Transcript
00:00:00 The following is a conversation with Gustav Sorenstrom.
00:00:03 He’s the chief research and development officer at Spotify,
00:00:07 leading their product design, data technology and engineering teams.
00:00:11 As I’ve said before, in my research and in life in general,
00:00:15 I love music, listening to it and creating it.
00:00:18 And using technology, especially personalization through machine learning,
00:00:23 to enrich the music discovery and listening experience.
00:00:27 That is what Spotify has been doing for years, continually innovating,
00:00:31 defining how we experience music as a society in the digital age.
00:00:36 That’s what Gustav and I talk about, among many other topics,
00:00:39 including our shared appreciation of the movie True Romance,
00:00:43 in my view, one of the great movies of all time.
00:00:46 This is the Artificial Intelligence Podcast.
00:00:49 If you enjoy it, subscribe on YouTube, give it five stars on iTunes,
00:00:53 support on Patreon or simply connect with me on Twitter at Lex Friedman,
00:00:58 spelled F R I D M A N.
00:01:01 And now, here’s my conversation with Gustav Sorenstrom.
00:01:06 Spotify has over 50 million songs in its catalog.
00:01:10 So let me ask the all important question.
00:01:14 I feel like you’re the right person to ask.
00:01:16 What is the definitive greatest song of all time?
00:01:19 It varies for me, personally.
00:01:22 So you can’t speak definitively for everyone?
00:01:26 I wouldn’t believe very much in machine learning if I did, right?
00:01:30 Because everyone had the same taste.
00:01:32 So for you, what is… you have to pick. What is the song?
00:01:36 All right, so it’s pretty easy for me.
00:01:39 There’s this song called You’re So Cool, Hans Zimmer, a soundtrack to True Romance.
00:01:46 It was a movie that made a big impression on me.
00:01:49 And it’s kind of been following me through my life.
00:01:51 I actually had it play at my wedding.
00:01:55 I sat with the organist and helped him play it on an organ,
00:01:58 which was a pretty interesting experience.
00:02:01 That is probably my, I would say, top three movie of all time.
00:02:06 Yeah, this is an incredible movie.
00:02:07 Yeah, and it came out during my formative years.
00:02:10 And as I’ve discovered in music, you shape your music taste during those years.
00:02:15 So it definitely affected me quite a bit.
00:02:17 Did it affect you in any other kind of way?
00:02:20 Well, the movie itself affected me back then.
00:02:23 It was a big part of culture.
00:02:25 I didn’t really adopt any characters from the movie,
00:02:27 but it was a great story of love, fantastic actors.
00:02:33 And really, I didn’t even know who Hans Zimmer was at the time, but fantastic music.
00:02:39 And so that song has followed me.
00:02:42 And the movie actually has followed me throughout my life.
00:02:43 That was Quentin Tarantino, actually, I think, director or producer.
00:02:48 So it’s not Stairway to Heaven or Bohemian Rhapsody.
00:02:52 Those are great.
00:02:53 They’re not my personal favorites, but I’ve realized that people have different tastes.
00:02:57 And that’s a big part of what we do.
00:03:00 Well, for me, I would have to stick with Stairway to Heaven.
00:03:04 So 35,000 years ago, I looked this up on Wikipedia,
00:03:09 flute like instruments started being used in caves as part of hunting rituals.
00:03:13 And primitive cultural gatherings, things like that.
00:03:15 This is the birth of music.
00:03:18 Since then, we had a few folks, Beethoven, Elvis, Beatles, Justin Bieber, of course, Drake.
00:03:25 So in your view, let’s start like high level philosophical.
00:03:29 What is the purpose of music on this planet of ours?
00:03:35 I think music has many different purposes.
00:03:38 I think there’s certainly a big purpose, which is the same as much of entertainment,
00:03:44 which is escapism and to be able to live in some sort of other mental state for a while.
00:03:52 But I also think you have the opposite of escaping,
00:03:54 which is to help you focus on something you are actually doing.
00:03:57 Because I think people use music as a tool to tune the brain
00:04:02 to the activities that they are actually doing.
00:04:05 And it’s kind of like, in one sense, maybe it’s the rawest signal.
00:04:10 If you think about the brain as neural networks,
00:04:13 it’s maybe the most efficient hack we can do to actually actively tune it
00:04:16 into some state that you want to be.
00:04:18 You can do it in other ways.
00:04:19 You can tell stories to put people in a certain mood.
00:04:22 But music is probably very effective to get you to a certain mood very fast, I think.
00:04:27 You know, there’s a social component historically to music,
00:04:30 where people listen to music together.
00:04:32 I was just thinking about this, that to me, and you mentioned machine learning,
00:04:36 but to me personally, music is a really private thing.
00:04:43 I’m speaking for myself, I listen to music,
00:04:45 like almost nobody knows the kind of things I have in my library,
00:04:50 except people who are really close to me and they really only know a certain percentage.
00:04:54 There’s like some weird stuff that I’m almost probably embarrassed by, right?
00:04:58 It’s called the guilty pleasures, right?
00:05:00 Everyone has the guilty pleasures, yeah.
00:05:02 Hopefully they’re not too bad, but for me, it’s personal.
00:05:06 Do you think of music as something that’s social or as something that’s personal?
00:05:12 Or does it vary?
00:05:14 So I think it’s the same answer that you use it for both.
00:05:20 We’ve thought a lot about this during these 10 years at Spotify, obviously.
00:05:25 In one sense, as you said, music is incredibly
00:05:27 social, you go to concerts and so forth.
00:05:30 On the other hand, it is your escape and everyone has these things that are very personal to them.
00:05:38 So what we’ve found is that when it comes to, most people claim that they have a friend or two
00:05:47 that they are heavily inspired by and that they listen to.
00:05:50 So I actually think music is very social, but in a smaller group setting,
00:05:54 it’s an intimate form of, it’s an intimate relationship.
00:06:00 It’s not something that you necessarily share broadly.
00:06:03 Now, at concerts, you can argue you do, but then you’ve gathered a lot of people
00:06:07 that you have something in common with.
00:06:08 I think this broadcast sharing of music is something we tried on social networks and so forth.
00:06:16 But it turns out that people aren’t super interested in sharing their music.
00:06:23 They aren’t super interested in what their friends listen to.
00:06:28 They’re interested in understanding if they have something in common perhaps with a friend,
00:06:32 but not just as information.
00:06:35 Right, that’s really interesting.
00:06:38 I was just thinking of it this morning, listening to Spotify.
00:06:41 I really have a pretty intimate relationship with Spotify, with my playlists, right?
00:06:48 I’ve had them for many years now and they’ve grown with me together.
00:06:53 There’s an intimate relationship you have with a library of music that you’ve developed.
00:06:59 And we’ll talk about different ways we can play with that.
00:07:02 Can you do the impossible task and try to give a history of music listening
00:07:09 from your perspective from before the internet and after the internet
00:07:14 and just kind of everything leading up to streaming with Spotify and so on?
00:07:18 I’ll try.
00:07:19 It could be a 100 year podcast.
00:07:22 I’ll try to do a brief version.
00:07:24 There are some things that I think are very interesting during the history of music,
00:07:28 which is that before recorded music, to be able to enjoy music,
00:07:33 you actually had to be where the music was produced
00:07:35 because you couldn’t record it and time shift it, right?
00:07:38 Creation and consumption had to happen at the same time, basically concerts.
00:07:41 And so you either had to get to the nearest village to listen to music.
00:07:46 And while that was cumbersome and it severely limited the distribution of music,
00:07:51 it also had some different qualities,
00:07:53 which was that the creator could always interact with the audience.
00:07:56 It was always live.
00:07:58 And also there was no time cap on the music.
00:08:00 So I think it’s not a coincidence that these early classical works,
00:08:04 they’re much longer than the three minutes.
00:08:06 The three minutes came in as a restriction of the first wax disc that could only contain
00:08:11 a three minute song on one side, right?
00:08:14 So actually the recorded music severely limited or put constraints.
00:08:20 I won’t say limit.
00:08:21 I mean, constraints are often good,
00:08:22 but it put very hard constraints on the music format.
00:08:24 So you kind of said, instead of doing this opus on many tens of minutes or something,
00:08:31 now you get three and a half minutes because then you’re out of wax on this disc.
00:08:34 But in return, you get an amazing distribution.
00:08:37 Your reach will widen, right?
00:08:39 Just on that point real quick.
00:08:42 Without the mass scale distribution, there’s a scarcity component
00:08:47 where you kind of look forward to it.
00:08:51 We had that, it’s like the Netflix versus HBO Game of Thrones.
00:08:56 You like wait for the event because you can’t really listen to it.
00:09:00 So you like look forward to it and then it’s like,
00:09:02 you derive perhaps more pleasure because it’s more rare for you to listen to a particular piece.
00:09:07 You think there’s value to that scarcity?
00:09:10 Yeah, I think that that is definitely a thing.
00:09:12 And there’s always this component of if you have something in infinite amounts,
00:09:17 will you value it as much?
00:09:20 Probably not.
00:09:20 Humanity is always seeking some, it’s relative.
00:09:24 So you’re always seeking something you didn’t have.
00:09:25 And when you have it, you don’t appreciate it as much.
00:09:27 So I think that’s probably true.
00:09:29 But I think that that’s probably true.
00:09:31 But I think that’s why concerts exist.
00:09:33 So you can actually have both.
00:09:35 But I think net, if you couldn’t listen to music in your car driving, that’d be worse.
00:09:42 That cost will be bigger than the benefit of the anticipation I think that you would have.
00:09:47 So, yeah, it started with live concerts.
00:09:50 Then it’s being able to, you know, the phonograph invented, right?
00:09:56 That you start to be able to record music.
00:09:59 Exactly.
00:09:59 So then you got this massive distribution that made it possible to create two things.
00:10:04 I think, first of all, cultural phenomenons, they probably need distribution to be able to happen.
00:10:10 But it also opened access to, you know, for a new kind of artist.
00:10:15 So you started to have these phenomenons like Beatles and Elvis and so forth.
00:10:18 That would really, a function of distribution, I think, obviously of talent and innovation.
00:10:23 But there was also technical component.
00:10:25 And of course, the next big innovation to come along was radio.
00:10:29 Broadcast radio.
00:10:30 And I think radio is interesting because it started not as a music medium.
00:10:36 It started as an information medium for news.
00:10:39 And then radio needed to find something to fill the time with so that they could honestly
00:10:45 play more ads and make more money.
00:10:47 And music was free.
00:10:48 So then you had this massive distribution where you could program to people.
00:10:52 I think those things, that ecosystem, is what created the ability for hits.
00:10:59 But it was also a very broadcast medium.
00:11:01 So you would tend to get these massive, massive hits, but maybe not such a long tail.
00:11:07 In terms of choice of everybody listens to the same stuff.
00:11:10 Yeah.
00:11:10 And as you said, I think there are some social benefits to that.
00:11:14 I think, for example, there’s a high statistical chance that if I talk about the latest episode
00:11:19 of Game of Thrones, we have something to talk about, just statistically.
00:11:23 In the age of individual choice, maybe some of that goes away.
00:11:26 So I do see the value of shared cultural components, but I also obviously love personalization.
00:11:36 And so let’s catch this up to the internet.
00:11:39 So maybe Napster, well, first of all, there’s MP3s, tapes, CDs.
00:11:44 There was a digitalization of music with a CD, really.
00:11:47 It was physical distribution, but the music became digital.
00:11:51 And so they were files, but basically boxed software, to use a software analogy.
00:11:56 And then you could start downloading these files.
00:11:59 And I think there are two interesting things that happened.
00:12:02 Back to music used to be longer before it was constrained by the distribution medium.
00:12:08 I don’t think that was a coincidence.
00:12:09 And then really the only music genre to have developed mostly after music was a file again
00:12:15 on the internet is EDM.
00:12:17 And EDM is often much longer than the traditional music.
00:12:20 I think it’s interesting to think about the fact that music is no longer constrained in
00:12:26 minutes per song or something.
00:12:27 It’s a legacy of an old distribution technology.
00:12:31 And you see some of this new music that breaks the format.
00:12:33 Not so much as I would have expected actually by now, but it still happens.
00:12:38 So first of all, I don’t really know what EDM is.
00:12:41 Electronic dance music.
00:12:42 Yeah.
00:12:42 You could say Avicii.
00:12:44 Avicii was one of the biggest in this genre.
00:12:46 So the main constraint is of time.
00:12:49 Something like a three, four, five minute song.
00:12:52 So you could have songs that were eight minutes, 10 minutes and so forth.
00:12:56 Because it started as a digital product that you downloaded.
00:13:01 So you didn’t have this constraint anymore.
00:13:03 So I think it’s something really interesting that I don’t think has fully happened yet.
00:13:08 We’re kind of jumping ahead a little bit to where we are, but I think there’s tons of format
00:13:12 innovation in music that should happen now, that couldn’t happen when you needed to really
00:13:18 adhere to the distribution constraints.
00:13:20 If you didn’t adhere to that, you would get no distribution.
00:13:24 So Björk, for example, the Icelandic artist, she made a full iPad app as an album.
