back to indexGustav Soderstrom: Spotify | Lex Fridman Podcast #29
link |
The following is a conversation with Gustav Sonastrom.
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 a 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 five stars on iTunes,
link |
support on Patreon, or simply connect with me on Twitter at Lex Freedman,
link |
spelled F R I D M A N.
link |
And now here's my conversation with Gustav Sonastrom.
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,
link |
right? 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 it's pretty easy for me.
link |
There is this song called You're So Cool,
link |
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 |
Actually, had to play at my wedding.
link |
I sat with the organist and helped him play it on an organ,
link |
which was a pretty 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,
link |
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 it was a great story of love.
link |
It's fantastic actors.
link |
And and really, I didn't even know who Hans Zimmer was at the time,
link |
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,
link |
I think directed or produced that.
link |
So it's not Stairway to Heaven or Bohemian Rhapsody.
link |
It's those are those are great.
link |
They're not my personal favorites,
link |
but I've realized that people have different tastes.
link |
And that's that's a big part of what we do.
link |
Well, for me, I don't 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
link |
as part of hunting rituals and primitive cultural gatherings,
link |
This is the birth of music.
link |
Since then, we had a few folks, Beethoven, Elvis, Beatles,
link |
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 there's certainly a big purpose,
link |
which is the same as much of entertainment,
link |
which is escapism and to be able to live in some sort of
link |
other mental state for a while.
link |
But I also think you have the the opposite of escaping,
link |
which is to help you focus on something you are actually doing.
link |
As I think people use music as a tool to to tune the brain
link |
to the activities that they are actually doing.
link |
And it's kind of like in one sense,
link |
maybe it's the rawest signal.
link |
If you if you think about the brain as neural networks,
link |
it's maybe the most efficient hack we can do
link |
to actually actively tune it 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
link |
to get to a certain mood very fast.
link |
You know, there's a there's a social component historically
link |
to music where people listen to music together.
link |
I was just thinking about this,
link |
that to me, and you mentioned machine learning,
link |
but to me personally, music is a really private thing.
link |
Like I'm speaking for myself.
link |
I listen to music like almost nobody knows
link |
the kind of things I have in my library,
link |
except people who are really close to me
link |
and they really only know a certain percentage.
link |
There's like some weird stuff
link |
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.
link |
For me, it's personal.
link |
Do you think of music as something that's social
link |
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
link |
during these 10 years at Spotify, obviously.
link |
In one sense, as you said, music is incredibly social.
link |
You go to concerts and so forth.
link |
On the other hand, it is your escape
link |
and everyone has these things that are very personal to them.
link |
So what we've found is that when it comes to,
link |
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,
link |
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,
link |
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
link |
is something we tried on social networks and so forth,
link |
but it turns out that people aren't super interested in
link |
what their friends listen to.
link |
They're interested in understanding
link |
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 that this morning listening to Spotify,
link |
I really have a pretty intimate relationship with Spotify,
link |
with my playlists.
link |
I've had them for many years now
link |
and they've grown with me together.
link |
There's an intimate relationship you have
link |
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
link |
and try to give a history of music listening
link |
from your perspective, from before the internet
link |
and after the internet?
link |
And just kind of everything leading up to streaming
link |
with Spotify and so on.
link |
I'll try, 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
link |
during the history of music,
link |
which is that before recorded music,
link |
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,
link |
basically concerts.
link |
And so you either had to get to the nearest village
link |
to listen to music.
link |
And while that was cumbersome
link |
and it severely limited the distribution of music,
link |
it also had some different qualities,
link |
which was that the creator could always interact
link |
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
link |
that these early classical works,
link |
they're much longer than the three minutes.
link |
The three minutes came in as a restriction
link |
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 the,
link |
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, like instead of doing this opus
link |
of like many, you know, tens of minutes or something,
link |
now you get three and a half minutes
link |
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,
link |
there's a scarcity component 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,
link |
you derive perhaps more pleasure
link |
because it's more rare for you to listen to 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
link |
if you have something in infinite amount,
link |
so will you value it as much?
link |
Humanity is always seeking some, is 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's why concerts exist.
link |
So you can actually have both.
link |
But I think net, if you couldn't listen to music
link |
in your car driving, that'd be worse.
link |
That cost will be bigger than the benefit of the anticipation,
link |
I think, that you would have.
link |
So, yeah, it started with live concerts.
link |
Then it's being able to, you know, the photograph invented, right?
link |
You start to be able to record music.
link |
So then you got this massive distribution
link |
that made it possible to create two things.
link |
I think, first of all, cultural phenomenons.
link |
They probably need distribution.
link |
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,
link |
like Beatles and Elvis and so forth,
link |
that would really a function of distribution, I think.
link |
Obviously, of talent and innovation,
link |
but there was also taking a component.
link |
And of course, the next big innovation to come along
link |
was radio, broadcast radio.
link |
And I think radio is interesting
link |
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
link |
so that they could honestly play more ads and make more money.
link |
And music was free.
link |
So then you had this massive distribution
link |
where you could program to people.
link |
I think those things, that ecosystem,
link |
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,
link |
but maybe not such a long tail.
link |
In terms of choice of everybody listening to the same stuff.
link |
Yeah. And as you said, I think there are some social benefits to that.
link |
I think, for example, there is a high statistical chance
link |
that if I talk about the latest episode of Game of Thrones,
link |
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.
link |
But I also obviously love personalization.
link |
And so let's catch this up to the internet.
link |
Well, first of all, there's like MP3s.
link |
Exactly. There's tape, 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,
link |
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
link |
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
link |
music was a file again 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
link |
music is no longer constrained in 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,
link |
but it still happens.
link |
So first of all, I don't really know what EDM is.
link |
Electronic dance music, you could say.
link |
Avicii was one of the biggest in this genre.
link |
So the main constraint is of time,
link |
something like three, four, five minutes song.
link |
So you could have songs that were eight minutes,
link |
10 minutes, and so forth.
link |
Because it started as a digital product
link |
that you downloaded, so you didn't have this constraint anymore.
link |
So I think it's something really interesting
link |
that I don't think has fully happened yet.
link |
We're kind of jumping ahead a little bit to where we are,
link |
but I think there's tons of formal innovation in music
link |
that should happen now, that couldn't happen
link |
when you needed to really adhere to the distribution constraints.
link |
If you didn't adhere to that, you would get no distribution.
link |
So Björk, for example, an Icelandic artist,
link |
she made a full iPad app as an album.
link |
That was very expensive, even though the App Store
link |
has great distribution, she gets nowhere near the distribution
link |
versus staying within the three minute format.
link |
So I think now that music is fully digital
link |
inside these streaming services,
link |
there is the opportunity to change the format again
link |
and allow creators to be much more creative
link |
without limiting their distribution ability.
link |
That's interesting that you're right.
link |
You're right, it's surprising that we don't see
link |
that taking advantage more often.
link |
It's almost like the constraints of the distribution
link |
from the 50s and 60s have molded the culture
link |
to where we want the three to five minute song,
link |
that anything else, not just...
link |
So we want the song as consumers and as artists,
link |
like, because I write a lot of music
link |
and I never even thought about writing something
link |
longer than 10 minutes.
link |
It's really interesting that those constraints...
link |
Because all your training data
link |
has been three and a half minutes long, right?
link |
Okay, so yeah, digitization of data led to then MP3s.
link |
Yeah, so I think you had this file then
link |
that was distributed physically,
link |
but then you had the components of digital distribution
link |
and then the internet happened and there was this vacuum
link |
where you had a format that could be digitally shipped
link |
but there was no business model.
link |
And then all these pirate networks happened.
link |
Napster and in Sweden, Pirate Bay,
link |
which was one of the biggest.
link |
And I think from a consumer point of view,
link |
which leads up to the inception of Spotify
link |
from a consumer point of view,
link |
consumers for the first time had this access model to music
link |
where they could, without any marginal cost,
link |
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.
link |
They have access to all the music ever made.
link |
I think it was fantastic.
link |
But it was also horrible for artists
link |
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
link |
before there was a business model.
link |
And then there were these download stores
link |
that allowed you to download files.
link |
Which was a solution,
link |
but it didn't solve the access problem.
link |
There was still a marginal cost of 99 cents
link |
to try one more track.
link |
And I think that heavily limits how you listen to music.
link |
The example I always give is in Spotify,
link |
a huge amount of people listen to music while they sleep,
link |
while they go to sleep and while they sleep.
link |
If that costed you 99 cents per three minutes,
link |
you probably wouldn't do that.
link |
And you would be much less adventurous
link |
if there was a real dollar cost to explore music.
link |
So the access model is interesting in that
link |
it changes your music behavior.
link |
You can take much more risk
link |
because there's no marginal cost to it.
link |
Maybe let me linger on piracy for a second.
link |
Because I find, especially coming from Russia,
link |
piracy is something that's very interesting.
link |
To me, not me of course ever,
link |
but I have friends who've partook in piracy
link |
of music, software, TV shows, sporting events.
link |
And usually, to me, what that shows
link |
is not that 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
link |
is a business opportunity in all these domains.
link |
And that's where I think you're right.
link |
Spotify stepped in is basically,
link |
piracy is an experience.
link |
You can explore, find music you like.
link |
And actually, the interface of piracy is horrible.
link |
Because it's, I mean...
