh, that was over 10 years ago, right?

    27. GS

      That was in 2008, the product was launched in Sweden. It was in a beta, I think 2007.

    28. LF

      And it was on the desktop, right? So-

    29. GS

      It was desktop only.

    30. LF

      There's no phone then.

  3. 30:00 – 45:00

    I like it how…

    1. GS

      And once you go through that mental journey of like, "It's still my files, they're just over there, and I just have $40 million on them, or $50 million on them or something now," then people are like, "Okay. That- that's good." The, the problem is I think because you paid us a subscription, if we hadn't had the free tier, where you would feel like, "Even if I don't want to pay anymore..."... I still get to keep them. You keep your playlist forever. They don't disappear even though you stop paying. I think that was really important. If we would have started as, you know, you can put in all this time but if you stop paying, you lose all your work. I think that would have been a big challenge and was the big challenge for a lot of our competitors. That's another reason why I think the free tier is really important. That people need to feel the security that the work they put in, it will never disappear, even if they decide not to pay.

    2. LF

      I like it how you put the work you put in. I actually stopped to even think of it that way. I just, actually Spotify taught me to just enjoy music.

    3. GS

      That's great.

    4. LF

      As oppo- as opposed to (laughs) as opposed to, uh, what I was doing before, which is like in an unhealthy way, hoarding music. Which I found that because I was doing that, I was listening to a small selection of songs way too much to where, where I was getting sick of them.

    5. GS

      Yeah.

    6. LF

      Whereas Spotify, the more liberating kind of approach is I was just enjoying... Of course I listened to Stairway to Heaven over and over, but I, because of the extra variety, I don't get as sick of them. There's an interesting statistic I saw th- so Spotify has, maybe you can correct me, but over 50 million songs, tracks, and over three billion playlists. So...

    7. GS

      Yes.

    8. LF

      Fifty million songs and three billion playlists. 60 times more playlists than songs. (laughs) Wh- what do you make of that?

    9. GS

      Yeah, so the way I think about it is that from a, um, from a statistician or machine learning point of view, you have all these, um, if you want to think about reinforcement learning, but you have this state space of all the tracks and you can take different journeys through this, through this world. And, um, these, I think of these as like people helping themselves and each other, creating interesting vectors through this space of tracks.

    10. LF

      Mm-hmm.

    11. GS

      And then it's not so surprising that across, you know, many tens of millions of kind of atomic units, there will be billions of paths that make sense, and we're probably pretty quite far away from having found all of them. So kind of our job now is users... When, when Spotify started, it was really a search box that was for the time pretty powerful, and then, uh, I like to refer to it as this programming language called playlisting-

    12. LF

      (laughs)

    13. GS

      ... where if you, as you probably were pretty good at music, you knew your new releases, you knew your back catalog, you knew your Stairway to Heaven, you could create a soundtrack for yourself using this playlisting tool that's like meta programming language for music to soundtrack your life. And people who are good at music, it's back to how do you scale the product. For people who are good at music, that was ac- actually enough. If you had the catalog and a good search tool and you can create your own sessions, you could create really good... A soundtrack for your entire life. Probably perfectly personalized because you did it yourself. But the problem was most people, many people aren't that good at music, they just can't spend the time. Even if you're very good at music, it's going to be hard to, to keep up. So what we did to try to scale this was to essentially try to build, you can think of them as agents that this, this friend that some people had that helped them navigate this music catalog, that's what we're trying to do for you.

    14. LF

      But also there is something like 200 million active users on Spotify.

    15. GS

      Yes.

    16. LF

      So there... Okay, so from the machine learning perspective, you have these 200 million people plus, uh, that are creating... It's, it's really interesting to think of, uh, playlists as, um... I mean, I don't know if you meant it that way, but it's almost like a programming language. It's, um, or at least a trace of, um, exploration of those individual agents, uh, th- the listeners, and you have all this new tracks coming in. So it's a fascinating space that, uh, is ripe for machine learning. So th- is there mo- is there, is it poss- how can playlists be used as data in terms of, uh, machine learning and, and to s- to help Spotify organize the music?