00:13:30 That was very expensive.
00:13:33 Even though the app store has great distribution, she gets nowhere near the distribution versus
00:13:38 staying within the three minute format.
00:13:39 So I think now that music is fully digital inside these streaming services, there is
00:13:44 the opportunity to change the format again and allow creators to be much more creative
00:13:50 without limiting their distribution ability.
00:13:52 That’s interesting that you’re right.
00:13:54 It’s surprising that we don’t see that taken advantage more often.
00:13:59 It’s almost like the constraints of the distribution from the 50s and 60s have molded the culture
00:14:06 to where we want the five, three to five minute song than anything else, not just.
00:14:12 So we want the song as consumers and as artists, because I write a lot of music and I never
00:14:18 even thought about writing something longer than 10 minutes.
00:14:23 It’s really interesting that those constraints.
00:14:26 Because all your training data has been three and a half minute songs, right?
00:14:29 It’s right.
00:14:30 Okay, so yes, digitization of data led to then mp3s.
00:14:36 Yeah, so I think you had this file then that was distributed physically, but then you had
00:14:42 the components of digital distribution and then the internet happened and there was this
00:14:46 vacuum where you had a format that could be digitally shipped, but there was no business
00:14:51 model.
00:14:51 And then all these pirate networks happened, Napster and in Pirate Island.
00:14:58 Napster and in Sweden Pirate Bay, which was one of the biggest.
00:15:02 And I think from a consumer point of view, which kind of leads up to the inception of
00:15:10 Spotify, from a consumer point of view, consumers for the first time had this access model to
00:15:15 music where they could, without kind of any marginal cost, they could try different tracks.
00:15:25 You could use music in new ways.
00:15:27 There was no marginal cost.
00:15:28 And that was a fantastic consumer experience to have access to all the music ever made,
00:15:32 I think was fantastic.
00:15:34 But it was also horrible for artists because there was no business model around it.
00:15:38 So they didn’t make any money.
00:15:39 So the user need almost drove the user interface before there was a business model.
00:15:46 And then there were these download stores that allowed you to download files, which
00:15:52 was a solution, but it didn’t solve the access problem.
00:15:55 There was still a marginal cost of 99 cents to try one more track.
00:15:58 And I think that that heavily limits how you listen to music.
00:16:01 The example I always give is, you know, in Spotify, a huge amount of people listen to
00:16:07 music while they sleep, while they go to sleep and while they sleep.
00:16:11 If that costed you 99 cents per three minutes, you probably wouldn’t do that.
00:16:15 And you would be much less adventurous if there was a real dollar cost to exploring
00:16:18 music.
00:16:19 So the access model is interesting in that it changes your music behavior.
00:16:22 You can be, you can take much more risk because there’s no marginal cost to it.
00:16:27 Maybe let me linger on piracy for a second, because I find, especially coming from Russia,
00:16:33 piracy is something that’s very interesting to me.
00:16:39 Not me, of course, ever, but I have friends who have partook in piracy of music, software,
00:16:49 TV shows, sporting events.
00:16:52 And usually to me, what that shows is not that they’re, they can actually pay the money
00:16:58 and they’re not trying to save money.
00:17:00 They’re choosing the best experience.
00:17:03 So what to me, piracy shows is a business opportunity in all these domains.
00:17:08 And that’s where I think you’re right.
00:17:11 Spotify stepped in is basically piracy was an experience.
00:17:15 You can explore with fine music you like, and actually the interface of piracy is horrible
00:17:23 because it’s, I mean, it’s bad metadata, long download times, all kinds of stuff.
00:17:29 And what Spotify does is basically first rewards artists and second makes the experience of
00:17:37 exploring music much better.
00:17:38 I mean, the same is true, I think for movies and so on.
00:17:42 That piracy reveals in the software space, for example, I’m a huge user and fan of Adobe
00:17:48 products and there was much more incentive to pirate Adobe products before they went
00:17:54 to a monthly subscription plan.
00:17:57 And now all of the said friends that used to pirate Adobe products that I know now actually
00:18:04 pay gladly for the monthly subscription.
00:18:06 Yeah, I think you’re right.
00:18:08 I think it’s a sign of an opportunity for product development.
00:18:11 And that sometimes there’s a product market fit before there’s a business model fit in
00:18:19 product development.
00:18:19 I think that’s a sign of it.
00:18:21 In Sweden, I think it was a bit of both.
00:18:24 There was a culture where we even had a political party called the Pirate Party.
00:18:30 And this was during the time when people said that information should be free.
00:18:35 It was somehow wrong to charge for ones and zeros.
00:18:38 So I think people felt that artists should probably make some money somehow else and
00:18:43 concerts or something.
00:18:44 So at least in Sweden, it was part really social acceptance, even at the political level.
00:18:49 But that also forced Spotify to compete with free, which I don’t think would actually
00:18:56 could have happened anywhere else in the world.
00:18:58 The music industry needed to be doing bad enough to take that risk.
00:19:03 And Sweden was like the perfect testing ground.
00:19:04 It had government funded high bandwidth, low latency broadband, which meant that the product
00:19:10 would work.
00:19:11 And it was also there was no music revenue anyway.
00:19:14 So they were kind of like, I don’t think this is going to work, but why not?
00:19:18 So this product is one that I don’t think could have happened in America, the world’s
00:19:21 largest music market, for example.
00:19:23 So how do you compete with free?
00:19:25 Because that’s an interesting world of the internet where most people don’t like to
00:19:30 pay for things.
00:19:31 So Spotify steps in and tries to, yes, compete with free.
00:19:36 How do you do it?
00:19:37 So I think two things.
00:19:38 One is people are starting to pay for things on the internet.
00:19:41 I think one way to think about it was that advertising was the first business model because
00:19:47 no one would put a credit card on the internet.
00:19:49 Transactional with Amazon was the second.
00:19:51 And maybe subscription is the third.
00:19:52 And if you look offline, subscription is the biggest of those.
00:19:56 So that may still happen.
00:19:57 I think people are starting to pay for things.
00:19:59 But definitely back then, we needed to compete with free.
00:20:02 And the first thing you need to do is obviously to lower the price to free and then you need
00:20:07 to be better somehow.
00:20:09 And the way that Spotify was better was on the user experience, on the actual performance,
00:20:15 the latency of, you know, even if you had high bandwidth broadband, it would still take
00:20:24 you 30 seconds to a minute to download one of these tracks.
00:20:30 So the Spotify experience of starting within the perceptual limit of immediacy, about 250
00:20:35 milliseconds, meant that the whole trick was it felt as if you had downloaded all of Pirate
00:20:41 Bay.
00:20:41 It was on your hard drive.
00:20:42 It was that fast, even though it wasn’t.
00:20:45 And it was still free.
00:20:46 But somehow you were actually still being a legal citizen.
00:20:50 And that was the trick that Spotify managed to pull off.
00:20:54 So I’ve actually heard you say this or write this.
00:20:58 And I was surprised that I wasn’t aware of it because I just took it for granted.
00:21:02 You know, whenever an awesome thing comes along, you’re just like, of course, it has
00:21:05 to be this way.
00:21:07 That’s exactly right.
00:21:08 That it felt like the entire world’s libraries at my fingertips because of that latency being
00:21:14 reduced.
00:21:15 What was the technical challenge in reducing the latency?
00:21:18 So there was a group of really, really talented engineers, one of them called Ludwig Strigius.
00:21:25 He wrote the, actually from Gothenburg, he wrote the initial, the uTorrent client, which
00:21:32 is kind of an interesting backstory to Spotify, that we have one of the top developers from
00:21:38 uTorrent clients as well.
00:21:39 So he wrote uTorrent, the world’s smallest uTorrent client.
00:21:42 And then he was acquired very early by Daniel and Martin, who founded Spotify, and they
00:21:49 actually sold the uTorrent client to BitTorrent, but kept Ludwig.
00:21:53 So Spotify had a lot of experience within peer to peer networking.
00:21:59 So the original innovation was a distribution innovation, where Spotify built an end to
00:22:04 end media distribution system up until only a few years ago, we actually hosted all the
00:22:08 music ourselves.
00:22:09 So we had both the service side and the client, and that meant that we could do things such
00:22:13 as having a peer to peer solution to use local caching on the client side, because back then
00:22:19 the world was mostly desktop.
00:22:20 But we could also do things like hack the TCP protocols, things like Nagel’s algorithm
00:22:26 for kind of exponential back off, or ramp up and just go full throttle and optimize
00:22:31 for latency at the cost of bandwidth.
00:22:33 And all of this end to end control meant that we could do an experience that felt like a
00:22:39 step change.
00:22:40 These days, we actually are on GCP, we don’t host our own stuff, and everyone is really
00:22:46 fast these days.
00:22:47 So that was the initial competitive advantage.
00:22:49 But then obviously, you have to move on over time.
00:22:51 And that was over 10 years ago, right?
00:22:54 That was in 2008.
00:22:55 The product was launched in Sweden.
00:22:57 It was in a beta, I think, 2007.
00:22:59 And it was on the desktop, right?
00:23:00 It was desktop only.
00:23:01 There’s no phone.
00:23:03 There was no phone.
00:23:04 The iPhone came out in 2008.
00:23:07 But the App Store came out one year later, I think.
00:23:10 So the writing was on the wall, but there was no phone yet.
00:23:14 You’ve mentioned that people would use Spotify to discover the songs they like, and then
00:23:19 they would torrent those songs to so they can copy it to their phone.
00:23:24 Just hilarious.
00:23:25 Exactly.
00:23:26 Not torrent, pirate.
00:23:27 Seriously, piracy does seem to be like a good guide for business models.
00:23:33 Video content.
00:23:34 As far as I know, Spotify doesn’t have video content.
00:23:37 Well, we do have music videos, and we do have videos on the service.
00:23:42 But the way we think about ourselves is that we’re an audio service, and we think that
00:23:48 if you look at the amount of time that people spend on audio, it’s actually very similar
00:23:52 to the amount of time that people spend on music.
00:23:55 It’s very similar to the amount of time that people spend on video.
00:23:58 So the opportunity should be equally big.
00:24:02 But today, it’s not at all valued.
00:24:03 Videos value much higher.
00:24:05 So we think it’s basically completely undervalued.
00:24:08 So we think of ourselves as an audio service.
00:24:10 But within that audio service, I think video can make a lot of sense.
00:24:14 I think when you’re discovering an artist, you probably do want to see them and understand
00:24:19 who they are, to understand their identity.
00:24:21 You won’t see that video every time.
00:24:22 90% of the time, the phone is going to be in your pocket.
00:24:25 For podcasters, you use video.
00:24:27 I think that can make a ton of sense.
00:24:28 So we do have video, but we’re an audio service where, think of it as we call it internally,
00:24:33 backgroundable video.
00:24:35 Video that is helpful, but isn’t the driver of the narrative.
00:24:39 I think also, if we look at YouTube, there’s quite a few folks who listen to music on YouTube.
00:24:48 So in some sense, YouTube is a bit of a competitor to Spotify, which is very strange to me that
00:24:55 people use YouTube to listen to music.
00:24:57 They play essentially the music videos, right?
00:25:00 But don’t watch the videos and put it in their pocket.
00:25:03 Well, I think it’s similar to what, strangely, maybe it’s similar to what we were for the
00:25:12 piracy networks, where YouTube, for historical reasons, have a lot of music videos.
00:25:20 So people use YouTube for a lot of the discovery part of the process, I think.
00:25:25 But then it’s not a really good sort of, quote unquote, MP3 player, because it doesn’t even
00:25:29 background.
00:25:29 Then you have to keep the app in the foreground.
00:25:31 So it’s not a good consumption tool, but it’s a decently good discovery.
00:25:36 I mean, I think YouTube is a fantastic product.
00:25:38 And I use it for all kinds of purposes.
00:25:40 That’s true.
00:25:41 If I were to admit something, I do use YouTube a little bit to assist in the discovery process
00:25:46 of songs.
00:25:47 And then if I like it, I’ll add it to Spotify.
00:25:50 But that’s OK.
00:25:51 That’s OK with us.
00:25:53 OK, so sorry, we’re jumping around a little bit.
00:25:55 So it’s kind of incredible.
00:25:58 You look at Napster, you look at the early days of Spotify.
00:26:03 One fascinating point is how do you grow a user base?
00:26:06 So you’re there in Sweden.
00:26:08 You have an idea.
00:26:10 I saw the initial sketches that look terrible.
00:26:14 How do you grow a user base from a few folks to millions?
00:26:19 I think there are a bunch of tactical answers.
00:26:22 So first of all, I think you need a great product.
00:26:24 I don’t think you take a bad product and market it to be successful.
00:26:30 So you need a great product.
00:26:31 But sorry to interrupt, but it’s a totally new way to listen to music, too.
00:26:34 So it’s not just did people realize immediately that Spotify is a great product?
00:26:38 No, I think they did.
00:26:40 So back to the point of piracy, it was a totally new way to listen to music legally.
00:26:45 But people had been used to the access model in Sweden
00:26:48 and the rest of the world for a long time through piracy.