link |
Bad metadata, long download times, all kinds of stuff.
link |
And what Spotify does is basically
link |
first, rewards artists,
link |
and second, makes the experience
link |
of exploring music much better.
link |
I mean, the same is true, I think, for movies and so on.
link |
Piracy reveals, in the software space, for example,
link |
I'm a huge user and fan of Adobe products.
link |
And there was much more incentive
link |
to pirate Adobe products
link |
before they went to a monthly subscription plan.
link |
And now, all of the said friends
link |
that used to pirate Adobe products that I know,
link |
now actually pay, gladly, for the monthly subscription.
link |
Yeah, I think you're right.
link |
I think it's a sign of an opportunity
link |
for product development.
link |
And that sometimes there's a product market fit
link |
before there's a business model fit in 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
link |
a political party called the Pirate Party.
link |
And this was during the time when people said that
link |
information should be free.
link |
It was somehow wrong to charge for ones and zeros.
link |
So I think people felt that artists
link |
should probably make some money somehow else
link |
in concerts or something.
link |
So at least in Sweden, it was part really social acceptance
link |
even at the political level.
link |
But that also forced Spotify to compete with free,
link |
which I don't think could have happened
link |
anywhere else in the world.
link |
The music industry needed to be doing bad enough
link |
to take that risk.
link |
And Sweden was like the perfect testing ground.
link |
It had government funded high bandwidth, low latency broadband,
link |
which meant that the product would work.
link |
And it was also, there was no music revenue anyway.
link |
So they were kind of like,
link |
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
link |
in America, the world's largest music market, for example.
link |
So how do you compete with free?
link |
Because that's an interesting world of the internet
link |
where most people don't like to pay for things.
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
link |
was the first business model
link |
because no one would put the 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.
link |
But definitely back then, we needed to compete with free.
link |
And the first thing you need to do is obviously
link |
to lower the price to free.
link |
And then you need to be better somehow.
link |
And the way that Spotify was better was on the user experience,
link |
on the actual performance, the latency of, you know,
link |
even if you had high band with broadband,
link |
it would still take you 30 seconds to a minute
link |
to download one of these tracks.
link |
So the Spotify experience of starting within the
link |
perceptual limit of immediacy, about 250 milliseconds,
link |
meant that the whole trick was,
link |
it felt as if you had downloaded all of Pirate Bay.
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 |
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,
link |
because I just took it for granted.
link |
You know, whenever an awesome thing comes along,
link |
you just like, oh, of course it has to be this way.
link |
That's exactly right.
link |
That it felt like the entire world's libraries
link |
at my fingertips because of that latency being reduced.
link |
What was the technical challenge in reducing the late?
link |
So there was a group of really, really talented engineers.
link |
One of them called Ludwig Strigius.
link |
He wrote the, actually from Gothenburg,
link |
he wrote the initial, the UTorrent Clients,
link |
which is kind of an interesting backstory to Spotify,
link |
you know, that we have one of the top developers
link |
from BitTorrent Clients as well.
link |
So he wrote UTorrent, the world's smallest BitTorrent Clients.
link |
And then he was acquired very early by Daniel and Martin,
link |
who founded Spotify, and they actually sold the UTorrent Client
link |
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,
link |
where Spotify built an end to end media distribution system.
link |
Up until only a few years ago, we actually hosted all the music ourselves.
link |
So we had both the server side and the client,
link |
and that meant that we could do things such as having a peer to peer solution
link |
to use local caching on the client side,
link |
because back then the world was mostly desktop.
link |
But we could also do things like hack the TCP protocols,
link |
things like Nagle's algorithm for kind of exponential back off,
link |
or ramp up and just go full throttle and optimize for latency
link |
at the cost of bandwidth.
link |
And all of this end to end control meant that we could do
link |
an experience that felt like a step change.
link |
These days, we actually are on on GCP.
link |
We don't host our own stuff, and everyone is really fast these days.
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 |
That was in 2008, 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
link |
to discover the songs they like,
link |
and then they would torrent those songs
link |
so they can copy it to their phone.
link |
Seriously, piracy does seem to be a good guide
link |
for business models.
link |
As far as I know, Spotify doesn't have video content.
link |
Well, we do have music videos,
link |
and we do have videos on the service,
link |
but the way we think about ourselves is that
link |
we're an audio service,
link |
and we think that if you look at the amount of time
link |
that people spend on audio,
link |
it's actually very similar to the amount of time
link |
that people spend on video.
link |
So the opportunity should be equally big,
link |
but today it's not at all valued.
link |
Video is valued much higher.
link |
So we think it's basically completely undervalued.
link |
We think of ourselves as an audio service,
link |
but within that audio service, I think video
link |
can make a lot of sense.
link |
I think when you're discovering an artist,
link |
you probably do want to see them
link |
and understand who they are,
link |
to understand their identity.
link |
Do you want to 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,
link |
but we're an audio service where,
link |
think of it as we call it internally,
link |
backgroundable video.
link |
Video that is helpful,
link |
but isn't the driver of the narrative.
link |
I think also if you look at YouTube,
link |
there's quite a few folks who listen to music on YouTube.
link |
YouTube is a bit of a competitor to Spotify,
link |
which is very strange to me that people use YouTube
link |
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
link |
strangely maybe it's similar to what we were
link |
for the Piracy Networks,
link |
where YouTube for historical reasons
link |
have a lot of music videos.
link |
people use YouTube for a lot of the discovery part
link |
of the process, I think.
link |
But then it's not a really good
link |
sort of quote unquote MP3 player,
link |
because it doesn't even background.
link |
Then you have to keep the app in the foreground.
link |
So the consumption,
link |
it's not a good consumption tool,
link |
but it's a decently good discovery.
link |
I mean, I think YouTube is fantastic products.
link |
And I use it for all kinds of purposes.
link |
If I were to admit something,
link |
I do use YouTube a little bit for the discovery,
link |
to assist in the discovery process of songs.
link |
And then if I like it,
link |
I'll add it to Spotify.
link |
That's okay with us.
link |
So sorry, we're jumping around a little bit.
link |
So this kind of incredible,
link |
you look at Napster,
link |
you look at the early days of Spotify.
link |
one fascinating point is how do you grow
link |
So you're there in Sweden.
link |
I saw the initial sketches that looked terrible.
link |
How do you grow a user base
link |
from a few folks to millions?
link |
I think there are a bunch of tactical answers.
link |
I think you need a great product.
link |
I don't think you take a bad product
link |
and market it to be successful.
link |
So you need a great product.
link |
Sorry to interrupt,
link |
but it's a totally new way to listen to music.
link |
did people realize immediately that Spotify is a great product?
link |
So back to the point of piracy,
link |
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,
link |
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 legal
link |
because it seemed too fast and too good to be true,
link |
which I think is a great product proposition
link |
if you can be too good to be true.
link |
But what I saw again and again
link |
was people showing each other,
link |
clicking the song, showing how fast it started
link |
and say, can you believe this?
link |
So I really think it was about speed.
link |
Then we also had an invite program
link |
that was really meant for scaling
link |
because we hosted our own servers.
link |
We needed to control scaling,
link |
but that built a lot of expectation
link |
and I don't want to say hype
link |
because hype implies that it wasn't true.
link |
Expectations, excitement around the product.
link |
And we replicated that when we launched in the US.
link |
We also built up an invite only program first.
link |
There are lots of tactics,
link |
but I think you need a great product
link |
to solve some problem.
link |
And basically the key innovation,
link |
there was technology, but on a meta level,
link |
the innovation was really the access model
link |
versus the ownership model.
link |
And that was tricky.
link |
A lot of people said that 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
link |
helped quite a lot.
link |
So this is an interesting psychological point
link |
that maybe you can speak to.
link |
It was a big shift for me.
link |
It's almost like 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,
link |
is basically hoarding music.
link |
As you're slowly one song by one song,
link |
gathering a collection of music that you love.
link |
Especially with CDs or tape,
link |
you physically had it.
link |
what I had to come to grips with,
link |
and it was kind of liberating actually,
link |
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,
link |
so actually we've seen the user data.
link |
When Spotify started,
link |
a lot of people did the exact same thing.
link |
They started hoarding,
link |
as if the music would disappear.
link |
Almost the equivalent of downloading.
link |
And so we had these playlists that had limits
link |
of a few hundred thousand tracks
link |
and we figured no one would ever.
link |
It's in hundreds and hundreds of thousands of tracks.
link |
some people want to actually save,
link |
and play the entire catalog.
link |
But I think that the therapy session
link |
goes something like,
link |
instead of throwing away your music,
link |
if you took your files
link |
and you stored them in a locker at Google,
link |
it'd be a streaming service.
link |
It's just that in that locker,
link |
you have all the world's music now for free.
link |
So instead of giving away your music,
link |
you got all the music.
link |
You could think of it as having a copy
link |
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
link |
and you sent it to someone who stored it for you.
link |
And once you go through that mental journey of,
link |
like, still my files, they're just over there.
link |
And I just have 40 million or 50 million or something now.
link |
Then people are like, okay, that's good.
link |
The problem is, I think,
link |
because you paid us a subscription,
link |
if we hadn't had the free tier,
link |
where you would feel like,
link |
even if I don't want to pay anymore,
link |
I still get to keep them.
link |
You keep your playlists 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,
link |
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
link |
and was the big challenge for a lot of our competitors.