    17. GS

      So we found in our data, not surprising that people who playlisted lots, they retained much better. They had a great experience. And so our first attempt was to playlist for users. And so we acquired this company called Tunigo of editors and professional playlisters and kind of leveraged the maximum of, of, um, human intelligence to help, to help, uh, build li- li- kind of these vectors through the track space-

    18. LF

      Mm-hmm.

    19. GS

      ... for, for people. Uh, and that, that broadened the product. Then the, the obvious next... And, and we, you know, we used statistical means where they could see wha- when they created a playlist, how did that playlist perform, you know? They could see skips of the songs, they could see how the songs perform, and they manually iterated the playlist to maximize performance for a large group of people. But there were never enough editors to playlist for you personally. So the promise of machine learning was to go from kind of group personalization using editors and, and tools and ta- statistics to individualization. And then what's so interesting about the, the three billion playlists we have is we... And the, the truth is we lucked out. This was not a priority strategy as is often the case. It looks really smart in hindsight, but it's, it was dumb luck. Uh, we looked at these playlists and we had some people in the company, um, a person named Erik Bernhardsson, who was really good at machine learning already back in, in, back then in like 2007, 2008. Uh, back then it was mostly collaborative filtering and so forth. But we realized that what, what this is, is people are grouping tracks for themselves that have some semantic meaning to them, and then they actually label it with a playlist name as well. So in a sense, people were grouping tracks along semantic dimensions and labeling them. And so could you, could you, uh, use that information to find that, that latent embedding? And so we started playing around with collaborative filtering.And, uh, we saw tremendous success with it, basically trying to extract some of these, uh, some of these dimensions. And, and if you think about it, it's not surprising at all. It'd be quite surprising if playlists were actually random, if they had no semantic meaning.

    20. LF

      Okay.

    21. GS

      For, for most people, they group these tracks for some reason. So we just happened to cross this incredible data set where people are taking- taken these tens of millions of tracks and grouped them along different semantic vectors.

    22. LF

      Uh, and the semantics being outside the individual users, so it's some kind of universal... there's a universal embedding that holds across people on this earth.

    23. GS

      Yes. I, I do think that, um, the embeddings you find are gonna be reflective of the people who playlisted. So if, if you have a lot of indie lovers who playlist, your, your embedding is gonna perform better there. But what we found was that, yes, uh, there were these, these, uh, latent similarities. They were very powerful. And we, we had the... it was interesting because I think that the people who playlisted the most initially were the so-called music aficionados who, who were really into music, and they often had a certain... their taste was of- of- of often cert- geared towards a certain type of music. And so what surprised us, if, if you look at the problem from the outside, you might expect that the algorithms would start performing best with mainstreamers first, because it somehow feels like an easier problem to solve mainstream taste than really particular taste. It was the complete opposite for us.

    24. LF

      Hmm.

    25. GS

      The recommendations performed fantastically for people who saw themselves as having very unique taste. That's probably because all of them playlisted (laughs) and they didn't perform so well for mainstreamers. They actually thought they were a bit too particular and unorthodox. So we had the complete opposite of what we expected, success within the hardest problem first, and then had to try to scale to more mainstream recommendations.

    26. LF

      So, uh, you've also acquired Echo Nest, that analyzes song data. So, in your view, maybe you can talk about, so what kind of data is there from a machine learning perspective? There's a, like a huge amount w- we're talking about playlisting and just user data of, of what people are listening to, the playlist they're constructing and so on. Uh, and then there's the, the actual data within a song, what makes a song, I don't know, the, the actual waveforms, right? Is there any... th- how do you mix the two? How much value is there in each? To me, it seems like user data is, uh... well, it's a romantic notion that the song itself would contain useful information. But I, if I were to guess, user data would be much more powerful. Like playlists would be much more powerful.

    27. GS

      Yeah. So we use both. Uh, our biggest success initially w- was with playlist data without understanding anything about the structure of the song. But when we acquired the Echo Nest, they had the inverse problem. They actually didn't have any play data. They were just... they were a provider of recommendations, but they didn't actually have any play data. So they, they looked at the structure of songs sonically, and they looked at Wikipedia for cultural references and so forth, right?