00:26:50 So one way to think about Spotify, it was just legal and fast piracy.
00:26:54 And so people have been using it for a long time.
00:26:56 So they weren’t alien to it.
00:26:59 They didn’t really understand how it could be illegal
00:27:01 because it seemed too fast and too good to be true,
00:27:03 which I think is a great product proposition if you can be too good to be true.
00:27:06 But what I saw again and again was people showing each other,
00:27:09 clicking the song, showing how fast it started and say, can you believe this?
00:27:13 So I really think it was about speed.
00:27:16 Then we also had an invite program that was really meant for scaling
00:27:22 because we hosted our own service.
00:27:23 We needed to control scaling.
00:27:25 But that built a lot of expectation.
00:27:27 And I don’t want to say hype because hype implies that it wasn’t true.
00:27:32 Excitement around the product. And we’ve replicated that when we launched in the US.
00:27:38 We also built up an invite only program first.
00:27:41 There are lots of tactics, but I think you need a great product to solve some problem.
00:27:46 And basically the key innovation, there was technology,
00:27:51 but on a meta level, the innovation was really the access model versus the ownership model.
00:27:55 And that was tricky.
00:27:56 A lot of people said that they wanted to be able to do it.
00:28:01 I mean, they wanted to own their music.
00:28:04 They would never kind of rent it or borrow it.
00:28:07 But I think the fact that we had a free tier,
00:28:09 which meant that you get to keep this music for life as well, helped quite a lot.
00:28:14 So this is an interesting psychological point that maybe you can speak to.
00:28:18 It was a big shift for me.
00:28:22 It’s almost like I had to go to therapy for this.
00:28:26 I think I would describe my early listening experience,
00:28:29 and I think a lot of my friends do, as basically hoarding music.
00:28:33 As you’re like slowly, one song by one song,
00:28:35 or maybe albums, gathering a collection of music that you love.
00:28:40 And you own it.
00:28:42 It’s like often, especially with CDs or tape, you like physically had it.
00:28:46 And what Spotify, what I had to come to grips with,
00:28:50 it was kind of liberating actually, is to throw away all the music.
00:28:55 I’ve had this therapy session with lots of people.
00:28:58 And I think the mental trick is, so actually we’ve seen the user data.
00:29:02 When Spotify started, a lot of people did the exact same thing.
00:29:05 They started hoarding as if the music would disappear.
00:29:09 Almost the equivalent of downloading.
00:29:10 And so we had these playlists that had limits of like a few hundred thousand tracks.
00:29:16 We figured no one will ever.
00:29:17 Well, they do.
00:29:18 Nuts and hundreds and hundreds of thousands of tracks.
00:29:20 And to this day, some people want to actually save, quote unquote,
00:29:25 and then play the entire catalog.
00:29:26 But I think the therapy session goes something like instead of throwing away your music,
00:29:34 if you took your files and you stored them in the locker at Google,
00:29:38 it’d be a streaming service.
00:29:39 It’s just that in that locker, you have all the world’s music now for free.
00:29:42 So instead of giving away your music, you got all the music.
00:29:45 It’s yours.
00:29:46 You could think of it as having a copy of the world’s catalog there forever.
00:29:50 So you actually got more music instead of less.
00:29:52 It’s just that you just took that hard disk and you sent it to someone who stored it for you.
00:29:58 And once you go through that mental journey, I’m like, it’s still my files.
00:30:01 They’re just over there.
00:30:02 And I just have 40 million or 50 million or something now.
00:30:05 Then people are like, OK, that’s good.
00:30:07 The problem is, I think, because you paid us a subscription,
00:30:11 if we hadn’t had the free tier where you would feel like,
00:30:14 even if I don’t want to pay anymore, I still get to keep them.
00:30:17 You keep your playlist forever.
00:30:18 They don’t disappear even though you stop paying.
00:30:20 I think that was really important.
00:30:21 If we would have started as, you know, you can put in all this time,
00:30:25 but if you stop paying, you lose all your work.
00:30:27 I think that would have been a big challenge and was the big challenge for a lot of our competitors.
00:30:31 That’s another reason why I think the free tier is really important.
00:30:34 That people need to feel the security, that the work they put in,
00:30:37 it will never disappear, even if they decide not to pay.
00:30:40 I like how you put the work you put in.
00:30:42 I actually stopped even thinking of it that way.
00:30:44 I just actually Spotify taught me to just enjoy music as opposed to.
00:30:50 As opposed to what I was doing before, which is like in an unhealthy way, hoarding music.
00:30:58 Which I found that because I was doing that,
00:31:01 I was listening to a small selection of songs way too much to where I was getting sick of them.
00:31:07 Whereas Spotify, the more liberating kind of approach is I was just enjoying.
00:31:11 Of course, I listened to Stairway to Heaven over and over,
00:31:13 but because of the extra variety, I don’t get as sick of them.
00:31:18 There’s an interesting statistic I saw.
00:31:21 So Spotify has, maybe you can correct me, but over 50 million songs, tracks,
00:31:27 and over 3 billion playlists.
00:31:31 So 50 million songs and 3 billion playlists.
00:31:35 60 times more playlist songs.
00:31:38 What do you make of that?
00:31:39 Yeah.
00:31:40 So the way I think about it is that from a statistician or machine learning point of view,
00:31:48 you have all these, if you want to think about reinforcement learning,
00:31:52 you have this state space of all the tracks.
00:31:54 You can take different journeys through this world.
00:32:00 I think of these as people helping themselves and each other,
00:32:05 creating interesting vectors through this space of tracks.
00:32:08 And then it’s not so surprising that across many tens of millions of atomic units,
00:32:14 there will be billions of paths that make sense.
00:32:17 And we’re probably pretty quite far away from having found all of them.
00:32:21 So kind of our job now is users, when Spotify started,
00:32:26 it was really a search box that was for the time pretty powerful.
00:32:30 And then I’d like to refer to it as this programming language called playlisting,
00:32:34 where if you, as you probably were pretty good at music,
00:32:36 you knew your new releases, you knew your back catalog,
00:32:39 you knew your star with the heaven,
00:32:40 you could create a soundtrack for yourself using this playlisting tool,
00:32:43 this like meta programming language for music to soundtrack your life.
00:32:47 And people who were good at music, it’s back to how do you scale the product.
00:32:50 For people who are good at music, that wasn’t actually enough.
00:32:53 If you had the catalog and a good search tool,
00:32:55 and you can create your own sessions,
00:32:57 you could create really good a soundtrack for your entire life.
00:33:01 Probably perfectly personalized because you did it yourself.
00:33:04 But the problem was most people, many people aren’t that good at music.
00:33:06 They just can’t spend the time.
00:33:08 Even if you’re very good at music, it’s going to be hard to keep up.
00:33:10 So what we did to try to scale this was to essentially try to build,
00:33:16 you can think of them as agents that this friend that some people had
00:33:20 that helped them navigate this music catalog.
00:33:22 That’s what we’re trying to do for you.
00:33:24 But also there is something like 200 million active users.
00:33:32 1 million active users on Spotify.
00:33:35 So there it’s okay.
00:33:36 So from the machine learning perspective,
00:33:39 you have these 200 million people plus they’re creating.
00:33:45 It’s really interesting to think of a playlist as,
00:33:51 I mean, I don’t know if you meant it that way,
00:33:53 but it’s almost like a programming language.
00:33:54 It’s or at least a trace of exploration of those individual agents.
00:34:01 The listeners and you have all this new tracks coming in.
00:34:06 So it’s a fascinating space that is ripe for machine learning.
00:34:11 So is there, is it possible, how can playlists be used as data
00:34:18 in terms of machine learning and to help Spotify organize the music?
00:34:24 So we found in our data, not surprising that people who play listed lots
00:34:29 they retain much better.
00:34:30 They had a great experience.
00:34:32 And so our first attempt was to playlist for users.
00:34:35 And so we acquired this company called Tunigo of editors and professional playlisters
00:34:41 and kind of leveraged the maximum of human intelligence
00:34:45 to help build kind of these vectors through the track space for people.
00:34:52 And that broadened the product.
00:34:54 But then the obvious next, and we use statistical means,
00:34:57 where they could see when they created a playlist, how did that playlist perform?
00:35:02 They could see skips of the songs, they could see how the songs perform,
00:35:04 and they manually iterated the playlist to maximize performance for a large group of people.
00:35:10 But there were never enough editors to playlists for you personally.
00:35:14 So the promise of machine learning was to go from kind of group personalization
00:35:18 using editors and tools and statistics to individualization.
00:35:22 And then what’s so interesting about the 3 billion playlists we have is we ended,
00:35:28 the truth is we lucked out.
00:35:29 This was not a priority strategy, as is often the case.
00:35:32 It looks really smart in hindsight, but it was dumb luck.
00:35:37 We looked at these playlists and we had some people in the company,
00:35:42 a person named Eric Beranodson.
00:35:43 He was really good at machine learning already back then in like 2007, 2008.
00:35:48 Back then it was mostly collaborative filtering and so forth.
00:35:51 But we realized that what this is, is people are grouping tracks for themselves
00:35:57 that have some semantic meaning to them.
00:36:00 And then they actually label it with a playlist name as well.
00:36:04 So in a sense, people were grouping tracks along semantic dimensions and labeling them.
00:36:09 And so could you use that information to find that latent embedding?
00:36:15 And so we started playing around with collaborative filtering
00:36:20 and we saw tremendous success with it.
00:36:24 Basically trying to extract some of these dimensions.
00:36:28 And if you think about it, it’s not surprising at all.
00:36:30 It’d be quite surprising if playlists were actually random,
00:36:34 if they had no semantic meaning.
00:36:36 For most people, they group these tracks for some reason.
00:36:39 So we just happened across this incredible data set.
00:36:43 Where people are taking these tens of millions of tracks
00:36:46 and group them along different semantic vectors.
00:36:49 And the semantics being outside the individual users.
00:36:52 So it’s some kind of universal.
00:36:54 There’s a universal embedding that holds across people on this earth.
00:36:59 Yes, I do think that the embeddings you find are going to be reflective of the people who play listed.
00:37:05 So if you have a lot of indie lovers who play list,
00:37:09 your embedding is going to perform better there.
00:37:14 But what we found was that yes, there were these latent similarities.
00:37:20 They were very powerful.
00:37:22 And it was interesting because I think that the people who play listed the most initially
00:37:28 were the so called music aficionados who were really into music.
00:37:32 And they often had a certain…
00:37:34 Their taste was often geared towards a certain type of music.
00:37:38 And so what surprised us, if you look at the problem from the outside,
00:37:42 you might expect that the algorithms would start performing best with mainstreamers first.
00:37:47 Because it somehow feels like an easier problem to solve mainstream taste
00:37:51 than really particular taste.
00:37:53 It was the complete opposite for us.
00:37:55 The recommendations performed fantastically for people who saw themselves as
00:37:59 having very unique taste.
00:38:00 That’s probably because all of them play listed.
00:38:03 And they didn’t perform so well for mainstreamers.
00:38:05 They actually thought they were a bit too particular and unorthodox.
00:38:09 So we had the complete opposite of what we expected.
00:38:12 Success within the hardest problem first,
00:38:13 and then had to try to scale to more mainstream recommendations.
00:38:17 So you’ve also acquired Echo Nest that analyzes song data.
00:38:24 So in your view, maybe you can talk about,
00:38:28 so what kind of data is there from a machine learning perspective?
00:38:31 From a machine learning perspective, there’s a huge amount.
00:38:35 We’re talking about playlisting and just user data of what people are listening to,
00:38:40 the playlist they’re constructing, and so on.
00:38:44 And then there’s the actual data within a song.
00:38:48 What makes a song, I don’t know, the actual waveforms.
00:38:54 How do you mix the two?
00:38:55 How much value is there in each?
00:38:57 To me, it seems like user data is a romantic notion
00:39:03 that the song itself would contain useful information.
00:39:05 But if I were to guess, user data would be much more powerful,
00:39:09 like playlists would be much more powerful.
00:39:11 Yeah, so we use both.
00:39:14 Our biggest success initially was with playlist data
00:39:18 without understanding anything about the structure of the song.
00:39:22 But when we acquired Echo Nest, they had the inverse problem.
00:39:25 They actually didn’t have any play data.
00:39:27 They were just, they were a provider of recommendations,
00:39:29 but they didn’t actually have any play data.
00:39:31 So they looked at the structure of songs, sonically,
00:39:36 and they looked at Wikipedia for cultural references and so forth, right?
00:39:40 And did a lot of NLU and so forth.
00:39:41 So we got that skill into the company and combined kind of our user data
00:39:47 with their kind of content based.
00:39:51 So you can think of it as we were user based
00:39:53 and they were content based in their recommendations.
00:39:54 And we combined those two.
00:39:56 And for some cases where you have a new song that has no play data,
00:40:00 obviously you have to try to go by either who the artist is
00:40:04 or the sonic information in the song or what it’s similar to.
00:40:09 So there’s definitely a value in both and we do a lot in both,
00:40:12 but I would say, yes, the user data captures things
00:40:16 that have to do with culture in the greater society
00:40:19 that you would never see in the content itself.