link |
That's another reason why I think the free tier
link |
is really important,
link |
that people need to feel the security
link |
that the work they put in, it will never disappear,
link |
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 |
Spotify taught me to just enjoy music
link |
as opposed to what I was doing before,
link |
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
link |
way too much to where I was getting sick of them.
link |
Whereas Spotify, the more liberating kind of approach,
link |
I was just enjoying, of course,
link |
I listened to Stairway to Heaven over and over,
link |
but because of the extra variety,
link |
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,
link |
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 playlists than songs.
link |
What do you make of that?
link |
So the way I think about it is that
link |
from a statistical machine learning point of view,
link |
you have all these,
link |
if you want to think about reinforcement learning,
link |
you have this state space of all the tracks,
link |
and you can take different journeys through this world.
link |
And I think of these as like people
link |
helping themselves and each other,
link |
creating interesting vectors through this space of tracks.
link |
And then it's not so surprising that across
link |
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
link |
having found all of them.
link |
So kind of our job now is,
link |
when Spotify started, it was really a search box
link |
that was for the time pretty powerful,
link |
and then I like to refer to this programming language
link |
called playlisting, where if you,
link |
as you probably were pretty good at music,
link |
you knew your new releases, you knew your back catalog,
link |
you knew you're starting with the heaven,
link |
you could create a soundtrack for yourself
link |
using this playlisting tool that's like metaprogramming language
link |
for music to soundtrack your life.
link |
And people who are good at music,
link |
it's back to how do you scale the product.
link |
For people who are good at music,
link |
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 |
The problem was, 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
link |
was to essentially try to build,
link |
you can think of them as agents,
link |
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
link |
200 million active users on Spotify.
link |
So from the machine learning perspective,
link |
you have these 200 million people plus.
link |
it's really interesting to think of playlists
link |
as, I don't know if you meant it that way,
link |
but it's almost like a programming language.
link |
It's released a trace of exploration
link |
of those individual agents, the listeners.
link |
And you have all this new tracks coming in.
link |
So it's a fascinating space that is ripe for machine learning.
link |
So how can playlists be used as data
link |
in terms of machine learning
link |
to help Spotify organize the music?
link |
So we found in our data,
link |
not surprising that people who playlisted lots,
link |
they retained much better.
link |
We 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
link |
and professional playlisters
link |
and kind of leveraged the maximum of human intelligence
link |
to help build kind of these vectors
link |
through the track space for people.
link |
And that brought in the product.
link |
Then the obvious next,
link |
and we used statistical means
link |
where they could see when they created a playlist,
link |
how did that playlist perform?
link |
They could see skips of the songs,
link |
they could see how the songs perform,
link |
and they manually iterated the playlist
link |
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
link |
was to go from kind of group personalization,
link |
using editors and tools and statistics
link |
to individualization.
link |
And then what's so interesting
link |
about three billion playlists we have is,
link |
the truth is we lucked out.
link |
This was not a priori strategy,
link |
as is often the case.
link |
It looks really smart in hindsight,
link |
but it was dumb luck.
link |
We looked at these playlists
link |
and we had some people in the company,
link |
a person named Erik Bernhardtson,
link |
who was really good at machine learning
link |
already back then, in like 2007, 2008.
link |
Back then it was mostly collaborative filtering and so forth,
link |
we realized that what this is,
link |
is people are grouping tracks for themselves
link |
that have some semantic meaning to them.
link |
And then they actually label it
link |
with a playlist name as well.
link |
So in a sense, people were grouping tracks
link |
along semantic dimensions and labeling them.
link |
And so could you use that information
link |
to find that latent embedding?
link |
And so we started playing around
link |
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 would be quite surprising
link |
if playlists were actually random,
link |
if they had no semantic meaning.
link |
For most people, they grouped these tracks for some reason.
link |
So we just happened across this incredible data set
link |
where people had taken these tens of millions of tracks
link |
and grouped them along different semantic vectors.
link |
And the semantics being outside the individual user,
link |
so it's some kind of universal.
link |
There's a universal embedding
link |
that holds across people on this earth.
link |
Yes, I do think that the embeddings you find
link |
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,
link |
there were these latent similarities.
link |
They were very powerful.
link |
And it was interesting because
link |
I think that the people who play listed the most initially
link |
were the so called music aficionados
link |
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,
link |
if you look at the problem from the outside,
link |
you might expect that the algorithms
link |
would start performing best with mainstreamers first
link |
and now feels like an easier problem to solve mainstream taste
link |
than really particular taste.
link |
It was a complete opposite for us.
link |
The recommendations performed fantastically
link |
for people who saw themselves as 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
link |
So we had a 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
link |
that analyzes song data.
link |
in your view, maybe you can talk about
link |
so what kind of data is there
link |
from a machine learning perspective?
link |
There's a huge amount,
link |
what we're talking about, play listing
link |
and just user data
link |
of what people are listening to, the playlist they're constructing
link |
And then there is the actual data within a song
link |
what makes a song,
link |
I don't know, the actual waveforms.
link |
How do you mix the two?
link |
How much values are in each?
link |
To me, it seems like user data
link |
is a romantic notion
link |
that the song itself would contain useful information
link |
but if I were to guess,
link |
user data would be much more powerful.
link |
Playlists would be much more powerful.
link |
Yeah, so we use both.
link |
Our biggest success initially was with
link |
playlist data without understanding
link |
anything about the structure of the song.
link |
But when we acquired the Echo Nest,
link |
they had the inverse problem.
link |
They actually didn't have any play data.
link |
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
link |
and they looked at Wikipedia for cultural references
link |
and so forth, right?
link |
And did a lot of NLU and so forth.
link |
They had that skill into the company
link |
and combined our user data
link |
with their 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
link |
that has no play data,
link |
obviously you have to try to go by
link |
either who the artist is
link |
or the sonic information in the song
link |
or what it's similar to.
link |
And we do a lot in both.
link |
But I would say yes.
link |
The user data captures things that
link |
have to do with culture in the greater society
link |
that you would never see
link |
in the content itself.
link |
But that said, we have seen
link |
we have a research lab in Paris
link |
when we can talk about
link |
more about that on
link |
kind of machine learning on the creator side.
link |
What it can do for creators, not just for the consumers.
link |
But where we looked at
link |
how does the structure of a song actually affect
link |
the listening behavior.
link |
And it turns out that
link |
we can predict things like skips
link |
based on the song itself.
link |
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
link |
because it is some sort of mind hack.
link |
So there should be structure.
link |
That's probably what we respond to.
link |
You just blew my mind actually
link |
from the creator perspective.
link |
That probably most creators
link |
are not taken advantage of.
link |
So I have recently
link |
got to interact with a few
link |
obsessed with this idea
link |
do to make sure people
link |
keep watching the video.
link |
And then you look at the analytics
link |
of which point do people turn it off
link |
and so on. First of all,
link |
I don't think that's healthy.
link |
It's because you can do it a little too much.
link |
But it is a really
link |
powerful tool for helping the creative process.
link |
realize you could do the same thing for
link |
creation of music.
link |
So is that something you've looked
link |
to how much opportunity there is for that kind of thing?
link |
I listened to the podcast
link |
with Zirash and I thought it was
link |
fantastic and reacted to the same thing
link |
where he said he posted something
link |
in the morning, immediately
link |
watched the feedback where the drop off was
link |
and then responded to that in the afternoon.
link |
Which is quite different
link |
from how people make podcasts, for example.
link |
The feedback loop is almost
link |
non existent. So if we
link |
back out at one level, I think
link |
actually both for music
link |
and podcasts, which we also
link |
do at Spotify, I think there's
link |
tremendous opportunity just
link |
for the creation workflow.
link |
I think it's really interesting
link |
speaking to you, because you're
link |
a musician, a developer and a
link |
podcaster, if you think about those
link |
three different roles,
link |
if you make the leap as a musician,
link |
if you think about it
link |
as a software tool chain, really,
link |
your door with the stems,
link |
that's the IDE, right?
link |
That's where you work in source code format
link |
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
link |
some sort of, you know, AAC,
link |
or something, you do that because
link |
you get distribution. There are so many run times
link |
for that MP3 across the world in car stairs and stuff.
link |
So you kind of compile this executable and you ship it out
link |
in kind of an old fashioned
link |
box software analogy.
link |
And then you hope for the best,
link |
a software developer,
link |
you would never do that. First you go
link |
on GitHub and you collaborate with other creators.
link |
And then, you know,
link |
you think it would be crazy to just ship one version
link |
of your software without doing an A.B. test,
link |
without any feedback loop.
link |
And then, issue tracking.
link |
Exactly. And then you would look at the
link |
feedback loops and try to optimize that thing, right?
link |
So I think if you think
link |
about it as a very specific software tool chain,
link |
The tools that a music creator has versus
link |
what a software developer has.
link |
So that's kind of how we think about it.
link |
have something like GitHub where you could collaborate
link |
much more easily? So we bought
link |
this company called Soundtrap,
link |
which has a kind of
link |
Google Docs for music approach where you can
link |
collaborate with other people on the kind of
link |
source code format with Stems.
link |
And I think introducing things like
link |
AI tools there to help you
link |
as you're creating music,
link |
put a component to your music,
link |
like drums or something,
link |
master and mix automatically,
link |
help you understand how this track will perform.
