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