00:40:23 But that said, we have seen, we have a research lab in Paris
00:40:28 when we can talk more about that on machine learning on the creator side,
00:40:32 what it can do for creators, not just for the consumers,
00:40:35 but where we looked at how does the structure of a song
00:40:38 actually affect the listening behavior?
00:40:40 And it turns out that there is a lot of,
00:40:43 we can predict things like skips based on the song itself.
00:40:48 We could say that maybe you should move that chorus a bit
00:40:50 because your skip is going to go up here.
00:40:52 There is a lot of latent structure in the music,
00:40:54 which is not surprising because it is some sort of mind hack.
00:40:58 So there should be structure. That’s probably what we respond to.
00:41:00 You just blew my mind actually from the creator perspective.
00:41:05 So that’s a really interesting topic
00:41:08 that probably most creators aren’t taking advantage of, right?
00:41:11 So I’ve recently got to interact with a few folks,
00:41:15 YouTubers who are like obsessed with this idea of what do I do
00:41:24 to make sure people keep watching the video?
00:41:27 And they like look at the analytics of which point do people turn it off and so on.
00:41:32 First of all, I don’t think that’s healthy,
00:41:35 but it’s because you can do it a little too much.
00:41:38 But it is a really powerful tool for helping the creative process.
00:41:42 You just made me realize you could do the same thing for creation of music.
00:41:47 And so is that something you’ve looked into?
00:41:51 And can you speak to how much opportunity there is for that kind of thing?
00:41:54 Yeah, so I listened to the podcast with Ziraj and I thought it was fantastic
00:41:59 and I reacted to the same thing where he said he posted something in the morning,
00:42:04 immediately watched the feedback where the drop off was
00:42:06 and then responded to that in the afternoon,
00:42:08 which is quite different from how people make podcasts, for example.
00:42:12 Yes, exactly.
00:42:12 I mean, the feedback loop is almost non existent.
00:42:15 So if we back out one level, I think actually both for music and podcasts,
00:42:21 which we also do at Spotify,
00:42:23 I think there’s a tremendous opportunity just for the creation workflow.
00:42:27 And I think it’s really interesting speaking to you who,
00:42:30 because you’re a musician, a developer, and a podcaster.
00:42:34 If you think about those three different roles,
00:42:36 if you make the leap as a musician,
00:42:38 if you think about it as a software tool chain, really,
00:42:42 your DAW with the stems, that’s the IDE, right?
00:42:46 That’s where you work in source code format with what you’re creating.
00:42:51 Then you sit around and you play with that.
00:42:52 And when you’re happy, you compile that thing into some sort of AAC or MP3 or something.
00:42:57 You do that because you get distribution.
00:42:59 There are so many runtimes for that MP3 across the world in car stairs and stuff.
00:43:02 So if you kind of compile this execution,
00:43:03 you ship it out in kind of an old fashioned boxed software analogy.
00:43:09 And then you hope for the best, right?
00:43:11 But as a software developer, you would never do that.
00:43:16 First, you go on GitHub and you collaborate with other creators.
00:43:19 And then you think it’d be crazy to just ship one version of your software
00:43:22 without doing an A B test, without any feedback loop.
00:43:26 Issue tracking.
00:43:28 Exactly.
00:43:28 And then you would look at the feedback loop and say,
00:43:31 try to optimize that thing, right?
00:43:34 So I think if you think of it as a very specific software tool chain,
00:43:38 it looks quite arcane, the tools that a music creator has
00:43:42 versus what a software developer has.
00:43:45 So that’s kind of how we think about it.
00:43:48 Why wouldn’t a music creator have something like GitHub
00:43:52 where you could collaborate much more easily?
00:43:54 So we bought this company called Soundtrap,
00:43:56 which has a kind of Google Docs for music approach, where you can collaborate
00:44:01 with other people on the kind of source code format with Stems.
00:44:05 And I think introducing things like AI tools there to help you
00:44:09 as you’re creating music, both in helping you put accompaniment to your music,
00:44:19 like drums or something, help you master and mix automatically,
00:44:24 help you understand how this track will perform.
00:44:26 Exactly what you would expect as a software developer.
00:44:29 I think it makes a lot of sense.
00:44:30 And I think the same goes for a podcaster.
00:44:33 I think podcasters will expect to have the same kind of feedback loop
00:44:36 that Siraj has, like, why wouldn’t you?
00:44:39 Maybe it’s not healthy, but…
00:44:41 Sorry, I wanted to criticize the fact because you can overdo it
00:44:45 because a lot of the, and we’re in a new era of that.
00:44:49 So you can become addicted to it and therefore, what people say,
00:44:56 you become a slave to the YouTube algorithm or sort of,
00:45:00 it’s always a danger of a new technology as opposed to say,
00:45:04 if you’re creating a song, becoming too obsessed about the intro riff to the song
00:45:11 that keeps people listening versus actually the entirety of the creation process.
00:45:15 It’s a balance.
00:45:16 But the fact that there’s zero, I mean, you’re blowing my mind right now,
00:45:19 because you’re completely right that there is no signal whatsoever.
00:45:24 There’s no feedback whatsoever on the creation process and music or podcasting,
00:45:30 almost at all.
00:45:31 And are you saying that Spotify is hoping to help create tools to, not tools, but…
00:45:39 No, tools actually.
00:45:41 Actually, tools.
00:45:42 Tools for creators.
00:45:47 Absolutely.
00:45:48 So we’ve made some acquisitions the last few years around music creation,
00:45:53 this company called Soundtrap, which is a digital audio workstation,
00:45:57 but that is browser based.
00:45:59 And their focus was really the Google Docs approach.
00:46:01 We can collaborate with people much more easily than you could in previous tools.
00:46:06 So we have some of these tools that we’re working with that we want to make accessible
00:46:09 and then we can connect it with our consumption data.
00:46:12 We can create this feedback loop where we could help you understand,
00:46:16 we could help you create and help you understand how you will perform.
00:46:20 We also acquired this other company within podcasting called Anchor,
00:46:24 which is one of the biggest podcasting tools, mobile focused.
00:46:28 So really focused on simple creation or easy access to creation.
00:46:32 But that also gives us this feedback loop.
00:46:34 And even before that, we invested in something called Spotify for Artists
00:46:40 and Spotify for Podcasters, which is an app that you can download,
00:46:43 you can verify that you are that creator.
00:46:46 And then you get things that software developers have had for years.
00:46:51 You can see where, if you look at your podcast, for example, on Spotify
00:46:55 or a song that you released, you can see how it’s performing,
00:46:58 which cities it’s performing in, who’s listening to it,
00:47:01 what’s the demographic breakup.
00:47:02 So similar in the sense that you can understand
00:47:05 how you’re actually doing on the platform.
00:47:08 So we definitely want to build tools.
00:47:10 I think you also interviewed the head of research for Adobe.
00:47:15 And I think that’s an, back to Photoshop that you like,
00:47:19 I think that’s an interesting analogy as well.
00:47:22 Photoshop, I think, has been very innovative in helping photographers and artists.
00:47:28 And I think there should be the same kind of tools for music creators,
00:47:32 where you could get AI assistance, for example, as you’re creating music,
00:47:36 as you can do with Adobe, where you can,
00:47:38 I want a sky over here and you can get help creating that sky.
00:47:42 The really fascinating thing is what Adobe doesn’t have
00:47:47 is a distribution for the content you create.
00:47:50 So you don’t have the data of if I create, if I, you know,
00:47:55 whatever creation I make in Photoshop or Premiere,
00:47:59 I can’t get like immediate feedback like I can on YouTube,
00:48:02 for example, about the way people are responding.
00:48:05 And if Spotify is creating those tools, that’s a really exciting actually world.
00:48:11 But let’s talk a little about podcasts.
00:48:16 So I have trouble talking to one person.
00:48:20 So it’s a bit terrifying and kind of hard to fathom,
00:48:23 but on average, 60 to 100,000 people will listen to this episode.
00:48:30 Okay, so it’s intimidating.
00:48:32 Yeah, it’s intimidating.
00:48:34 So I hosted on Blueberry.
00:48:36 I don’t know if I’m pronouncing that correctly, actually.
00:48:39 It looks like most people listen to it on Apple Podcasts,
00:48:42 Cast Box and Pocket Casts, and only about a thousand listen on Spotify.
00:48:48 It’s just my podcast, right?
00:48:53 So where do you see a time when Spotify will dominate this?
00:49:00 So Spotify is relatively new into this podcasting site.
00:49:06 Yeah, in podcasting.
00:49:07 What’s the deal with podcasting and Spotify?
00:49:10 How serious is Spotify about podcasting?
00:49:13 Do you see a time where everybody would listen to, you know,
00:49:16 probably a huge amount of people, majority perhaps listen to music on Spotify?
00:49:22 Do you see a time when the same is true for podcasting?
00:49:26 Well, I certainly hope so.
00:49:28 That is our mission.
00:49:29 Our mission as a company is actually to enable a million creators to live off of their art,
00:49:34 and a billion people be inspired by it.
00:49:35 And what I think is interesting about that mission is it actually puts the creators first,
00:49:40 even though it started as a consumer focused company,
00:49:43 and it’s just to be able to live off of their art,
00:49:44 not just make some money off of their art as well.
00:49:47 So it’s quite an ambitious project.
00:49:51 So we think about creators of all kinds,
00:49:53 and we kind of expanded our mission from being music to being audio a while back.
00:50:01 And that’s not so much because we think we made that decision.
00:50:08 We think that decision was made for us.
00:50:10 We think the world made that decision.
00:50:12 Whether we like it or not, when you put in your headphones,
00:50:16 you’re going to make a choice between music and a new episode of your podcast or something else.
00:50:25 We’re in that world whether we like it or not.
00:50:26 And that’s how radio works.
00:50:28 So we decided that we think it’s about audio.
00:50:32 You can see the rise of audiobooks and so forth.
00:50:34 We think audio is a great opportunity.
00:50:36 So we decided to enter it.
00:50:37 And obviously, Apple and Apple Podcasts is absolutely dominating in podcasting,
00:50:45 and we didn’t have a single podcast only like two years ago.
00:50:49 What we did though was we looked at this and said,
00:50:54 can we bring something to this?
00:50:56 We want to do this, but back to the original Spotify,
00:50:59 we have to do something that consumers actually value to be able to do this.
00:51:03 And the reason we’ve gone from not existing at all to being quite a wide margin,
00:51:09 the second largest podcast consumption, still wide gap to iTunes, but we’re growing quite fast.
00:51:16 I think it’s because when we looked at the consumer problem,
00:51:20 people said surprisingly that they wanted their podcasts and music in the same application.
00:51:26 So what we did was we took a little bit of a different approach where we said,
00:51:29 instead of building a separate podcast app,
00:51:31 we thought, is there a consumer problem to solve here?
00:51:33 Because the others are very successful already.
00:51:35 And we thought there was in making a more seamless experience
00:51:38 where you can have your podcast and your music in the same application,
00:51:43 because we think it’s audio to you.
00:51:45 And that has been successful.
00:51:46 And that meant that we actually had 200 million people to offer this to instead of starting from zero.
00:51:52 So I think we have a good chance because we’re taking a different approach than the competition.
00:51:56 And back to the other thing I mentioned about
00:51:59 creators, because we’re looking at the end to end flow.
00:52:02 I think there’s a tremendous amount of innovation to do around podcast as a format.
00:52:07 When we have creation tools and consumption, I think we could start improving what podcasting is.
00:52:12 I mean, podcast is this opaque, big, like one, two hour file that you’re streaming,
00:52:19 which it really doesn’t make that much sense in 2019 that it’s not interactive.
00:52:24 There’s no feedback loops, nothing like that.
00:52:26 So I think if we’re going to win, it’s going to have to be because we build a better product
00:52:29 for creators and for consumers.
00:52:32 So we’ll see, but it’s certainly our goal.
00:52:34 We have a long way to go.
00:52:36 Well, the creators part is really exciting.
00:52:38 You already, you got me hooked there.
00:52:40 Cause the only stats I have,
00:52:42 Blueberry just recently added the stats of whether it’s listened to the end or not.
00:52:48 And that’s like a huge improvement, but that’s still
00:52:52 nowhere to where you could possibly go in terms of statistics.
00:52:54 You just download the Spotify podcasters up and verify.
00:52:57 And then, then you’ll know where people dropped out in this episode.
00:52:59 Oh, wow.
00:53:00 Okay.
00:53:01 The moment I started talking.
00:53:02 Okay.
00:53:03 I might be depressed by this, but okay.
00:53:06 So one, um, one other question is the original Spotify for music.
00:53:14 And I have a question about podcasting in this line is the idea of podcasting
00:53:19 about podcasting in this line is the idea of albums.
00:53:23 I have, uh, what did you, uh, music aficionados, uh, friends who are really,
00:53:29 uh, big fans of music often, uh, really enjoy albums,
00:53:33 listening to entire albums of, of an artist.
00:53:36 Correct me if I’m wrong, but I feel like Spotify has helped
00:53:40 replace the idea of an album with playlists.