link |
Exactly what you would expect as a software
link |
developer. I think it makes a lot of sense.
link |
And I think the same goes for
link |
a podcaster. I think podcasters will expect
link |
to have the same kind of feedback loop that Sirosh
link |
why wouldn't you? Maybe it's not healthy, but
link |
Sorry, I wanted to
link |
criticize the fact because you can overdo it.
link |
Because a lot of the
link |
and we're in a new era
link |
addicted to it and
link |
what people say you become a slave to the YouTube
link |
It's always a danger
link |
of a new technology as opposed to
link |
say if you're creating a song
link |
becoming too obsessed
link |
intro riff to the song
link |
that keeps people listening versus
link |
actually the entirety of the creation process. It's a balance.
link |
there's zero, I mean you're blowing my mind right now
link |
completely right that there's no
link |
signal whatsoever. There's no feedback
link |
whatsoever on the creation process
link |
and musical podcasting
link |
And are you saying
link |
is hoping to help create tools
link |
to, not tools, but
link |
No, tools actually.
link |
Actually tools for creators.
link |
we've made some acquisitions the last few years around music creation
link |
this company called Soundtrap
link |
which is a digital audio workstation
link |
that is browser based
link |
and their focus was really the Google Docs approach.
link |
We can collaborate with people much more easily
link |
than you could in previous tools.
link |
So we have some of these tools
link |
that we're working with that we want to make accessible
link |
and then we can connect it
link |
with our consumption data.
link |
We can create this feedback loop where
link |
we could help you understand
link |
we could help you create
link |
and help you understand how you will perform.
link |
We also acquired this other company
link |
with a podcasting called Anchor
link |
which is one of the biggest podcasting tools
link |
So really focused on simple creation
link |
or easy access to creation
link |
but that also gives us this feedback loop
link |
and even before that
link |
we invested in something called
link |
Spotify for Artists
link |
and Spotify for Podcasters
link |
which is an app that you can download
link |
you can verify that you are that creator
link |
software developers have had for years.
link |
if you look at your podcast for example on Spotify
link |
or a song that you released
link |
you can see how it's performing
link |
which citizen it's performing in
link |
who's listening to it, what's the demographic breakup
link |
so similar in the sense that
link |
you can understand how you're actually doing on the platform.
link |
So we definitely want
link |
to build tools. I think
link |
you also interviewed
link |
the head of research for Adobe
link |
and I think that's
link |
back to Photoshop that you like.
link |
I think that's an interesting analogy as well.
link |
Photoshop I think has been
link |
very innovative in helping
link |
photographers and artists
link |
and I think there should be
link |
the same kind of tools for
link |
music creators where you could get
link |
AI assistants for example as you're creating music
link |
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
link |
is a distribution for the content
link |
you create. So you don't have
link |
the data of if I create
link |
you know whatever creation I make
link |
in Photoshop or Premiere
link |
I can't get like immediate feedback
link |
like I can on YouTube for example about
link |
the way people are responding
link |
and if Spotify is creating those tools
link |
that's a really exciting
link |
let's talk a little about podcasts
link |
talking to one person
link |
so it's a bit terrifying
link |
and kind of hard to fathom
link |
people will listen to this episode
link |
so I hosted on Blueberry
link |
I don't know if I'm pronouncing
link |
that correctly actually
link |
it looks like most people listen to it on Apple
link |
Podcast, Cast Box and Pocket
link |
Cast and only about
link |
just my podcast right
link |
do you see a time when Spotify
link |
will dominate this so Spotify
link |
so yeah in podcasting
link |
what's the deal with podcasting in Spotify
link |
is Spotify about podcasting
link |
do you see a time where everybody would listen
link |
to you know probably
link |
a huge amount of people, majority perhaps
link |
listen to music on Spotify
link |
when the same is true for
link |
well I certainly hope so that is our mission
link |
our mission as a company is actually to
link |
enable a million creators to live
link |
off of their art and a billion people inspired
link |
by it and what I think is interesting about that mission
link |
is it actually puts the creators
link |
first even though it started as a
link |
consumer focused company and it says
link |
to be able to live off of their art not just
link |
make some money off of their art as well
link |
so it's quite an ambitious
link |
so we think about creators of all kinds
link |
we kind of expanded our mission from
link |
being music to being audio
link |
we think we made that decision
link |
we think that decision
link |
we think the world made that decision
link |
whether we like it or not
link |
when you put in your headphones
link |
you're going to make a choice between
link |
and a new episode of
link |
your podcast or something else
link |
we're in that world whether we like it or not
link |
that's how radio work
link |
so we decided that
link |
we think it's about audio you can see the
link |
rights of audio books and so forth
link |
we think audio is this great opportunity
link |
so we decided to enter it and obviously
link |
and Apple podcast is absolutely
link |
podcasting and we didn't have
link |
a single podcast only like two years ago
link |
what we did though was
link |
we looked at this and said
link |
bring something to this
link |
we want to do this but back to the original
link |
Spotify we had to do something that consumers actually value
link |
to be able to do this and the reason
link |
we've gone from not existing at all
link |
quite a wide margin the second largest podcast
link |
still wide gap to iTunes but we're growing
link |
I think it's because when we looked at the consumer
link |
people said surprisingly that they wanted their podcasts
link |
and music in the same
link |
in the same application
link |
so what we did was we took a little bit of a different
link |
approach what we said instead of building a separate podcast
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
link |
seamless experience where you can have your podcasts
link |
in the same application
link |
because we think it's audio to you
link |
and that has been successful and that meant that
link |
we actually had 200 million people
link |
to offer this to instead of starting from zero
link |
so I think we have a good
link |
chance because we're taking a different approach
link |
than the competition and back to the other thing
link |
because we're looking at the end to end flow
link |
I think there's a tremendous amount of innovation
link |
to do around podcast as a format
link |
when we have creation tools and consumption
link |
start improving what podcasting is
link |
I mean podcast is this
link |
this opaque big like one to
link |
that you're streaming
link |
which it really doesn't make that much sense in 2019
link |
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
link |
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 got me hooked there
link |
it's the only stats I have
link |
Blueberry just recently added the stats of
link |
it's listen to the end
link |
and that's like a huge improvement
link |
nowhere to where you could possibly
link |
go into statistics
link |
just download the Spotify podcasters up and verify
link |
and then you'll know where people dropped out in this episode
link |
the moment I started talking
link |
I might be depressed by this
link |
one other question
link |
and I have a question about
link |
podcasting in this line
link |
is the idea of albums
link |
friends who are really
link |
often really enjoy albums
link |
listening to entire albums of
link |
an artist correct me if I'm wrong
link |
Spotify has helped
link |
replace the idea of an album with playlists
link |
so you create your own albums
link |
that's kind of the way
link |
at least I've experienced music
link |
and I've really enjoyed it that way
link |
one of the things that was missing
link |
in podcasting for me
link |
I don't know if it's missing I don't know
link |
it's an open question for me
link |
but the way I listen to podcasts is the way I would listen to albums
link |
Joe Rogan Experience
link |
and that's an album and I listen
link |
and I listen one episode after the next
link |
and there's a sequence and so on
link |
what Spotify did for music
link |
sort of this kind of playlisting
link |
idea of breaking apart from podcasting
link |
podcasts and creating kind of
link |
or have you thought about that
link |
it's a great question so I think in
link |
you're right basically you bought an album
link |
so it was like you bought a small catalog of
link |
again it was actually a lot of consumption
link |
you think it's about what you like
link |
but it's based on the business model
link |
you paid for this 10 track
link |
service and then you listen to that for a while
link |
and then when everything was flat
link |
priced you tended to listen differently
link |
the album is still tremendously important that's why we have it
link |
and you can save albums and so forth
link |
and you have a huge amount of people who really listen
link |
according to albums and I like that because
link |
it is a creator format you can tell a longer
link |
story over several tracks
link |
and so some people listen to just one track
link |
some people actually want to hear that whole
link |
now in podcast I think
link |
I think it's different
link |
you can argue that podcasts might be more
link |
like shows on Netflix
link |
you have like a full season of Narcos
link |
and you're probably not going to do like one
link |
episode of Narcos and then one of House of Cards
link |
there's a narrative there
link |
and you love the cast
link |
and you love these characters
link |
so I think people love shows
link |
and I think they will
link |
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
link |
whatever the 5, 6, 10 things that you're into
link |
I think people are going to listen
link |
and love those hosts
link |
for a long time because I think there's something
link |
different with podcasts
link |
experience of the audience
link |
is actually standing here right between us
link |
whereas if you look at something on TV
link |
the audio actually would come from
link |
you would sit over there and the audio would come to you
link |
from both of us as if you were watching
link |
not as if you were part of the conversation
link |
so my experience of having to listen to podcasts
link |
like yours and Joe Rogan
link |
I feel like I know all of these people
link |
and I have no idea who I am
link |
but I feel like I've listened to so many hours of them
link |
it's very different from me watching
link |
like a TV show or an interview
link |
so I think you kind of
link |
fall in love with people
link |
and experience in a different way
link |
so I think shows and hosts
link |
are going to be very important
link |
I don't think that's going to go away into some sort of thing
link |
where you 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
link |
discovery opportunity
link |
in podcasts because
link |
the catalogue is growing quite quickly
link |
I think podcasts is only
link |
a few like five, six hundred thousand
link |
if you look back to YouTube as another analogy
link |
for creators, no one really knows
link |
if you would lift the lid on
link |
YouTube but it's probably billions
link |
and so I think the podcast catalogue would probably grow
link |
tremendously because the creation
link |
tools are getting easier
link |
and you're going to have this
link |
discovery opportunity that I think is really big
link |
so a lot of people tell me that they love their shows
link |
podcasts kind of suck
link |
it's really hard to get into new show
link |
they're usually quite long, it's a big time investment
link |
so I think there's plenty of opportunity
link |
in the discovery part
link |
yeah for sure, a hundred percent
link |
and even the dumbest
link |
there's so many low hanging fruit too
link |
what episode to listen to first
link |
to try out a podcast
link |
exactly because most podcasts
link |