    28. LF

      Oh, cool. Cool.

    29. GS

      And did a lot of NLU and so forth. So we got that skill into the company and combined kind of our user data with their, with their kind of, uh, uh, content-based. So y- you can think of it as we were user-based and they were content-based in their recommendations. And we combined those two. And for some cases, where you have a new song that has no, no play data, obviously you have to try to go by either, you know, who the artist is or, or the sonic information in the song or what it's similar to. So, so there's definitely value in, in both. And we do a lot in both. But I would say yes, the user data captures things that, that have to do with culture in the greater society that you would never see in the, in the content itself. But that said, we have seen... uh, we have a research lab in, in Paris when, you know, we can talk about more about that on kind of machine learning on the creator side, what it can do for creators, not just for the consumers. But what, where we looked at how does the structure of a song actually affect the listening behavior? And it turns out that there is a lot of... we can, we can predict things like skips based on this, you know, based on, on the song itself. We could say that maybe you should move that chorus a bit 'cause your skip is gonna go up here. There, there is a lot of latent structure in the music, which is not surprising 'cause it is some sort of mind hack. So there should be structure, that's probably what we respond to.

    30. LF

      You just blew my mind actually for, uh, from the creator perspective. Um, so that's really interesting topic, uh, that probably most creators aren't taking advantage of, right? So there's... I've recently got to interact with a few folks, YouTubers, who are, uh, like obsessed with this idea of what do I do to make sure people keep watching the video? And they like look at the analytics of, of which point do people turn it off and so on. First of all, I don't think that's healthy, but, uh, it's, it's... 'cause you can do it a little too much. But it is a really powerful tool for helping the creative process. You just made me realize you could do the same thing for creation of music. And so is that something you've looked into of, uh, how-

  4. 45:00 – 1:00:00

    Yeah. …

    1. LF

      say you become a slave to the YouTube algorithm or sort of, uh, it's a, it's always a danger of a new technology as opposed to, say, if you're creating a song, becoming too obsessed about the intro riff to the song that keeps people listening versus actually the entirety of the creation process.

    2. GS

      Yeah.

    3. LF

      It is a balance.

    4. GS

      Absolutely.

    5. LF

      But the fact that there's zero ... I mean, you're blowing my mind right now because you're, you're, uh, completely right that there is no signal whatsoever, there's no feedback whatsoever on the creation process in music or podcasting, uh, almost at all. And, uh, y- are, are you saying that Spotify is hoping to help create tools to, um, not tools but also-

    6. GS

      No tu- tools actually.

    7. LF

      Actually tools for creators?

    8. GS

      Yeah. Absolutely.

    9. LF

      Wow.

    10. GS

      So, we have, um, we've, we've made some acquisitions the last few years around music creation, this company called Soundtrap which is a DAW, digital audio workstation, but, uh, that is browser-based, and th- their focus was really the Google Docs approach where you can collaborate with people much more easily than you could in previous tools. So, we have some of these tools that we're working with that we want to make accessible. And then we can connect it with our, with our consumption data. We can create this feedback loop where we could help you understand, we could help you, uh, create and help you understand how you will perform. We also acquired this other company within podcasting called Anchor-

    11. LF

      Mm.

    12. GS

      ... which is one of the biggest podcasting tools. Uh, mobile-focused, so really focused on simple creation or easy access to creation, but that also gives us this feedback loop. And even before that, we, we invested in something called, uh, Spotify for Artists and Spotify for Podcasters-

    13. LF

      Mm-hmm.

    14. GS

      ... which is an app that you can download, you can verify that you are that creator. And then you get, you get things that, uh, you know, software developers have had for years. You can see where if you look at your podcast, for example, on Spotify or, or something you release, you can see how it's performing, which cities it's performing in, who's listening to it, what's the demographic breakup. So, so similar in the sense that you can understand how you're actually doing on the, on the platform. So, we, we definitely want to build tools. I think you also interviewed the, um, the head of research for Adobe, and I think that's an, that's an, back to Photoshop that you liked, I think that's an interesting analogy as well. Uh, Photoshop I think has been very innovative in helping photographers and, and artists. And I think there should be the same kind of tools for, for music creators where you could get, you know, AI assistance, for example, as you're creating music, uh, as you can do with, with Adobe where you can, "I want a sky over here," and you can get help creating that sky.