00:53:44 So you create your own albums.
00:53:46 It’s, it’s kind of the way, at least I’ve experienced music
00:53:48 and I’ve really enjoyed it that way.
00:53:51 One of the things that was missing in podcasting for me,
00:53:54 I don’t know if it’s missing.
00:53:56 I don’t know.
00:53:56 It’s an open question for me, but the way I listened to podcasts is
00:53:59 the way I would listen to albums.
00:54:02 So I take a Joe Rogan experience and that’s an album.
00:54:05 And I listened, you know, I like, I, I put that on and I listened one
00:54:09 episode after the next, then there’s a sequence and so on.
00:54:12 Is there a room for doing what you did for music or doing what
00:54:17 Spotify did for music, but, uh, creating playlists, sort of, uh,
00:54:22 this kind of playlisting idea of breaking apart from podcasting,
00:54:27 uh, from individual podcasts and creating kind of, uh, this interplay
00:54:31 or, or have you thought about that space?
00:54:33 Uh, it’s a great question.
00:54:34 So I think in, um, in music, you’re right.
00:54:38 Basically you bought an album.
00:54:39 So it was like, you bought a small catalog of like 10 tracks, right?
00:54:42 It was, it was, again, it was actually a lot of, a lot of consumption.
00:54:46 You think it’s about what you like, but it’s based on the business model.
00:54:49 So you paid for this 10 track service and then you listened to that for a while.
00:54:54 And then when, when everything was flat priced, you tended to listen differently.
00:54:58 Now, so, so I think the, I think the album is still tremendously important.
00:55:01 That’s why we have it and you can save albums and so forth.
00:55:03 And you have a huge amount of people who really listen according to albums.
00:55:06 And I like that because it is a creator format, you can tell a longer story
00:55:10 over several tracks.
00:55:12 And so some people listen to just one track.
00:55:13 Some people actually want to hear that whole story.
00:55:17 Now in podcast, I think, I think it’s different.
00:55:21 You can argue that podcasts might be more like shows on Netflix.
00:55:25 Have like a full season of Narcos and you’re probably not going to do like
00:55:29 one episode of Narcos and then one of House of Cards, like, like, you know,
00:55:33 there’s a narrative there.
00:55:34 And you, you, you love the cast and you love these characters.
00:55:37 So I think people will, people love shows.
00:55:42 And I think they will, they will listen to those shows.
00:55:44 I do think you follow a bunch of shows at the same time.
00:55:46 So there’s certainly an opportunity to bring you the latest episode of, you
00:55:50 know, whatever the five, six, 10 things that, that you’re into.
00:55:54 But, but I think, I think people are going to listen to specific hosts and love
00:56:00 those hosts for a long time.
00:56:01 Because I think there’s something different with podcasts where, um, this
00:56:06 format of the, the, the, the, the, the experience of the, of the audience is
00:56:11 actually sitting here right between us.
00:56:13 Whereas if you look at something on TV, the audio actually would come from, you
00:56:16 would sit over there and the audio would come to you from both of us as if you
00:56:20 were watching, not as you were part of the conversation.
00:56:22 So my experience is having listened to podcasts like yours and Joe Rogan is, I
00:56:27 feel like I know all of these people.
00:56:28 They, they have a lot of experience.
00:56:30 I know all of these people, they have no idea who I am, but I feel like I’ve
00:56:33 listened to so many hours of that.
00:56:35 It’s very different from me watching a, watching like a TV show or an interview.
00:56:39 So I think you, you kind of, um, fall in love with people and, um, experience
00:56:44 in a, in a different way.
00:56:45 So I think, I think shows and hosts are going to be very, uh, very important.
00:56:49 I don’t think that’s going to go away into some sort of thing where, where you
00:56:52 don’t even know who you’re listening to.
00:56:53 I don’t think that’s going to happen.
00:56:55 What I do think is I think there’s a tremendous discovery opportunity in
00:56:59 podcast because the catalog is growing quite quickly.
00:57:03 And I think podcast is only a few, like five, 600,000 shows right now.
00:57:11 If you look back to YouTube as another analogy of creators, no one really knows
00:57:16 if you would lift the lid on YouTube, but it’s probably billions of episodes.
00:57:21 And so I think the podcast catalog would probably grow tremendously because the
00:57:24 creation tools are getting easier.
00:57:27 And then you’re going to have this discovery opportunity that I think is
00:57:30 really big.
00:57:31 So, so a lot of people tell me that they love their shows, but discovering
00:57:35 podcasts kind of suck.
00:57:36 It’s really hard to get into new show.
00:57:38 They’re usually quite long.
00:57:39 It’s a big time investment.
00:57:40 So I think there’s plenty of opportunity in the discovery part.
00:57:45 Yeah, for sure.
00:57:46 A hundred percent in, in even the dumbest, there’s so many low hanging fruit too.
00:57:51 Uh, for example, just knowing what episode to listen to first to try out a podcast.
00:57:59 Exactly.
00:58:00 Uh, because most podcasts don’t have an order to them.
00:58:03 Uh, they, they can be listened to out of order and sorry to say some are better
00:58:10 than others episodes.
00:58:12 So some episodes of Joe Rogan are better than others.
00:58:15 And it’s nice to know, uh, which you should listen to, to try it out.
00:58:20 And there’s, uh, as far as I know, almost no information, uh, in terms of like, uh,
00:58:26 upvotes on how good an episode is.
00:58:28 Exactly.
00:58:29 So I think part of the problem is, uh, you, it’s kind of like music.
00:58:33 There isn’t one answer.
00:58:34 People use music for different things and there’s actually many different types of music.
00:58:37 There’s workout music and there’s classical piano music and focus music and,
00:58:41 and, and, uh, so forth.
00:58:42 I think the same with podcasts.
00:58:44 Some podcasts are sequential.
00:58:45 They’re supposed to be listened to in, in order.
00:58:48 It’s actually, it’s actually telling a narrative.
00:58:51 Some podcasts are one topic, uh, kind of like yours, but different guests.
00:58:55 So you could jump in anywhere.
00:58:57 Some podcasts actually have completely different topics.
00:58:59 And for those podcasts, it might be that I want, you know, we should recommend one episode
00:59:04 because it’s about AI from someone, but then they talk about something that you’re not
00:59:09 interested in the rest of the episodes.
00:59:10 So I think our, what we’re spending a lot of time on now is just first understanding
00:59:15 the domain and creating kind of the knowledge graph of how do these objects relate and how
00:59:21 do people consume.
00:59:22 And I think we’ll find that it’s going to be, it’s going to be different.
00:59:26 I’m excited because you’re the, uh, Spotify is the first people I’m aware of that are
00:59:32 trying to do this for podcasting.
00:59:34 Podcasting has been like a wild west up until now.
00:59:38 It’s been a very, we want to be very careful though, because it’s been a very good wild
00:59:43 west, I think it’s this fragile ecosystem.
00:59:46 And I, we want to make sure that you don’t barge in and say like, Oh, we’re going to
00:59:52 internetize this thing.
00:59:53 And you have to think about the creators.
00:59:56 You have to understand how they get distribution today, who listens to how they make money
01:00:01 today, try to, you know, make sure that their business model works, that they understand.
01:00:06 I think it’s back to doing something to improving their products, like feedback loops and
01:00:10 distribution.
01:00:11 So jumping back into terms of this fascinating world of a recommender system and listening
01:00:17 to music and using machine learning to analyze things, do you think it’s better to what
01:00:24 currently, correct me if I’m wrong, but currently Spotify lets people pick what they listen
01:00:30 to the most part.
01:00:31 There’s a discovery process, but you kind of organize playlists.
01:00:35 Is it better to let people pick what they listen to or recommend what they should listen
01:00:39 to something like stations by Spotify that I saw that you’re playing around with?
01:00:44 Maybe you can tell me what’s the status of that.
01:00:47 This is a Pandora style app that just kind of, as opposed to you select the music you
01:00:52 listen to, it kind of feeds you the music you listen to.
01:00:58 What’s the status of stations by Spotify?
01:01:00 What’s its future?
01:01:01 The story of Spotify, as we have grown, has been that we made it more accessible to different
01:01:07 audiences and stations is another one of those where the question is, some people want to
01:01:14 be very specific.
01:01:14 They actually want to hear Starway to Heaven right now, that needs to be very easy to do.
01:01:19 And some people, or even the same person, at some point might say, I want to feel upbeat
01:01:26 or I want to feel happy or I want songs to sing in the car.
01:01:32 So they put in the information at a very different level and then we need to translate that into
01:01:38 what that means musically.
01:01:40 So stations is a test to create like a consumption input vector that is much simpler where you
01:01:45 can just tune it a little bit and see if that increases the overall reach.
01:01:49 But we’re trying to kind of serve the entire gamut of super advanced so called music aficionados
01:01:56 all the way to people who they love listening to music but it’s not their number one priority
01:02:02 in life.
01:02:03 They’re not going to sit and follow every new release from every new artist.
01:02:06 They need to be able to influence music at a different level.
01:02:11 So you can think of it as different products and I think one of the interesting things
01:02:17 to answer your question on if it’s better to let the user choose or to play, I think
01:02:22 the answer is the challenge when machine learning kind of came along, there was a lot of thinking
01:02:28 about what does product development mean in a machine learning context.
01:02:33 People like Andrew Ng, for example, when he went to Baidu, he started doing a lot of practical
01:02:38 machine learning, went from academia and he thought a lot about this and he had this notion
01:02:43 that a product manager, designer and engineer, they used to work around this wireframe to
01:02:47 kind of describe what the product should look like.
01:02:49 It was something to talk about when you’re doing a chatbot or a playlist, what are you
01:02:54 going to say?
01:02:54 It should be good.
01:02:55 That’s not a good product description.
01:02:57 So how do you do that?
01:02:58 And he came up with this notion that the test set is the new wireframe.
01:03:03 The job of the product manager is to source a good test set that is representative of
01:03:06 what, like if you say I want to play this, that is songs to sing in the car.
01:03:11 The job of the product manager is to go and source a good test set of what that means.
01:03:15 So then you can work with engineering to have algorithms to try to produce that.
01:03:20 So we try to think a lot about how to structure product development for a machine learning
01:03:25 age.
01:03:26 And what we discovered was that a lot of it is actually in the expectation.
01:03:30 And you can go two ways.
01:03:33 So let’s say that if you set the expectation with the user that this is a discovery product,
01:03:40 like Discover Weekly, you’re actually setting the expectation that most of what we show
01:03:45 you will not be relevant.
01:03:46 When you’re in the discovery process, you’re going to accept that actually if you find
01:03:50 one gem every Monday that you totally love, you’re probably going to be happy.
01:03:55 Even though the statistical meaning, one out of 10 is terrible or one out of 20 is terrible
01:04:00 from a user point of view because the setting was discovery is fine.
01:04:03 Sorry to interrupt real quick.
01:04:05 I just actually learned about Discover Weekly, which is a Spotify, I don’t know, it’s a
01:04:11 feature of Spotify that shows you cool songs to listen to.
01:04:16 Maybe I can do issue tracking.
01:04:18 I couldn’t find it on my Spotify app.
01:04:20 It’s in your library.
01:04:21 It’s in the library.
01:04:22 It’s in the list of library.
01:04:23 Because I was like, whoa, this is cool.
01:04:25 I didn’t know this existed.
01:04:26 And I tried to find it.
01:04:27 But okay.
01:04:28 I will show it to you and feedback to our product team.
01:04:31 There you go.
01:04:32 But yeah, so yeah, sorry.
01:04:34 Just to mention the expectation there is basically that you’re going to discover new songs.
01:04:42 Yeah.
01:04:42 So then you can be quite adventurous in the recommendations you do.
01:04:47 But we have another product called Daily Mix, which kind of implies that these are only
01:04:53 going to be your favorites.
01:04:54 So if you have one out of 10 that is good and nine out of 10 that doesn’t work for you,
01:04:58 you’re going to think it’s a horrible product.
01:04:59 So actually a lot of the product development we learned over the years is about setting
01:05:03 the right expectations.
01:05:04 So for Daily Mix, you know, algorithmically, we would pick among things that feel very
01:05:09 safe in your taste space.
01:05:11 Whereas Discover Weekly, we go kind of wild because the expectation is most of this is
01:05:15 not going to.
01:05:16 So a lot of that, a lot of to answer your question there, a lot of should you let the
01:05:20 user pick or not?
01:05:21 It depends.
01:05:23 We have some products where the whole point is that the user can click play, put the phone
01:05:26 in the pocket, and it should be really good music for like an hour.
01:05:30 We have other products where you probably need to say like, no, no, save, no, no.
01:05:35 And it’s very interactive.
01:05:37 I see.
01:05:37 That makes sense.
01:05:38 And then the radio product, the stations product is one of these like click play, put in your
01:05:41 pocket for hours.
01:05:43 That’s really interesting.
01:05:44 So you’re thinking of different test sets for different users and trying to create products
01:05:50 that sort of optimize for those test sets that represent a specific set of users.