don't have an order to them
link |
they can be listened to out of order
link |
some are better than others
link |
episodes so some episodes of Joe Rogan
link |
are better than others
link |
and it's nice to know
link |
which you should listen to
link |
to try it out and there's as far
link |
as I know almost no information
link |
on how good an episode is
link |
so I think part of the problem is
link |
it's kind of like music
link |
there isn't one answer, people use music for different things
link |
and there's actually many different types of music
link |
there's workout music and there's classical piano music
link |
and focus music and
link |
I think the same with podcasts, some podcasts are sequential
link |
they're supposed to be listened to
link |
it's actually telling and narrative
link |
podcasts are one topic
link |
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
link |
we should recommend
link |
one episode because it's about AI
link |
but then they talk about something that you're not interested in
link |
the rest of the episode
link |
so I think what we're spending a lot of time on now
link |
is just first understanding the domain
link |
and creating kind of the knowledge graph
link |
how do these objects relate
link |
and how do people consume and I think we'll find that it's going to be
link |
it's going to be different
link |
Spotify is the first
link |
people I'm aware of that are
link |
podcasting has been like a wild west
link |
we want to be very careful though
link |
because it's been a very good wild west
link |
I think it's this fragile ecosystem
link |
we want to make sure that
link |
you don't barge in and say
link |
we're going to internetize this thing
link |
and you have to think about
link |
the creators, you have to understand
link |
how they get distribution today
link |
who listens to how they make money today
link |
make sure that their business model works, that they understand
link |
I think it's back to doing something
link |
improving their products
link |
like feedback loops and distribution
link |
jumping back into terms of this
link |
fascinating world of
link |
recommender system and listening to music
link |
and using machine learning to analyze things
link |
do you think it's better
link |
correct me if I'm wrong but
link |
Spotify lets people pick what they listen to
link |
there's a discovery process but you kind of
link |
organize playlists
link |
is it better to let people pick
link |
what they listen to or recommend
link |
what they should listen to
link |
something like stations by Spotify
link |
that I saw that you're playing around with
link |
maybe you can tell me
link |
what's the status of that
link |
this is a Pandora style app
link |
that just kind of as opposed to you select
link |
the music you listen to
link |
it kind of feeds you
link |
the music you listen to
link |
what's the status of stations by Spotify
link |
the store is Spotify as we have grown
link |
has been that we made it more accessible
link |
to different audiences
link |
stations is another one of those where
link |
some people want to be very specific
link |
they actually want to start with heaven right now
link |
that needs to be very easy to do
link |
and some people or even the same person
link |
at some point might say
link |
I want to feel upbeat
link |
or I want to feel happy
link |
or I want songs to sing in the car
link |
so they put in the information
link |
at a very different level
link |
and then we need to translate that into
link |
so stations is a test to
link |
create like a consumption input vector
link |
that is much simpler where you can just tune it a little bit
link |
and see if that increases the overall reach
link |
but we're trying to kind of serve
link |
the entire gamut of super advanced
link |
so called music aficionados
link |
they love listening to music but it's not
link |
their number one priority in life
link |
they're not going to sit and follow every new release
link |
from every new artist
link |
to influence music at
link |
at a different level
link |
so we're trying, you can think of it as different
link |
products and I think when
link |
one of the interesting things
link |
to answer your question on
link |
if it's better to lift the user choose
link |
or to play I think the answer is
link |
the challenge when you
link |
when machine learning kind of came along
link |
there was a lot of thinking about
link |
what does product development mean
link |
in a machine learning context
link |
people like Andrew Eng for example
link |
when he went to Baidu
link |
he started doing a lot of practical machine learning
link |
went from academia and he thought a lot about this
link |
and he had this notion that
link |
you know product manager, designer
link |
and they used to work around this wireframe
link |
kind of describe what the product should look like
link |
was something to talk about when you're doing
link |
like a chat bot or a playlist
link |
what are you going to say like it should be good
link |
that's not a good product description
link |
so how do you do that and he came up with this notion
link |
the test set is the new wireframe
link |
and the job of the product manager is to source
link |
a good test set that is representative of what
link |
like if you say like I want to play this
link |
that is songsticing in the car
link |
job of the product manager to go and source
link |
like a good test set of what that means
link |
then you can work with engineering to have algorithms
link |
to try to produce that right
link |
so we try to think a lot about
link |
how to structure product development
link |
for machine learning
link |
age and what we discovered
link |
was that a lot of it is actually in the expectation
link |
you can go two ways so
link |
if you set the expectation with the user
link |
that this is a discovery product like Discover Weekly
link |
you're actually setting
link |
the expectation that most of what we show you will not be
link |
relevant when you're in the discovery
link |
process you're going to accept that
link |
actually if you find one gem every Monday
link |
that you totally love
link |
you're probably going to be happy
link |
even though the statistical meaning
link |
1 out of 10 is terrible or 1 out of 20
link |
is terrible from a user point of view
link |
because the setting was discovered it's fine
link |
can I start to interrupt real quick
link |
I just actually learned about Discover Weekly
link |
which is a Spotify
link |
it's a feature of Spotify that shows you
link |
cool songs to listen
link |
issue tracking I couldn't find it on my Spotify app
link |
it's in your library
link |
it's in the library it's in the list of live
link |
because I was like whoa this is cool I didn't know this existed
link |
and I tried to find it
link |
but I would show it to you
link |
and feed it back to our product teams
link |
the expectation there is
link |
basically you're going to
link |
discover new songs
link |
so then you can be quite adventurous
link |
in the recommendations you do
link |
another product called Daily Mix
link |
which kind of implies that these are only going to be your favorites
link |
9 out of 10 that is good
link |
and 9 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
link |
we learned over the years is about setting the right expectations
link |
you know algorithmically
link |
we would pick among things that feel very safe
link |
in your taste space
link |
or discover weekly we go kind of wild
link |
because the expectation is
link |
most of this is not going to
link |
so a lot of that a lot of to answer your question there
link |
we have some products where the whole point is
link |
that the user can click play put the phone in the pocket
link |
and it should be really good music for like an hour
link |
we have other products where
link |
you probably need to say like no
link |
and it's very interactive
link |
that makes sense and then the radio product
link |
the stations product is one of these like click play
link |
put in your pocket for hours
link |
that's really interesting so you're thinking of
link |
different test sets
link |
for different users
link |
and trying to create products that sort of
link |
optimize for those test sets
link |
that represents a specific set of
link |
one thing that I think is interesting
link |
we invested quite heavily in editorial
link |
in people creating playlists
link |
using statistical data
link |
and that was successful for us and then we also
link |
invested in machine learning
link |
and for the longest time
link |
within Spotify and within the rest of the
link |
industry there was always this narrative of
link |
humans versus the machine
link |
algo versus editorial
link |
and editors would say like well
link |
if I had that data if I could see your
link |
playlisting history and I made a choice
link |
for you I would have made a better choice
link |
and they would have because they're
link |
much smarter than these algorithms
link |
human is incredibly smart compared to our
link |
algorithms they can take
link |
culture into account and so forth
link |
the problem is that they can't make 200 million
link |
per hour for every user that logs
link |
in so the algo may be
link |
not as sophisticated but much more efficient
link |
so there was this contradiction
link |
but then a few years ago
link |
focusing on this kind of human in the loop
link |
thinking around machine learning
link |
and we actually coined an internal
link |
term for it called algotorial
link |
the combination of algorithms and editors
link |
where if we take a concrete
link |
of the editor this
link |
paid expert that we have
link |
there's really good at something like
link |
EDM something right there are two experts
link |
no one in the industry
link |
so they have all the cultural knowledge
link |
you think of them as the product manager
link |
let's say that you want to create
link |
you think that there's a product
link |
need in the world for something like songs to sing
link |
in the car or songs to sing in the shower
link |
I'm taking that example because it exists
link |
people love to scream songs in the car
link |
when they drive right
link |
so you want to create that product and you have this product manager
link |
who's a musical expert
link |
they create, they come up with a concept
link |
like I think this is a missing thing in humanity
link |
like a playlist called songs to sing in the car
link |
the framing, the image
link |
the title and they create a test set
link |
they create a group of songs
link |
like a few thousand songs out of the catalog
link |
that they manually curate
link |
that are known songs that are great to sing in the car
link |
and they can take like
link |
through romance into account they understand things
link |
that algorithms do not at all
link |
so they have this huge set of tracks
link |
then when we deliver that to you
link |
we look at your taste vectors
link |
and you get the 20 tracks that are songs to sing in the car
link |
so you have personalization
link |
and editorial input
link |
in the same process
link |
if that makes sense
link |
it makes total sense and I have several questions around that
link |
so first it is a little bit surprising
link |
that the world expert
link |
humans are outperforming
link |
songs to sing in the car
link |
maybe you could talk to that a little bit
link |
I don't know if you can put it into words but
link |
how difficult is this problem
link |
I guess what I'm trying to ask
link |
is there how difficult is it to encode
link |
the cultural references
link |
of the song, the artists
link |
all those things together
link |
can machine learning really not do that
link |
I mean I think machine learning is great
link |
at replicating patterns
link |
if you have the patterns
link |
but if you try to write
link |
a spec of what song is great
link |
to sing in the car definition is
link |
does it have many choruses to have been in movies
link |
it quickly gets incredibly complicated
link |
and a lot of it may not be
link |
in the structure of the song or the title
link |
it could be cultural references because
link |
it was a hasty one
link |
so the definition problems
link |
and I think that was the insight of Andrew Eng
link |
when he said that job of the product manager
link |
is to understand these things that
link |
algorithms don't and then
link |
define what that looks like
link |
and then you have something to train towards
link |
then you have kind of the test set
link |
but today the editors create this
link |
pool of tracks and then we personalize
link |