    15. LF

      The really fascinating thing is what Adobe doesn't have is a distribution for the content you create. So, you don't have the data of if I create, if I, uh, uh, you know, the, whatever creation I make in Photoshop or Premiere, I can't get, like, immediate feedback like I can on YouTube, for example, about the way people are responding.

    16. GS

      Exactly.

    17. LF

      And if Spotify is creating those tools out, that's a, it's a really exciting actually world. But le- let's talk a little about podcasts. It's ... So I have trouble talking to one person.

    18. GS

      (laughs)

    19. LF

      (laughs) So it's a bit terrifying and, uh, kind of hard to fathom, but on average, 60 to 100,000 people will listen to this episode. Okay? So, uh-

    20. GS

      That's intimidating.

    21. LF

      It's intimidating. Uh, so I host it on Blubrry.... I don't know if I'm pronouncing that correctly, actually. It looks like most people listen to it on Apple Podcasts, Castbox, and Pocket Casts, and only about 1,000, uh, listen on Spotify, in th- just my podcast, right? So, h- how, wh- where ... (laughs) Do you see a time when Spotify will dominate this? So, Spotify is relatively new, uh, in, into this-

    22. GS

      In podcasting definitely.

    23. LF

      ... in podcasting, sorry, yeah, in podcasting. What's the deal with podcasting and Spotify? Uh, how serious is Spotify about podcasting? Do you see a time where everybody would listen to ... You know, probably a huge amount of people, majority perhaps, listen to music on Spotify. Do you see a time when the same is true for podcasting?

    24. GS

      Well, I certainly hope so. That is our mission. Our mission as a company is actually to enable a million creators to live off of their art, and a billion people be inspired by it. And what I think it- is interesting about that mission is, it actually puts the creators first, even though it started as a consumer-focused company, and it says to be able to live off of their art, not just make some money off of their art as well. So, it's a qu- it's quite, um, an ambitious project. And, um, so we think about creators of all kinds, and, uh, we kind of expanded our mission from being music to being audio a while back, and, uh, that's not so much because we think we made that decision. We think that m- decision was, was made for us; we think the world made that decision. Whether we like it or not, when you put in your headphones, you're going to make a choice between music and a new episode of, of, uh, of y- of your podcast or something else, right?

    25. LF

      Yeah.

    26. GS

      We're, we're in that world whether we like it or not.

    27. LF

      Yeah.

    28. GS

      And that, you know, that's how radio worked.

    29. LF

      Yes.

    30. GS

      So, we decided that, um, we think it's about audio. You, you can see the rise of audiobooks and so forth. We think audio is this great opportunity, so we decided to enter it, and, and obviously, uh, uh, Apple and Apple Podcasts is, is absolutely dominating in, in, um, podcasting and we didn't have a single podcast only, like, two years ago. What we did, though, was we, we, we looked at this and said, "You know, can we bring something to this?" Uh, you know, we, we want to do this, but the ... back to the original Spotify, we have to do something that consumers actually value-

  5. 1:00:00 – 1:15:00

    So jumping back into…

    1. GS

      listens to, how they make money today, try to, you know, uh, make sure that their business model works, that they understand. You, I think it's back to doing something, i- improving their product, like feedback loops and, and distribution.

    2. LF

      So jumping back into terms of this fascinating world of, uh, recommender system and listening to music and using machine learning to analyze things, do you think it, it's better to what currently, correct me if I'm wrong, but currently Spotify lets people pick what they listen to for the most part? There's a discovery process, but you kind of organize playlists. Uh, is it better to let people pick what they listen to or recommend what they should listen to?... something like Stations by Spotify-

    3. GS

      Yeah.

    4. LF

      ... that I saw that you're playing around with. Maybe you can tell me what's the status of that? This is, uh, Pandora style app that just kind of, as opposed to you select the music you listen to, it kind of feeds you the music you listen to. What's the status of Stations by Spotify? What's its future?