01:05:57 Yes, I think one thing that I think is interesting is we invested quite heavily in editorial
01:06:06 in people creating playlists using statistical data.
01:06:09 And that was successful for us.
01:06:10 And then we also invested in machine learning.
01:06:13 And for the longest time within Spotify and within the rest of the industry, there was
01:06:18 always this narrative of humans versus the machine, algo versus editorial.
01:06:23 And editors would say like, well, if I had that data, if I could see your
01:06:27 playlisting history and I made a choice for you, I would have made a better choice.
01:06:31 And they would have because they’re much smarter than these algorithms.
01:06:35 The human is incredibly smart compared to our algorithms.
01:06:38 They can take culture into account and so forth.
01:06:41 The problem is that they can’t make 200 million decisions per hour for every user that logs
01:06:47 in.
01:06:47 So the algo may be not as sophisticated, but much more efficient.
01:06:51 So there was this contradiction.
01:06:54 But then a few years ago, we started focusing on this kind of human in the loop thinking
01:07:00 around machine learning.
01:07:01 And we actually coined an internal term for it called algotorial, a combination of algorithms
01:07:07 and editors, where if we take a concrete example, you think of the editor, this paid
01:07:15 expert that we have that’s really good at something like soul, hip hop, EDM, something,
01:07:20 right?
01:07:20 They’re a true expert, no one in the industry.
01:07:22 So they have all the cultural knowledge.
01:07:24 You think of them as the product manager.
01:07:26 And you say that, let’s say that you want to create a, you think that there’s a product
01:07:32 need in the world for something like songs to sing in the car or songs to sing in the
01:07:36 shower.
01:07:36 I’m taking that example because it exists.
01:07:38 People love to scream songs in the car when they drive, right?
01:07:42 So you want to create that product and you have this product manager who’s a musical
01:07:45 expert.
01:07:46 They create, they come up with a concept, like I think this is a missing thing in humanity,
01:07:50 like a playlist called songs to sing in the car.
01:07:53 They create the framing, the image, the title, and they create a test set of, they create
01:07:59 a group of songs, like a few thousand songs out of the catalog that they manually curate
01:08:04 that are known songs that are great to sing in the car.
01:08:07 And they can take like true romance into account.
01:08:09 They understand things that our algorithms do not at all.
01:08:12 So they have this huge set of tracks.
01:08:14 Then when we deliver that to you, we look at your taste vectors and you get the 20 tracks
01:08:19 that are songs to sing in the car in your taste.
01:08:22 So you have personalization and editorial input in the same process, if that makes sense.
01:08:29 Yeah, it makes total sense.
01:08:30 And I have several questions around that.
01:08:32 This is like fascinating.
01:08:36 Okay.
01:08:36 So first, it is a little bit surprising to me that the world expert humans are outperforming
01:08:44 machines at specifying songs to sing in the car.
01:08:50 So maybe you could talk to that a little bit.
01:08:53 I don’t know if you can put it into words, but what is it?
01:08:57 How difficult is this problem?
01:09:01 Do you really, I guess what I’m trying to ask is there, how difficult is it to encode
01:09:06 the cultural references, the context of the song, the artists, all those things together?
01:09:14 Can machine learning really not do that?
01:09:17 I mean, I think machine learning is great at replicating patterns if you have the patterns.
01:09:23 But if you try to write with me a spec of what song’s greatest song to sing in the car
01:09:27 definition is, is it loud?
01:09:30 Does it have many choruses?
01:09:31 Should it have been in movies?
01:09:32 It quickly gets incredibly complicated, right?
01:09:35 Yeah.
01:09:36 And a lot of it may not be in the structure of the song or the title.
01:09:40 It could be cultural references because, you know, it was a history.
01:09:44 So the definition problems quickly get, and I think that was the insight of Andrew Ng
01:09:51 when he said the job of the product manager is to understand these things that algorithms
01:09:55 don’t and then define what that looks like.
01:09:58 And then you have something to train towards, right?
01:10:00 Then you have kind of the test set.
01:10:02 And then so today the editors create this pool of tracks and then we personalize.
01:10:06 You could easily imagine that once you have this set, you could have some automatic exploration
01:10:11 on the rest of the catalog because then you understand what it is.
01:10:14 And then the other side of it, when machine learning does help is this taste vector.
01:10:20 How hard is it to construct a vector that represents the things an individual human
01:10:26 likes, this human preference?
01:10:30 So you can, you know, music isn’t like, it’s not like Amazon, like things you usually buy.
01:10:38 Music seems more amorphous.
01:10:39 Like it’s this thing that’s hard to specify.
01:10:42 Like what is, you know, if you look at my playlist, what is the music that I love?
01:10:48 It’s harder.
01:10:49 It seems to be much more difficult to specify concretely.
01:10:54 So how hard is it to build a taste vector?
01:10:57 It is very hard in the sense that you need a lot of data.
01:11:00 And I think what we found was that, so it’s not a stationary problem.
01:11:06 It changes over time.
01:11:08 And so we’ve gone through the journey of, if you’ve done a lot of computer vision,
01:11:15 obviously I’ve done a bunch of computer vision in my past.
01:11:18 And we started kind of with the handcrafted heuristics for, you know, this is kind of
01:11:24 indie music.
01:11:24 This is this.
01:11:25 And if you consume this, you’d probably like this.
01:11:27 So we have, we started there and we have some of that still.
01:11:31 Then what was interesting about the playlist data was that you could find these latent
01:11:34 things that wouldn’t necessarily even make sense to you.
01:11:38 That could even capture maybe cultural references because they cooccurred.
01:11:42 Things that wouldn’t have appeared kind of mechanistically either in the content or so
01:11:48 forth.
01:11:48 So I think that, I think the core assumption is that there are patterns in almost
01:12:01 everything.
01:12:02 And if there are patterns, these embedding techniques are getting better and better now.
01:12:06 Now, as everyone else, we’re also using kind of deep embeddings where you can encode
01:12:12 binary values and so forth.
01:12:14 And what I think is interesting is this process to try to find things that do not
01:12:21 necessarily, you wouldn’t actually have guessed.
01:12:23 So it is very hard in an engineering sense to find the right dimensions.
01:12:28 It’s an incredible scalability problem to do for hundreds of millions of users and to
01:12:33 update it every day.
01:12:35 But in theory, in theory embeddings isn’t that complicated.
01:12:42 The fact that you try to find some principal components or something like that, dimensionality
01:12:46 reduction and so forth.
01:12:47 So the theory, I guess, is easy.
01:12:48 The practice is very, very hard.
01:12:50 And it’s a huge engineering challenge.
01:12:53 But fortunately, we have some amazing both research and engineering teams in this space.
01:12:58 Yeah, I guess the question is all, I mean, it’s similar.
01:13:03 I deal with it with autonomous vehicle spaces.
01:13:05 The question is how hard is driving?
01:13:07 And here is basically the question is of edge cases.
01:13:14 So embedding probably works, not probably, but I would imagine works well in a lot of
01:13:22 cases.
01:13:24 So there’s a bunch of questions that arise then.
01:13:25 So do song preferences, does your taste vector depend on context, like mood, right?
01:13:33 So there’s different moods, and so how does that take in it?
01:13:41 Is it possible to take that as a consideration?
01:13:44 Or do you just leave that as a interface problem that allows the user to just control it?
01:13:49 So when I’m looking for workout music, I kind of specify it by choosing certain playlists,
01:13:55 doing certain search.
01:13:56 Yeah, so that’s a great point.
01:13:58 Back to the product development.
01:14:00 You could try to spend a few years trying to predict which mood you’re in automatically
01:14:04 when you open Spotify, or you create a tab which is happy and sad, right?
01:14:08 And you’re going to be right 100% of the time with one click.
01:14:10 Now, it’s probably much better to let the user tell you if they’re happy or sad, or
01:14:14 if they want to work out.
01:14:15 On the other hand, if your user interface becomes 2,000 tabs, you’re introducing so
01:14:20 much friction so no one will use the product.
01:14:22 So then you have to get better.
01:14:24 So it’s this thing where you have to be able to get better.
01:14:26 So then you have to get better, so it’s this thing where I think maybe it was, I don’t
01:14:32 remember who coined it, but it’s called fault tolerant UIs, right?
01:14:35 You build a UI that is tolerant of being wrong, and then you can be much less right in your
01:14:42 algorithms.
01:14:43 So we’ve had to learn a lot of that.
01:14:45 Building the right UI that fits where the machine learning is, and a great discovery
01:14:52 there, which was by the teams during one of our hack days, was this thing of taking discovery,
01:14:58 packaging it into a playlist, and saying that these are new tracks that we think you might
01:15:04 like based on this.
01:15:05 And setting the right expectation made it a great product.
01:15:09 So I think we have this benefit that, for example, Tesla doesn’t have that we can change
01:15:15 the expectation.
01:15:16 We can build a fault tolerant setting.
01:15:18 It’s very hard to be fault tolerant when you’re driving at 100 miles per hour or something.
01:15:23 And we have the luxury of being able to say that of being wrong if we have the right UI,
01:15:30 which gives us different abilities to take more risk.
01:15:33 So I actually think the self driving problem is much harder.
01:15:37 Oh, yeah, for sure.
01:15:39 It’s much less fun because people die.
01:15:44 Exactly.
01:15:45 And in Spotify, it’s such a more fun problem because failure is beautiful in a way.
01:15:55 It leads to exploration.
01:15:56 So it’s a really fun reinforcement learning problem.
01:15:58 The worst case scenario is you get these WTF tweets like, how did I get this?
01:16:02 This song, yeah.
01:16:03 Which is a lot better than the self driving.
01:16:05 Exactly, so what’s the feedback that a user, what’s the signal that a user provides into
01:16:14 the system?
01:16:15 So you mentioned skipping.
01:16:19 What is like the strongest signal?
01:16:22 You didn’t mention clicking like.
01:16:24 So we have a few signals that are important.
01:16:27 Obviously playing, playing through.
01:16:30 So one of the benefits of music, actually, even compared to podcasts or movies is the
01:16:36 object itself is really only about three minutes.
01:16:39 So you get a lot of chances to recommend and the feedback loop is every three minutes instead
01:16:44 of every two hours or something.
01:16:45 So you actually get kind of noisy, but quite fast feedback.
01:16:50 And so you can see if people play through, which is the inverse of skip really.
01:16:55 That’s an important signal.
01:16:56 On the other hand, much of the consumption happens when your phone is in your pocket.
01:17:00 Maybe you’re running or driving or you’re playing on a speaker.
01:17:03 And so you not skipping doesn’t mean that you love that song.
01:17:05 It may be that it wasn’t bad enough that you would walk up and skip.
01:17:08 So it’s a noisy signal.
01:17:10 Then we have the equivalent of the like, which is you saved it to your library.
01:17:14 That’s a pretty strong signal of affection.
01:17:16 And then we have the more explicit signal of playlisting.
01:17:21 Like you took the time to create a playlist, you put it in there.
01:17:23 There’s a very little small chance that if you took all that trouble, this is not a really
01:17:28 important track to you.
01:17:30 And then we understand also what are the tracks it relates to.
01:17:34 So we have the playlisting, we have the like, and then we have the listening or skip.
01:17:39 And you have to have very different approaches to all of them because of different levels
01:17:43 of noise.
01:17:44 One is very voluminous, but noisy, and the other is rare, but you can probably trust it.
01:17:49 Yeah, it’s interesting because I think between those signals captures all the information
01:17:55 you’d want to capture.
01:17:57 I mean, there’s a feeling, a shallow feeling for me that there’s sometimes that I’ll hear
01:18:01 a song that’s like, yes, this is, you know, this was the right song for the moment.
01:18:05 But there’s really no way to express that fact except by listening through it all the
01:18:10 way and maybe playing it again at that time or something.
01:18:14 But there’s no need for a button that says this was the best song I could have heard
01:18:19 at this moment.
01:18:20 Well, we’re playing around with that, with kind of the thumbs up concept saying like,
01:18:24 I really like this.
01:18:25 Just kind of talking to the algorithm.
01:18:27 It’s unclear if that’s the best way for humans to interact.
01:18:30 Maybe it is.
01:18:31 Maybe they should think of Spotify as a person, an agent sitting there trying to serve you
01:18:35 and you can say like, bad Spotify, good Spotify.
01:18:38 Right now, the analogy we’ve had is more, you shouldn’t think of us.
01:18:42 We should be invisible.
01:18:44 And the feedback is if you save it, it’s kind of you work for yourself.
01:18:48 You do a playlist because you think it’s great and we can learn from that.
01:18:50 It’s kind of back to Tesla, how they kind of have this shadow mode.
01:18:55 They sit in what you drive.
01:18:56 We kind of took the same analogy.
01:18:58 We sit in what you playlist and then maybe we can offer you an autopilot where you can
01:19:02 take over for a while or something like that.
01:19:04 And then back off if you say like, that’s not good enough.
01:19:08 But I think it’s interesting to figure out what your mental model is.
01:19:11 If Spotify is an AI that you talk to, which I think might be a bit too abstract for many
01:19:18 consumers, or if you still think of it as it’s my music app, but it’s just more helpful.