you could easily imagine that once you have this set
link |
you could have some automatic exploration
link |
of the rest of the catalog because then you understand
link |
and then the other side of it when machine learning
link |
does help is this taste
link |
how hard is it to construct
link |
a vector that represents the things
link |
an individual human likes
link |
this human preference
link |
you know music isn't like
link |
it's not like Amazon
link |
like things you usually buy
link |
music seems more amorphous
link |
like it's this thing
link |
that's hard to specify like
link |
if you look at my playlist what is the music
link |
that I love it's harder
link |
much more difficult to specify concretely
link |
to build a taste vector
link |
it is very hard in the sense that you need a lot of
link |
and I think what we found was that
link |
so it's not a stationary problem
link |
it changes over time
link |
through the journey of if
link |
you've done a lot of computer vision
link |
obviously I've done a bunch of computer vision
link |
in my past and we started
link |
kind of with the handcrafted heuristics
link |
this is kind of in the music this is this
link |
and if you consume this you probably like this
link |
we started there and we have some of that still
link |
then what was interesting about the playlist data
link |
was that you could find these latent things that
link |
wouldn't necessarily even make sense to you
link |
could even capture maybe cultural references
link |
because they co occurred things that
link |
that wouldn't have appeared
link |
kind of mechanistically either in the content
link |
I think the core assumption
link |
there are patterns
link |
in almost everything
link |
and if there are patterns
link |
these embedding techniques are getting better and better
link |
now as everyone else
link |
kind of deep embeddings where you can encode binary
link |
values and so forth
link |
is interesting is this process
link |
to try to find things that
link |
necessarily you wouldn't actually have
link |
guest so it is very
link |
hard in an engineering
link |
sense to find the right dimensions
link |
it's an incredible scalability
link |
problem to do for hundreds of millions
link |
of users and to update it every day
link |
in theory embeddings isn't that
link |
complicated the fact
link |
that you try to find some principle components
link |
or something like that dimensionality reduction
link |
and so forth so the theory I guess is easy
link |
because it's very very hard
link |
engineering challenge but fortunately we have
link |
some amazing both research and
link |
engineering teams in this space
link |
I mean it's similar I deal with the autonomous
link |
vehicle space is the question is
link |
how hard is driving
link |
basically the question is of
link |
I would imagine works well
link |
so there's a bunch of questions that arise then
link |
preferences does your taste vector
link |
different moods and
link |
absolutely so how does that
link |
is it is it possible to take that
link |
consideration or do you just leave
link |
that as a interface
link |
problem that allows the user to just control
link |
it so when I'm looking for a work
link |
out music I kind of specify it
link |
by choosing certain playlists
link |
doing certain search yeah
link |
so that's a great point and back to the product
link |
development you could try
link |
to spend a few years trying to predict
link |
which mood you're in automatically when you open
link |
Spotify or you create a tab
link |
which is happy and sad right and you're going to be
link |
right 100% of the time with one click
link |
now probably much better to let
link |
the user tell you if they're happy or sad
link |
or if they want to work out on the other hand
link |
if your user interface become 2000
link |
tabs you're introducing so much friction
link |
so no one will use the product so then you have to
link |
get better so it's this
link |
thing where I think it maybe was
link |
I remember who coined it but it's
link |
called full tolerant UIs right to build a UI
link |
that is tolerant to being wrong
link |
and then you can be much
link |
less right in your
link |
in your algorithms so
link |
we had to learn a lot of that
link |
building the right UI that fits
link |
where the machine learning is
link |
discovery there which was
link |
by the teams during
link |
one of our hack days was this
link |
thing of taking discovery packaging
link |
into a playlist and saying that
link |
these are new tracks
link |
that we think you might like
link |
based on this and setting the right expectation
link |
made it a great product
link |
so I think we have this benefit that for example
link |
Tesla doesn't have
link |
we can change the expectation we can build
link |
a full tolerant setting it's very hard to be
link |
full tolerant when you're driving at
link |
100 miles per hour or something
link |
and we have the luxury of
link |
being able to say that
link |
of being wrong if we have the right
link |
different abilities to take more risk
link |
so I actually think the self driving
link |
problem is much harder
link |
it's much less fun
link |
people die exactly
link |
such a more fun problem because
link |
I mean failure is beautiful in a way
link |
at least exploration so it's a really fun
link |
reinforcement learning problem and the worst
link |
case scenario is you get these WTF tweets
link |
like how did I get this
link |
which is a lot better than the self
link |
what's the feedback that a user
link |
that a user provides
link |
into the system so
link |
you mentioned skipping
link |
what is like the strongest signal
link |
you didn't mention clicking
link |
so we have a few signals that are important
link |
so one of the benefits of music actually
link |
even compared to podcasts or
link |
movies is the object itself
link |
is really only about three minutes
link |
so you get a lot of chances to recommend
link |
and the feedback loop is
link |
every three minutes instead of every two hours
link |
or something so you actually get
link |
quite fast feedback and so you can see
link |
if people played through or if the
link |
which is the inverse of skip really
link |
that's an important signal on the other hand
link |
much of the consumption happens
link |
when your phone is in your pocket maybe you're running
link |
or driving or you're playing on a speaker
link |
and you're not skipping doesn't mean that you love that song
link |
it might be that it wasn't bad enough
link |
that you would walk up and skip
link |
so it's a noisy signal
link |
then we have the equivalent of the like
link |
which is you save it to your library
link |
that's a pretty strong signal of affection
link |
we have the more explicit signal
link |
of play listing like you took the time
link |
to create a playlist you put it in there
link |
there's a very little small chance that
link |
if you took all that trouble this is not
link |
a really important track to you
link |
what are the tracks it relates to so we have
link |
we have the play listing we have the like and then we have
link |
the listening or skip
link |
have very different approaches to all of them because
link |
at different levels of noise
link |
one is very voluminous but noisy
link |
and the other is rare but
link |
you can probably trust it
link |
yeah it's interesting because
link |
I think between those signals captures
link |
all the information you'd want to capture
link |
I mean there's a feeling
link |
for me that there's sometimes that I'll
link |
hear a song that's like yes this is
link |
you know this was the right song for the moment
link |
but there's really no way to express
link |
that fact except by
link |
listening through it all the way
link |
and maybe playing it again at that time
link |
or something but there's
link |
no need for a button that says
link |
this was the best song could have
link |
heard at this moment well we're playing around
link |
with that with kind of the thumbs up
link |
concept saying like I really like this
link |
just kind of talking to the algorithm
link |
it's unclear if that's the best way
link |
for humans to interact maybe it is
link |
maybe they should think of Spotify
link |
as a person an agent sitting there
link |
trying to serve you and you can say like
link |
that's Spotify, good Spotify
link |
right now the analogy we've had is more
link |
you shouldn't think of us
link |
we should be investable and the feedback
link |
you work for yourself you do a playlist
link |
because you think is great and we can learn from that
link |
it's kind of back to Tesla
link |
how they kind of have this shadow mode
link |
we sit in what you drive
link |
we kind of took the same analogy
link |
we sit in what you playlist
link |
and then maybe we can offer you an autopilot
link |
we can take over for a while or something like that
link |
and then back off if you say like
link |
that's not good enough
link |
but I think it's interesting to figure out
link |
what your mental model is
link |
if Spotify is an AI that you talk to
link |
which I think might be a bit too abstract
link |
for many consumers
link |
or if you still think of it as
link |
and depends on the device it's running on
link |
which brings us to
link |
so I have a lot of the Spotify listening I do
link |
things on devices I can talk to
link |
whether it's from Amazon, Google
link |
what's the role of Spotify in those devices
link |
how do you think of it differently than
link |
on the phone or on the desktop
link |
there are a few things
link |
to say about the first of all
link |
it's incredibly exciting they're growing like crazy
link |
especially here in the US
link |
a consumer need that I think is
link |
you can think of it as
link |
just remote interactivity
link |
you can control this thing from
link |
across the room and it may
link |
feel like a small thing but it turns out
link |
that friction matters to consumers
link |
being able to say play, pause
link |
and so forth from across the room
link |
is very powerful so basically
link |
you made the living room interactive
link |
what we see in our data is that
link |
the number one use case for these speakers
link |
is music, music and podcast
link |
fortunately for us it's been important
link |
to these companies to have those use case
link |
covered so they want to Spotify on this
link |
we have very good relationships with
link |
and we're seeing tremendous
link |
what I think is interesting
link |
it's already working
link |
we kind of had this
link |
many years ago back when we started
link |
using Sonos if you went through all the
link |
trouble of setting up your Sonos system
link |
you had this magical experience where you had
link |
all the music ever made in your living room
link |
made this assumption that
link |
at home everyone used to have a CD player at home
link |
but they never managed to get their files
link |
working in the home having this network attached
link |
storage was too cumbersome for most consumers
link |
so we made the assumption that
link |
the home would skip from the CD all the way to
link |
where you would buy the steering wheel without
link |
all the music built in that took longer than we
link |
thought but with the voice speakers that was the
link |
unlocking that made kind of the connected speaker
link |
happen in the home
link |
it really exploded and
link |
we saw this engagement that we
link |
predicted would happen
link |
what I think is interesting though is where it's going
link |
right now you think of them as voice speakers
link |
but I think if you look at
link |
Google I.O. for example
link |
they just added a camera
link |
to it where when the alarm goes off
link |
hey Google stop you can just
link |
so I think they're going to think more of it as
link |
as an assistant truly an assistant
link |
and an assistant that can see you
link |
is going to be much more effective than a blind
link |
assistant so I think these things will
link |
morph and we won't necessarily think of them as
link |
quote unquote voice speakers anymore
link |
access to the internet in the home
link |
but I still think that
link |
the biggest use case for those will be
link |
for that reason we're investing heavily in it
link |
and we built our own NLU stack
link |
the challenge here is
link |
how do you innovate in that world
link |
it lowers friction for consumers
link |
but it's also much more constrained
link |
you have no pixels to play with
link |
in an audio only world
link |
it's really the vocabulary that is the interface
link |
so we started investing
link |
and playing around quite a lot with that