    5. GS

      The story of Spotify, as we have grown, has been that we made it more accessible to different, different audiences.

    6. LF

      Mm-hmm.

    7. GS

      And, um, Stations is another one of those where the question is some people want to be very specific. They actually want to hear Stairway to Heaven right now. That needs to be very easy to do. And some people, or even the same person, at some point might say, "I want to feel upbeat," or, "I want to feel happy," or, "I want songs to sing in the car."

    8. LF

      Right.

    9. GS

      Right? So they put in, they put in the information at a very different level, and then we need to translate that into that, what that means musically. So Stations is a test to, to create like a consumption input vector that is much simpler, where you can just tune it a little bit and, and see if that increases the overall reach. But we're trying to kind of serve the entire gamut of super advanced, so-called afic- music aficionados, all the way to, to people who they love listening to music, but it's, it's not their number one priority in life, right? They're not gonna sit and follow every new release from every new artist. They need to be able to, to influence music at a, at a, at a different level. So we're trying... You can think of it as different products. And I think when one of the, one of the interesting things, uh, to answer your question on if it's better to let the user choose or to play, I think the answer is the, the challenge when you, um, when, when machine learning kind of came along, there was a lot of thinking about wha- what does product development mean-

    10. LF

      Mm-hmm.

    11. GS

      ... in a, in a machine learning context? People like Andrew Ng, for example, when he went to Baidu, he started doing a lot of practical machine learning, went from academia and, and, you know, he thought a lot about this. And, and he, he, he had this notion that, you know, a product manager, designer, and engi- they used to work around this wireframe, kind of describe what the product should look like or some talk about when you're doing like a chatbot or a playlist, how do you... What are you gonna say? Like it should be good?

    12. LF

      (laughs)

    13. GS

      That's not a good product description. So how do you, how do you do that? And he came up with this notion that, um, the test set is the new wireframe. The, the job of the product manager is to source a good test set that is representative of what... Like if you say like, "I want to play the status songs to sing in the car," the job of the product manager is to go and source like a good test set of what that means.

    14. LF

      Mm-hmm.

    15. GS

      So then you can work with engineering to have algorithms to try to produce that, right? So we, we try to think a lot about how to structure product development for, for a machine learning age. And, and what we discovered was that a lot of it is actually in the expectation. And you can go, you can go two ways. So let's say that if you, if you set the expectation with the user that this is a discovery product, like Discover Weekly-

    16. LF

      Mm-hmm.

    17. GS

      ... you're actually setting the expectation that most of what we show you will not be relevant. When you're in the discovery process, you're gonna accept that, actually, if you find one gem every Monday that you totally love, you're probably gonna be happy. Even though the statistical meaning one out of 10 is terrible or one out of 20 is terrible from a user point of view, because the setting was Discovery, it's fine, but-

    18. LF

      Can I... Yeah, sorry to interrupt real quick.

    19. GS

      Yeah.

    20. LF

      I just actually learned about Discover Weekly, which is a Spotify, I don't know, it's, it's a feature of Spotify that shows you cool songs to listen to. I, uh, maybe I can do issue tracking. I couldn't find it on my Spotify app.

    21. GS

      It's, it's in your library.

    22. LF

      It's in the library. It's in the list of live.

    23. GS

      Yeah.

    24. LF

      'Cause I was like, "Whoa, this is cool. I didn't know this existed," and I tried to find it, but, uh, okay. (laughs)

    25. GS

      I, I will show it to you and feedback to our product teams.

    26. LF

      (laughs) Yeah.

    27. GS

      Maybe you can find it.

    28. LF

      There you go. But yeah, it's a... So yeah, sorry. J- just to, uh, j- just to mention the expectation there is basically they, you're going to discover new songs.

    29. GS

      Yeah. So, so then you can be quite adventurous in, in the recommendations you do. But, but if you're... But we have another product called, uh, Daily Mix, which kind of implies that these are only gonna be your favorites.

    30. LF

      Mm-hmm.