01:19:24 And it depends on the device it’s running on, which brings us to smart speakers.
01:19:31 So I have a lot of the Spotify listening I do is on devices I can talk to, whether it’s
01:19:38 from Amazon, Google or Apple.
01:19:39 What’s the role of Spotify on those devices?
01:19:42 How do you think of it differently than on the phone or on the desktop?
01:19:47 There are a few things to say about the first of all, it’s incredibly exciting.
01:19:52 They’re growing like crazy, especially here in the US.
01:19:58 And it’s solving a consumer need that I think is, you can think of it as just remote interactivity.
01:20:09 You can control this thing from across the room.
01:20:11 And it may feel like a small thing, but it turns out that friction matters to consumers
01:20:16 being able to say play, pause and so forth from across the room is very powerful.
01:20:22 So basically, you made the living room interactive now.
01:20:26 And what we see in our data is that the number one use case for these speakers is music,
01:20:33 music and podcast.
01:20:34 So fortunately for us, it’s been important to these companies to have those use case
01:20:39 covered.
01:20:40 So they want to Spotify on this.
01:20:42 We have very good relationships with them.
01:20:45 And we’re seeing tremendous success with them.
01:20:51 What I think is interesting about them is it’s already working.
01:20:57 We kind of had this epiphany many years ago, back when we started using Sonos.
01:21:02 If you went through all the trouble of setting up your Sonos system, you had this magical
01:21:06 experience where you had all the music ever made in your living room.
01:21:10 And we made this assumption that the home, everyone used to have a CD player at home,
01:21:16 but they never managed to get their files working in the home.
01:21:19 Having this network attached storage was too cumbersome for most consumers.
01:21:22 So we made the assumption that the home would skip from the CD all the way to streaming
01:21:26 books, where you would buy the steering and would have all the music built in.
01:21:31 That took longer than we thought.
01:21:32 But with the voice speakers, that was the unlocking that made kind of the connected
01:21:36 speaker happen in the home.
01:21:39 So it really exploded.
01:21:41 And we saw this engagement that we predicted would happen.
01:21:45 What I think is interesting, though, is where it’s going from now.
01:21:49 Right now, you think of them as voice speakers.
01:21:51 But I think if you look at Google I.O., for example, they just added a camera to it, where
01:21:58 when the alarm goes off, instead of saying, hey, Google, stop, you can just wave your
01:22:04 hand.
01:22:05 So I think they’re going to think more of it as an agent or as an assistant, truly an
01:22:11 assistant.
01:22:12 And an assistant that can see you is going to be much more effective than a blind assistant.
01:22:17 So I think these things will morph.
01:22:18 And we won’t necessarily think of them as, quote unquote, voice speakers anymore.
01:22:22 Just as interactive access to the Internet in the home.
01:22:29 But I still think that the biggest use case for those will be audio.
01:22:34 So for that reason, we’re investing heavily in it.
01:22:36 And we built our own NLU stack to be able to the challenge here is, how do you innovate
01:22:43 in that world?
01:22:44 It lowers friction for consumers, but it’s also much more constrained.
01:22:48 You have no pixels to play with in an audio only world.
01:22:51 It’s really the vocabulary that is the interface.
01:22:54 So we started investing and playing around quite a lot with that, trying to understand
01:22:58 what the future will be of you speaking and gesturing and waving at your music.
01:23:03 And actually, you’re actually nudging closer to the autonomous vehicle space because from
01:23:08 everything I’ve seen, the level of frustration people experience upon failure of natural
01:23:14 language understanding is much higher than failure in other contexts.
01:23:18 People get frustrated really fast.
01:23:20 So if you screw that experience up even just a little bit, they give up really quickly.
01:23:25 Yeah.
01:23:26 And I think you see that in the data.
01:23:28 While it’s tremendously successful, the most common interactions are play, pause and next.
01:23:36 The things where if you compare it to taking up your phone, unlocking it, bringing up the
01:23:39 app and skipping, clicking skip, it was much lower friction.
01:23:44 But then for longer, more complicated things like, can you find me that song about the
01:23:49 people still bring up the phone and search and then play it on their speaker?
01:23:51 So we tried again to build a fault tolerant UI where for the more complicated things,
01:23:56 you can still pick up your phone, have powerful full keyboard search and then try to optimize
01:24:02 for where there is actually lower friction and try to it’s kind of like the test autopilot
01:24:07 thing.
01:24:07 You have to be at the level where you’re helpful.
01:24:11 If you’re too smart and just in the way, people are going to get frustrated.
01:24:15 And first of all, I’m not obsessed with stairway to heaven.
01:24:18 It’s just a good song.
01:24:19 But let me mention that as a use case because it’s an interesting one.
01:24:22 I’ve literally told one of I don’t want to say the name of the speaker because when people
01:24:28 are listening to it, it’ll make their speaker go off.
01:24:30 But I talked to the speaker and I say play stairway to heaven.
01:24:34 And every time it like not every time, but a large percentage of the time plays the wrong
01:24:40 stairway to heaven.
01:24:41 It plays like some cover of the and that part of the experience.
01:24:48 I actually wonder from a business perspective, does Spotify control that entire experience
01:24:55 or no?
01:24:56 It seems like the NLU, the natural language stuff is controlled by the speaker and then
01:25:01 Spotify stays at a layer below that.
01:25:04 It’s a good and complicated question.
01:25:07 Some of which is dependent on the on the partners.
01:25:11 So it’s hard to comment on the on the specifics.
01:25:13 But the question is the right one.
01:25:15 The challenge is if you can’t use any of the personalization, I mean, we know which stairway
01:25:21 to heaven.
01:25:21 And the truth is maybe for for one person, it is exactly the cover that they want.
01:25:26 And they would be very frustrated if a place I think we I think we default to the right
01:25:31 version.
01:25:31 But but you actually want to be able to do the cover for the person that just played
01:25:35 the cover 50 times.
01:25:36 Or Spotify is just going to seem stupid.
01:25:38 So you want to be able to leverage the personalization.
01:25:40 But you have this stack where you have the the ASR and this thing called the end best
01:25:46 list of the best guesses here.
01:25:48 And then the position comes in at the end.
01:25:50 You actually want the person to be here when you’re guessing about what they actually
01:25:53 meant.
01:25:54 So we’re working with these partners and it’s a complicated it’s a complicated thing
01:26:00 where you want to you want to be able.
01:26:02 So first of all, you want to be very careful with your users data.
01:26:06 You don’t want to share your users data without the permission.
01:26:09 But you want to share some data so that their experience gets better.
01:26:12 So that these partners can understand enough, but not too much and so forth.
01:26:16 So it’s really the trick is that it’s like a business driven relationship where you’re
01:26:21 doing product development across companies together, which is which is really complicated.
01:26:26 But this is exactly why we built our own NLU so that we actually can make personalized
01:26:32 guesses, because this is the biggest frustration from a user point of view.
01:26:36 They don’t understand about ASR and best list and and business deals.
01:26:40 They’re like, how hard can it be?
01:26:41 I was told this thing 50 times this version and still the place the wrong thing.
01:26:45 It can’t it can’t be hard.
01:26:47 So we try to take the user approach.
01:26:48 If the user the user is not going to understand the complications of business, we have to
01:26:53 solve it.
01:26:53 So let’s talk about sort of a complicated subject that I myself I’m quite torn about
01:27:02 the idea sort of of paying artists.
01:27:08 Right.
01:27:09 I saw as of August 31st, 2018, over 11 billion dollars were paid to rights holders.
01:27:17 So and further distributed to artists from Spotify.
01:27:21 So a lot of money is being paid to artists.
01:27:23 First of all, the whole time as a consumer for me, when I look at Spotify, I’m not sure
01:27:30 I’m remembering correctly, but I think you said exactly how I feel, which is this is
01:27:34 too good to be true.
01:27:36 Like when I start using Spotify, I assume you guys will go bankrupt in like a month.
01:27:43 It’s like this is too good.
01:27:44 A lot of people did.
01:27:47 I was like, this is amazing.
01:27:48 So one question I have is sort of the bigger question.
01:27:53 How do you make money in this complicated world?
01:27:55 How do you deal with the relationship with record labels who are complicated?
01:28:04 These big you’re essentially have the task of herding cats, but like rich and powerful
01:28:14 powerful cats, and also have the task of paying artists enough and paying those labels enough
01:28:21 and still making money in the Internet space where people are not willing to pay hundreds
01:28:26 of dollars a month.
01:28:27 So how do you navigate the space?
01:28:30 How do you navigate?
01:28:31 That’s a beautiful description.
01:28:32 Herding rich cats.
01:28:34 That before.
01:28:37 It is very complicated, and I think certainly actually betting against Spotify has been
01:28:42 statistically a very smart thing to do.
01:28:45 Just looking at the at the line of roadkill in music streaming services, it’s it’s kind
01:28:52 of I think if I understood the complexity when I joined Spotify, unfortunately, fortunately,
01:28:58 I didn’t know enough about the music industry to understand the complexities, because then
01:29:03 I would have made a more rational guess that it wouldn’t work.
01:29:06 So, you know, ignorance is bliss.
01:29:08 But I think there have been a few distinct challenges.
01:29:13 I think, as I said, one of the things that made it work at all was that Sweden and the
01:29:17 Nordics was a lost market.
01:29:19 So there was no risk for labels to try this.
01:29:25 I don’t think it would have worked if if the market was healthy.
01:29:29 So that was the initial condition.
01:29:33 Then we had this tremendous challenge with the model itself.
01:29:36 So now most people were pirating.
01:29:39 But for the people who bought a download or a CD, the artists would get all the revenue
01:29:45 for all the future plays then, right?
01:29:48 So you got it all up front, whereas the streaming model was like almost nothing day one, almost
01:29:51 nothing day two.
01:29:52 And then at some point, this curve of incremental revenue would intersect with your day one
01:29:58 payment.
01:29:59 And that took a long time to play out before before the music labels, they understood
01:30:05 that.
01:30:05 But on the artist side, it took a lot of time to understand that actually, if I have a big
01:30:09 hit that is going to be played for many years, this is a much better model because I get
01:30:14 paid based on how much people use the product, not how much they thought they would use it
01:30:18 day one or so forth.
01:30:20 So it was a complicated model to get across.
01:30:22 But time helped with that.
01:30:24 And now the revenues to the music industry actually are bigger again than it’s gone through
01:30:30 this incredible dip and now they’re back up.
01:30:32 And so we’re very proud of having been a part of that.
01:30:37 So there have been distinct problems.
01:30:39 I think when it comes to the labels, we have taken the painful approach.
01:30:46 Some of our competition at the time, they kind of looked at other companies and said,
01:30:52 if we just ignore the rights, we get really big, really fast.
01:30:56 We’re going to be too big for the labels to kind of, too big to fail.
01:31:00 They’re not going to kill us.
01:31:01 We didn’t take that approach.
01:31:02 We went legal from day one and we negotiated and negotiated and negotiated.
01:31:06 It was very slow.
01:31:07 It was very frustrating.
01:31:08 We were angry at seeing other companies taking shortcuts and seeming to get away with it.
01:31:12 It was this game theory thing where over many rounds of playing the game, this would be
01:31:18 the right strategy.
01:31:19 And even though clearly there’s a lot of frustrations at times during renegotiations, there is this
01:31:25 there is this weird trust where we have been honest and fair.
01:31:31 We’ve never screwed them.
01:31:32 They’ve never screwed us.
01:31:33 It’s 10 years, but there’s this trust and like they know that if music doesn’t get
01:31:39 really big, if lots of people do not want to listen to music and want to pay for it,
01:31:43 Spotify has no business model.
01:31:44 So we actually are incredibly aligned.
01:31:48 Other companies, not to be tense, but other companies have other business models where
01:31:51 even if they made no money from music, they’d still be profitable companies.
01:31:56 But Spotify won’t.
01:31:57 So I think the industry sees that we are actually aligned business wise.
01:32:03 So there is this trust that allows us to do product development, even if it’s scary,
01:32:11 taking risks.
01:32:12 The free model itself was an incredible risk for the music industry to take that they should
01:32:17 get credit for.
01:32:17 Now, some of it was that they had nothing to lose in the game.
01:32:20 Some of it was that they had nothing to lose in Sweden.
01:32:22 But frankly, a lot of the labels also took risk.
01:32:25 And so I think we built up that trust with I think herding of cats sounds a bit.
01:32:32 What’s the word?
01:32:33 It sounds like dismissive of the cats.
01:32:35 Dismissive.
01:32:35 No, every cat matters.
01:32:37 They’re all beautiful and very important.
01:32:39 Exactly.
01:32:39 They’ve taken a lot of risks and certainly it’s been frustrating.
01:32:44 So it’s really like playing it’s game theory.
01:32:47 If you play the game many times, then you can have the statistical outcome that you
01:32:53 bet on.
01:32:54 And it feels very painful when you’re in the middle of that thing.
01:32:57 I mean, there’s risk, there’s trust, there’s relationships.
01:33:00 From just having read the biography of Steve Jobs, similar kind of relationships were discussed
01:33:07 in iTunes.