link |
trying to understand what the future will be
link |
of you speaking and gesturing
link |
and waving at your music
link |
and actually you're actually nudging closer
link |
to autonomous vehicle space
link |
because from everything I've seen
link |
the level of frustration people experience
link |
upon failure of natural language
link |
understanding is much higher
link |
than failure in other context
link |
people get frustrated really fast
link |
so if you screw that
link |
experience up even just a little bit
link |
they give up really quickly
link |
and I think you see that in the data
link |
while it's tremendously successful
link |
the most common interactions
link |
the things where if you compare it to taking up your phone
link |
unlocking it, bringing up the app and skipping
link |
it was much lower friction
link |
for longer more complicated things
link |
can you find me that song about people still bring up their phone
link |
and search and then play it on their speaker
link |
so we tried again to build a fault tolerant UI
link |
where for the more
link |
complicated things you can still pick up your phone
link |
have powerful full keyboard search
link |
and then try to optimize
link |
for where there is actually lower friction
link |
and try to, it's kind of like the
link |
test autopilot thing
link |
you have to be at the level where
link |
you're helpful if you're too smart
link |
and just in the way people are going to get frustrated
link |
I'm not obsessed with where it happens
link |
it's just a good song but let me mention that as a use case
link |
because it's an interesting one
link |
I've literally told
link |
I don't want to say the name of the speaker
link |
because when people are listening to it
link |
it'll make their speaker go off
link |
but I talk to the speaker
link |
Stairway to Heaven
link |
but a large percentage of the time plays the wrong
link |
Stairway to Heaven
link |
it plays some cover of the
link |
that part of the experience
link |
I actually wonder from a business perspective
link |
the Spotify control
link |
that entire experience
link |
it seems like the NLU
link |
stuff is controlled by the speaker
link |
and then Spotify stays
link |
at a layer below that
link |
it's a good and complicated question
link |
some of which is dependent
link |
partners so it's hard to comment on the specifics
link |
the question is the right one
link |
if you can't use any of the personalization
link |
we know which Stairway to Heaven
link |
and the truth is maybe for one person
link |
it is exactly the cover that they want
link |
to be very frustrated
link |
we default to the right version
link |
but you actually want to be able to do the cover
link |
for the person that just played 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
link |
where you have the
link |
the ASR and this thing called the end best list
link |
or the end best guesses
link |
here and then the personalization comes in at the end
link |
you actually want the personalization to be here
link |
when you're guessing about what they actually meant
link |
we're working with these partners
link |
and it's a complicated
link |
it's a complicated thing
link |
so first of all you want to be very
link |
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
link |
can understand enough but not too much and so forth
link |
it's really the trick is that
link |
it's like a business driven relationship
link |
where you're doing product development across companies
link |
which is really complicated
link |
but this is exactly why we built our own NLU
link |
so that we actually
link |
can make personalized guesses
link |
because this is the biggest frustration
link |
from a user point of view they don't understand
link |
about ASRs and end best lists
link |
and business deals they're like how hard can it be
link |
I've told this thing 50 times
link |
this version and still it plays the wrong thing
link |
so we try to take that user approach
link |
if the user is not going to understand
link |
the implications of business we have to
link |
let's talk about sort of a complicated subject
link |
the idea sort of of
link |
I saw as of August 31
link |
over 11 billion dollars
link |
were paid to rights holders
link |
and further distributed
link |
to artists from Spotify
link |
so a lot of money is being paid to artists
link |
the whole time as a consumer for me
link |
when I look at Spotify
link |
I'm not sure I'm remembering
link |
correctly but I think you said exactly how I feel
link |
which is this is too good to be true
link |
when I started using Spotify
link |
I assumed you guys would go bankrupt
link |
it's like this is too good
link |
a lot of people did
link |
so one question I have
link |
is sort of the bigger question
link |
how do you make money in this complicated world
link |
how do you deal with the relationship
link |
with record labels
link |
you essentially have the task
link |
cats but like rich
link |
have the task of paying artists enough
link |
and paying those labels enough
link |
and still making money in the internet
link |
space where people are not willing to pay
link |
hundreds of dollars a month
link |
how do you navigate the space
link |
that's a beautiful description, hurting rich cats
link |
I've never heard that before
link |
very complicated and I think
link |
actually betting against Spotify has been statistically
link |
a very smart thing to do
link |
just looking at the
link |
line of roadkill in music streaming services
link |
I think if I understood the complexity
link |
when I joined Spotify
link |
fortunately I didn't know enough
link |
the music industry to understand the complexities
link |
because then I would have made a more rational guess
link |
that it wouldn't work
link |
so ignorance is bliss
link |
there have been a few distinct challenges
link |
one of the things that made it work at all
link |
was that Sweden and the Nordics
link |
no risk for labels to try this
link |
I don't think it would have worked
link |
so that was the initial condition
link |
then we had this tremendous
link |
challenge with the model itself
link |
we're pirating but for the people who bought
link |
a download or a CD
link |
the artist would get
link |
all the revenue for all the future plays
link |
then right so you got it all up front
link |
whereas the streaming model was like
link |
almost nothing they won almost nothing they too
link |
and then at some point this curve
link |
of incremental revenue
link |
would intersect with your day one payment
link |
and that took a long time to play out
link |
they understood that but on the artist side
link |
it took a lot of time to understand
link |
that actually if I have a big hit that is going to be played
link |
for many years this is a much better model
link |
because I get paid based
link |
on how much people use the product
link |
not how much they thought they would use it day one
link |
so it was a complicated model to get across
link |
but time helped with that
link |
the revenues to the music industry actually are bigger
link |
it's gone through this incredible dip and now they're back up
link |
and so we're very proud
link |
so there have been distinct problems
link |
I think when it comes to the
link |
we have taken the painful approach
link |
some of our competition at the time
link |
looked at other companies and said
link |
if we just ignore the rights
link |
we get really big really fast
link |
we're going to be too big for the
link |
they're not going to kill us
link |
we're going to take that approach
link |
we went legal from day one
link |
and negotiated and negotiated
link |
it was very slow, very frustrating
link |
we were angry at seeing other companies
link |
taking shortcuts and seeming to get away with it
link |
it was this game theory thing
link |
where over many rounds of playing the game
link |
this would be the right strategy
link |
clearly there's a lot of frustrations
link |
at times during renegotiations
link |
there is this weird trust
link |
we've never screwed them, they've never screwed us
link |
it's ten years but
link |
there's this trust in like
link |
they know that if music doesn't get really big
link |
if lots of people do not want to listen
link |
to music and want to pay for it
link |
Spotify has no business model
link |
so we actually are incredibly aligned
link |
other companies have other business models
link |
where even if they made new music
link |
no money for music
link |
they'd still be profitable companies
link |
and I think the industry sees that
link |
we are actually aligned
link |
so there is this trust
link |
that allows us to do
link |
product development even if it's scary
link |
the free model itself was an incredible risk
link |
for the music industry to take
link |
that they should get credit for
link |
now 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
link |
I think hurting a cat sounds a bit
link |
dismissive? no every cat matter
link |
they are all beautiful and very important
link |
exactly they've taken a lot of risks
link |
and certainly it's been frustrating on both sides
link |
it's really like playing
link |
it's game theory if you play the
link |
if you play the game many times
link |
then you can have the statistical outcome
link |
that you bet on and it feels very painful
link |
when you're in the middle of that
link |
I mean there's risk there's trust
link |
there's relationships from
link |
just having read the biography
link |
similar kind of relationship were discussed
link |
in iTunes the idea of selling
link |
a song for a dollar was very uncomfortable
link |
it was the same kind of thing it was trust
link |
it was game theory
link |
as a lot of relationships that had to be built
link |
terrifyingly difficult
link |
that Apple could go through a little bit
link |
because they could afford for that process
link |
for Spotify it seems terrifying because
link |
I think a lot of it comes down to
link |
honestly Daniel and his tenacity
link |
in negotiating which seems like an impossible
link |
he was completely unknown and so forth
link |
and also the reason that
link |
I think game theory is probably the best way to think about it
link |
you could straight go straight for this like Nash
link |
equilibrium that someone is going to defect
link |
or you play many times
link |
and you try to actually go for the top left
link |
the corporations sell
link |
is there any magical
link |
reason why Spotify
link |
so a lot of people have tried to do
link |
what Spotify tried to do
link |
and Spotify has come out
link |
well so the answer is that
link |
there's no magical reason because I don't believe in magic
link |
but I think there are
link |
and I think some of them are that
link |
a lot of what we actually do
link |
Spotify model is very complicated
link |
they've looked at the premium model
link |
like you can charge 9.99
link |
for music and people are going to pay
link |
but that's not what happened
link |
actually when we launched the original mobile product
link |
everyone said they would never pay
link |
what happened was they started on the free product
link |
and then their engagement
link |
grew so much that eventually
link |
they said maybe it is worth
link |
your propensity to pay gross with your engagement
link |
so we had this super complicated
link |
business model where you operate
link |
two different business model advertising and premium
link |
products at the same time
link |
and I think that is hard to replicate
link |
I struggle to think of other companies
link |
that run large scale advertising
link |
and subscription products at the same time
link |
so I think the business model is actually
link |
much more complicated than people think it is
link |
some people went after just the premium part
link |
without the free part and ran into a wall
link |
where no one wanted to pay
link |
some people went after just
link |
music should be free just ads
link |
which doesn't give you enough revenue
link |
for the music industry
link |
so I think that combination is
link |
kind of opaque from the outside
link |
so maybe I shouldn't say it here and reveal the secret
link |
but that turns out to be hard
link |
than you would think
link |
so there is a lot of 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 |
so let's call it brilliant
link |
yeah when the books are written it will be brilliant
link |
mentioned that your philosophy is to
link |
so how will the music streaming
link |
and music listening
link |
world change over the next
link |
10 years, 20 years
link |
you look out into the far future
link |
what do you think?