  6. 1:15:00 – 1:19:04

    Okay. …

    1. GS

      playlist and saying that, "These are new tracks that we think you might like based on this." And setting the right expectation made it, made it a great product. So I think we have this benefit that, for example, Tesla doesn't have, that we can, we can, we can change the expectation. We can, we can build a fault tolerant setting. It's very hard to be fault tolerant when you're driving at a, you know, 100 miles per hour or something.

    2. LF

      Okay.

    3. GS

      And, and we, we have the luxury of being able to say that, of being wrong if we have the right UI, which gives us different abilities to take more risk. So I actually think the self-driving problem is, is much harder.

    4. LF

      Oh, yeah, for sure. It's much less fun because people die.

    5. GS

      Exactly.

    6. LF

      And since Spotify, uh, it's just such a more fun problem because failure will, uh, I mean, failure is beautiful in a way. It leads to exploration. So it's, it's a really fun reinforcement learning problem, right?

    7. GS

      And the, the worst case scenario is you get these WTF tweets like-

    8. LF

      Right.

    9. GS

      ... "How the hell did I get this?"

    10. LF

      This song, yeah.

    11. GS

      Which is, which is a lot better than the self-driving fail case, right?

    12. LF

      Exactly. So what's the feedback that a user pro- what's the signal that a user provides into the system? So the f- the, you mentioned skipping. What is, like, the strongest signal? Is, uh, you didn't mention clicking Like.

    13. GS

      So, so we have a few signals that are important. Obviously playing, playing through. So, so one of the benefits of music actually, even compared to podcasts or, or, uh, movies is, the object itself is really only about three minutes. So you get a lot of chances to recommend and the feedback loop is, is every three minutes instead of every two hours or something. So you actually get kind of noisy but, but, uh, quite fast feedback. And so you can see if people played through or if the, which is, you know, the inverse of skip really. That's an important signal. On the other hand, much of the consumption happens when your phone is in your pocket, maybe you're running or driving or you're playing on a speaker.

    14. LF

      Mm-hmm.

    15. GS

      And so you not skipping doesn't mean that you love that song. It might be that it wasn't bad enough that you would walk up and skip. So it's a noisy signal. Then, then we have the equivalent of the Like, which is you saved it to your library. That's a pretty strong signal of affection. And then we have the more explicit signal of play listing. Like, you took the time to create a playlist, you put it in there. There's a very l- little, small chance that if you took all that trouble, this is not a really important track to you. And then we understand also what other tracks it relates to. So we have, we have the playlisting, we have the Like, and then we have the listening or skip. And, and you have to have very different approaches to all of them because they have different levels of, of noise. One, one is very voluminous but noisy, and the other's rare, but you can r- you can probably trust it.

    16. LF

      Yeah, it's interesting because, uh, I, I think between those signals captures all the information you'd want to capture. I mean, there's a feeling, a shallow feeling for me that there's sometimes that I'll hear a song and it's like, "Yes, this is, you know, this was the right song for the moment." But there's really no way to express that fact except by listening through it all the way.

    17. GS

      Yeah.

    18. LF

      And, and maybe playing it again at that time or something, but-

    19. GS

      Yeah.

    20. LF

      ... there, there's, there's no need for a button that says, uh, "This was the best song I could have heard at this moment."

    21. GS

      Well, we're, we're playing around with that, with, with kind of the thumbs up concept, saying like, "I really like this," just kind of talking to the algorithm. It's unclear if that's, um, the best way for humans to interact. Maybe it is. Maybe they should think of Spotify as a person, an agent sitting there trying to serve you and you can say, like, "Bad Spotify, good Spotify." Right now, the analogy we've had is more you shouldn't think of, of us, we should be investible. And the feedback is if you save it, kind of you work for yourself, you do a playlist because you think it's great and we can learn from that. It's kind of back to, back to Tesla, how they kind of have this shadow mode. They sit and watch you drive. We kind of took the same analogy. We sit and watch you playlist, and then maybe we can, we can offer you an autopilot where we can take over for a while or something like that-

Episode duration: 1:47:03

Install uListen for AI-powered chat & search across the full episode — Get Full Transcript

Transcript of episode v-9Mpe7NhkM

Get more out of YouTube videos.

High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.

Add to Chrome