01:33:08 The idea of selling a song for a dollar was very uncomfortable for labels.
01:33:12 Exactly.
01:33:13 And there was no, it was the same kind of thing.
01:33:16 It was trust, it was game theory as a lot of relationships that had to be built.
01:33:21 And it’s really a terrifyingly difficult process that Apple could go through a little
01:33:28 bit because they could afford for that process to fail.
01:33:32 For Spotify, it seems terrifying because you can’t.
01:33:37 Initially, I think a lot of it comes down to honestly Daniel and his tenacity in negotiating,
01:33:44 which seems like an impossible task because he was completely unknown and so forth.
01:33:50 But maybe that was also the reason that it worked.
01:33:56 But I think game theory is probably the best way to think about it.
01:34:03 You could go straight for this Nash equilibrium that someone is going to defect or you play
01:34:08 it many times, you try to actually go for the top left, the corporations sell.
01:34:14 Is there any magical reason why Spotify seems to have won this?
01:34:20 So a lot of people have tried to do what Spotify tried to do and Spotify has come out.
01:34:25 Well, so the answer is that there’s no magical reason because I don’t believe in magic.
01:34:30 But I think there are there are reasons.
01:34:32 And I think some of them are that people have misunderstood a lot of what we actually do.
01:34:40 The actual Spotify model is very complicated.
01:34:43 They’ve looked at the premium model and said, it seems like you can charge $9.99 for music
01:34:49 and people are going to pay, but that’s not what happened.
01:34:52 Actually, when we launched the original mobile product, everyone said they would never pay.
01:34:56 What happened was they started on the free product and then their engagement grew so
01:35:01 much that eventually they said, maybe it is worth $9.99, right?
01:35:05 It’s your propensity to pay gross with your engagement.
01:35:08 So we have this super complicated business model.
01:35:11 We operate two different business models, advertising and premium at the same time.
01:35:15 And I think that is hard to replicate.
01:35:17 I struggle to think of other companies that run large scale advertising and subscription
01:35:22 products at the same time.
01:35:24 So I think the business model is actually much more complicated than people think it is.
01:35:28 And so some people went after just the premium part without the free part and ran into a
01:35:32 wall where no one wanted to pay.
01:35:35 Some people went after just music should be free, just ads, which doesn’t give you enough
01:35:40 revenue and doesn’t work for the music industry.
01:35:42 So I think that combination is kind of opaque from the outside.
01:35:46 So maybe I shouldn’t say it here and reveal the secret, but that turns out to be hard
01:35:51 to replicate than you would think.
01:35:54 So there’s a lot of brilliant business strategies out there.
01:35:57 Brilliant business strategy here.
01:36:00 Brilliance or luck?
01:36:01 Probably more luck, but it doesn’t really matter.
01:36:03 It looks brilliant in retrospect.
01:36:05 Let’s call it brilliant.
01:36:07 Yeah, when the books are written, they’ll be brilliant.
01:36:10 You’ve mentioned that your philosophy is to embrace change.
01:36:16 So how will the music streaming and music listening world change over the next 10 years,
01:36:23 20 years?
01:36:24 You look out into the far future.
01:36:26 What do you think?
01:36:28 I think that music and for that matter, audio podcasts, audiobooks, I think it’s one of
01:36:35 the few core human needs.
01:36:37 I think it there is no good reason to me why it shouldn’t be at the scale of something
01:36:41 like messaging or social networking.
01:36:44 I don’t think it’s a niche thing to listen to music or news or something.
01:36:48 So I think scale is obviously one of the things that I really hope for.
01:36:50 I think I hope that it’s going to be billions of users.
01:36:54 I hope eventually everyone in the world gets access to all the world’s music ever made.
01:36:58 So obviously, I think it’s going to be a much bigger business.
01:37:01 Otherwise, we wouldn’t be betting this big.
01:37:05 Now, if you look more at how it is consumed, what I’m hoping is back to this analogy of
01:37:13 the software tool chain, where I think I sometimes internally I make this analogy to text messaging.
01:37:22 Text messaging was also based on standards in the area of mobile carriers.
01:37:28 You had the SMS, the 140 character, 120 character SMS.
01:37:33 And it was great because everyone agreed on the standards.
01:37:36 So as a consumer, you got a lot of distributions and interoperability, but it was a very constrained
01:37:40 format.
01:37:41 And when the industry wanted to add pictures to that format to do the MMS, I looked it
01:37:45 up and I think it took from the late 80s to early 2000s.
01:37:48 This is like a 15, 20 year product cycle to bring pictures into that.
01:37:53 Now, once that entire value chain of creation and consumption got wrapped in one software
01:38:00 stack within something like Snapchat or WhatsApp, the first week they added disappearing messages.
01:38:07 Then two weeks later, they added stories.
01:38:09 The pace of innovation when you’re on one software stack and you can affect both creation
01:38:14 and consumption, I think it’s going to be rapid.
01:38:17 So with these streaming services, we now, for the first time in history, have enough,
01:38:22 I hope, people on one of these services.
01:38:25 Actually, whether it’s Spotify or Amazon or Apple or YouTube, and hopefully enough
01:38:29 creators that you can actually start working with the format again.
01:38:32 And that excites me.
01:38:33 I think being able to change these constraints from 100 years, that could really do something
01:38:39 interesting.
01:38:40 I really hope it’s not just going to be the iteration on the same thing for the next 10
01:38:45 to 20 years as well.
01:38:47 Yeah, changing the creation of music, the creation of audio, the creation of podcasts
01:38:52 is a really fascinating possibility.
01:38:54 I myself don’t understand what it is about podcasts that’s so intimate.
01:38:59 It just is.
01:39:00 I listen to a lot of podcasts.
01:39:01 I think it touches on a deep human need for connection that people do feel like they’re
01:39:09 connected to when they listen.
01:39:12 I don’t understand what the psychology of that is, but in this world that’s becoming
01:39:17 more and more disconnected, it feels like this is fulfilling a certain kind of need.
01:39:24 And empowering the creator as opposed to just the listener is really interesting.
01:39:32 I’m really excited that you’re working on this.
01:39:34 Yeah, I think one of the things that is inspiring for our teams to work on podcasts is exactly
01:39:38 that, whether you think, like I probably do, that it’s something biological about perceiving
01:39:44 to be in the middle of the conversation that makes you listen in a different way.
01:39:47 It doesn’t really matter.
01:39:48 People seem to perceive it differently.
01:39:50 And there was this narrative for a long time that if you look at video, everything kind
01:39:55 of in the foreground, it got shorter and shorter and shorter because of financial pressures
01:39:59 and monetization and so forth.
01:40:01 And eventually, at the end, there’s almost like 20 seconds clip, people just screaming
01:40:06 something and I feel really good about the fact that you could have interpreted that
01:40:14 as people have no attention span anymore.
01:40:16 They don’t want to listen to things.
01:40:18 They’re not interested in deeper stories.
01:40:22 People are getting dumber.
01:40:23 But then podcasts came along and it’s almost like, no, no, the need still existed.
01:40:28 But maybe it was the fact that you’re not prepared to look at your phone like this for
01:40:32 two hours.
01:40:32 But if you can drive at the same time, it seems like people really want to dig deeper
01:40:36 and they want to hear like the more complicated version.
01:40:38 So to me, that is very inspiring that that podcast is actually long form.
01:40:42 It gives me a lot of hope for humanity that people seem really interested in hearing deeper,
01:40:48 more complicated conversations.
01:40:49 This is I don’t understand it.
01:40:52 It’s fascinating.
01:40:53 So the majority for this podcast, listen to the whole thing.
01:40:57 This whole conversation we’ve been talking for an hour and 45 minutes.
01:41:02 And somebody will I mean, most people will be listening to these words I’m speaking right
01:41:06 now.
01:41:06 It’s crazy.
01:41:07 You wouldn’t have thought that 10 years ago with where the world seemed to go.
01:41:10 That’s very positive, I think.
01:41:12 That’s really exciting.
01:41:13 And empowering the creator there is really exciting.
01:41:17 Last question.
01:41:18 You also have a passion for just mobile in general.
01:41:22 How do you see the smartphone world, the digital space of smartphones and just everything that’s
01:41:32 on the move, whether it’s Internet of Things and so on, changing over the next 10 years
01:41:39 and so on?
01:41:41 I think that one way to think about it is that computing might be moving out of these
01:41:47 multipurpose devices, the computer we had and the phone, into specific purpose devices.
01:41:55 And it will be ambient that at least in my home, you just shout something at someone
01:42:01 and there’s always one of these speakers close enough.
01:42:03 And so you start behaving differently.
01:42:06 It’s as if you have the Internet ambient, ambiently around you and you can ask it things.
01:42:11 So I think computing will kind of get more integrated and we won’t necessarily think
01:42:15 of it as connected to a device in the same way that we do today.
01:42:21 I don’t know the path to that.
01:42:22 Maybe we used to have these desktop computers and then we partially replaced that with the
01:42:30 laptops and left the desktop at home when I work.
01:42:32 And then we got these phones and we started leaving the mobile phones.
01:42:37 We had the desktop at home when I work and then we got these phones and we started leaving
01:42:41 the laptop at home for a while.
01:42:42 And maybe for stretches of time you’re going to start using the watch and you can leave
01:42:47 your phone at home for a run or something.
01:42:50 And we’re on this progressive path where I think what is happening with voice is that
01:43:00 you have an interaction paradigm that doesn’t require as large physical devices.
01:43:06 So I definitely think there’s a future where you can have your AirPods and your watch and
01:43:12 you can do a lot of computing.
01:43:15 And I don’t think it’s going to be this binary thing.
01:43:20 I think it’s going to be like many of us still have a laptop, we just use it less.
01:43:23 And so you shift your consumption over.
01:43:26 And I don’t know about AR glasses and so forth.
01:43:31 I’m excited about it.
01:43:32 I spent a lot of time in that area, but I still think it’s quite far away.
01:43:35 AR, VR, all of that.
01:43:37 Yeah, VR is happening and working.
01:43:39 I think the recent Oculus Quest is quite impressive.
01:43:43 I think AR is further away.
01:43:45 At least that type of AR.
01:43:48 But I do think your phone or watch or glasses understanding where you are and maybe what
01:43:54 you’re looking at and being able to give you audio cues about that.
01:43:56 Or you can say like, what is this?
01:43:58 And it tells you what it is.
01:44:00 That I think might happen.
01:44:02 You use your watch or your glasses as a mouse pointer on reality.
01:44:08 I think it might be a while before…
01:44:09 I might be wrong.
01:44:10 I hope I’m wrong.
01:44:10 I think it might be a while before we walk around with these big lab glasses that project
01:44:14 things.
01:44:15 I agree with you.
01:44:16 It’s actually really difficult when you have to understand the physical world enough to
01:44:23 project onto it.
01:44:25 I lied about the last question.
01:44:26 Go ahead, because I just thought of audio and my favorite topic, which is the movie
01:44:32 Her, do you think, whether it’s part of Spotify or not, we’ll have, I don’t know if you’ve
01:44:41 seen the movie Her.
01:44:42 Absolutely.
01:44:45 And there, audio is the primary form of interaction and the connection with another entity that
01:44:53 you can actually have a relationship with, that you fall in love with based on voice
01:44:59 alone, audio alone.
01:45:00 Do you think that’s possible, first of all, based on audio alone to fall in love with
01:45:04 somebody?
01:45:05 Somebody or…
01:45:06 Well, yeah, let’s go with somebody.
01:45:08 Just have a relationship based on audio alone.
01:45:11 And second question to that, can we create an artificial intelligence system that allows
01:45:18 one to fall in love with it and her, him with you?
01:45:21 So this is my personal answer, speaking for me as a person, the answer is quite unequivocally
01:45:29 yes on both.
01:45:32 I think what we just said about podcasts and the feeling of being in the middle of a
01:45:36 conversation, if you could have an assistant where, and we just said that feels like a
01:45:42 very personal setting.
01:45:43 So if you walk around with these headphones and this thing, you’re speaking with this
01:45:47 thing all of the time that feels like it’s in your brain.
01:45:49 I think it’s going to be much easier to fall in love with than something that would be
01:45:53 on your screen.
01:45:54 I think that’s entirely possible.
01:45:56 And then from the, you can probably answer this better than me, but from the concept
01:46:00 of if it’s going to be possible to build a machine that can achieve that, I think whether
01:46:07 you think of it as, if you can fake it, the philosophical zombie that assimilates it enough
01:46:12 or it somehow actually is, I think there’s, it’s only a question.
01:46:17 It’s only a question if you ask me about time, I’d have a different answer.
01:46:20 But if you say I’ve given some half infinite time, absolutely.
01:46:24 I think it’s just atoms and arrangement of information.
01:46:29 Well, I personally think that love is a lot simpler than people think.
01:46:33 So we started with true romance and ended in love.
01:46:37 I don’t see a better place to end.
01:46:39 Beautiful.
01:46:40 Gustav, thanks so much for talking today.
01:46:41 Thank you so much.
01:46:42 It was a lot of fun.
01:46:43 It was fun.