link |
I think that music
link |
and for that matter audio
link |
podcast audio books
link |
I think it's one of the few
link |
I think there is no good reason to me
link |
why it shouldn't be at the scale
link |
of something like messaging or social networking
link |
I don't think it's a niche thing
link |
to listen to music or news or something
link |
so I think scale is obviously one of the things
link |
that I really hope for
link |
I hope that it's going to be
link |
billions of users, I hope eventually
link |
everyone in the world gets access to
link |
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
link |
what I'm hoping is back to this
link |
software tool chain
link |
sometimes internally
link |
I make this analogy to
link |
text messaging was also based on
link |
in the area of mobile carriers
link |
you had the SMS, the 140 character
link |
because everyone agreed on the standard
link |
so as a consumer you got a lot of distributions
link |
and interoperability but it was a very constrained format
link |
and when the industry
link |
wanted to add pictures to that format
link |
to do the MMS, I looked it up
link |
and I think it took from the late 80s to early 2000
link |
this is like a 15, 20 year product cycle
link |
to bring pictures into that
link |
now once that entire
link |
of creation and consumption got wrapped
link |
in one software stack
link |
within something like Snapchat or WhatsApp
link |
they added disappearing messages
link |
then two weeks later they added stories
link |
the pace of innovation when you're on one software stack
link |
affect both creation and consumption
link |
I think it's going to be rapid
link |
so with these streaming services
link |
we now for the first time in history
link |
have enough I hope people
link |
on one of these services
link |
actually whether it's Spotify or Amazon
link |
or Apple or YouTube
link |
and hopefully enough creators
link |
can do the format again
link |
and that excites me
link |
I think being able to change these constraints from 100 years
link |
do something interesting
link |
I really hope it's not just going to be there
link |
iteration on the same thing
link |
for the next 10 to 20 years as well
link |
yeah changing the creation
link |
of music or creation of audio
link |
or creation of podcasts
link |
is a really fascinating possibility
link |
I myself don't understand
link |
what it is about podcasts that's so intimate
link |
it just is I listen to a lot of podcasts
link |
I think it touches
link |
on a human on a deep
link |
human need for connection
link |
that people do feel like they're
link |
to when they listen I don't understand
link |
what the psychology that is
link |
but in this world is becoming
link |
more and more disconnected
link |
this is fulfilling a certain kind of
link |
empowering the creator as opposed
link |
to just the listener
link |
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
link |
for our teams to work on podcasts
link |
whether you think like I probably do
link |
that it's something biological about
link |
perceiving to be in the middle of the conversation
link |
that makes you listen in a different way
link |
it doesn't really matter people seem to perceive it
link |
there was this narrative for a long time that
link |
you know if you look at video
link |
it's something kind of in the foreground it got shorter
link |
and shorter and shorter because of financial
link |
pressures and monetization and so forth
link |
and eventually at the end there's almost like
link |
people just screaming something
link |
I'm really I feel really good about
link |
you could have interpreted that as people have
link |
no attention span anymore
link |
they don't want to listen to things they're not interested
link |
like you know people are getting dumber
link |
but then podcast came along and it's almost like no
link |
no the need still existed
link |
once but maybe maybe it was
link |
the fact that you're not prepared to look at your
link |
phone like this for two hours
link |
but if you can drive at the same time it seems like
link |
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
link |
that podcast is actually long form
link |
it gives me a lot of hope for
link |
for humanity that people seem really interested
link |
in hearing deeper more complicated
link |
conversations this is
link |
I don't understand it
link |
it's fascinating so the majority
link |
for this podcast listen to the whole thing
link |
this whole conversation
link |
we've been talking for an hour and 45 minutes
link |
I mean most people will be listening
link |
to these words I'm speaking right now
link |
you wouldn't have thought that 10 years ago
link |
where the world seemed to go
link |
that's very positive I think
link |
that's really exciting and empowering the creator
link |
there's as really exciting
link |
do you also have a passion for
link |
just mobile in general
link |
how do you see the smartphone world
link |
and just everything that's on the move
link |
internet of things and so on
link |
changing over the next 10 years
link |
I think that one way to think about it is that
link |
moving out of these
link |
multi purpose devices
link |
the computer we had in the phone
link |
specific purpose devices
link |
and it will be ambient that
link |
at least in my home
link |
you just shout something at someone
link |
and there's always one of these speakers close enough
link |
you start behaving differently
link |
it's as if you have the internet ambience
link |
ambiently around you and you can ask it
link |
so I think computing
link |
will kind of get more integrated
link |
and we won't necessarily think of it
link |
connected to a device in the same way
link |
I don't know the path to that maybe
link |
we used to have these
link |
and then we partially replaced that with
link |
the laptops and left
link |
desktop at home and at work and then
link |
we got these phones and we started leaving
link |
the laptop at home for a while and maybe
link |
for stretches of time
link |
you're going to start using the watch and you can leave
link |
your phone at home like for a run
link |
or something and we're on this
link |
progressive path where
link |
what is happening with
link |
interaction paradigm that doesn't
link |
as large physical devices so I definitely
link |
think there's a future where you can have
link |
your watch and you can do
link |
a lot of computing
link |
it's going to be this binary
link |
thing, I think it's going to be like many of us
link |
still have a laptop, we just use it less
link |
and so you shift your consumption over
link |
I don't know about
link |
AR glasses and so forth
link |
I'm excited about it, I spent a lot of time in that area
link |
but I still think it's quite far away
link |
AR, VR, all of that
link |
Yeah, VR is happening and
link |
working, I think the recent
link |
Oculus Quest is quite impressive
link |
I think AR is further away
link |
at least that type of AR
link |
your phone or watch or glasses understanding
link |
where you are and maybe what you're looking at
link |
and being able to give you audio cues about that or you can say
link |
like what is this and it tells you what it is
link |
might happen, you use
link |
your watch or your glasses as
link |
a mouse pointer on reality
link |
I think it might be a while before
link |
I might be wrong, I hope I'm wrong but I think it might be a while
link |
before we walk around with these big lab glasses
link |
that project things
link |
I agree with you, it's actually really
link |
difficult when you have to
link |
understand the physical world
link |
I lied about the last question
link |
because I just thought
link |
and my favorite topic which is the movie
link |
whether it's part of Spotify or not
link |
I don't know if you've seen the movie Her
link |
the primary form of interaction
link |
and the connection with another entity
link |
that you can actually have a
link |
relationship with or fall in love with
link |
based on voice alone
link |
do you think that's possible first of all
link |
based on audio alone to fall in love with somebody
link |
somebody or well yeah let's go
link |
with somebody, just have a relationship
link |
based on audio alone
link |
and second question to that
link |
can we create an artificial intelligence system
link |
allows one to fall in love
link |
with it and her, him
link |
so this is my personal answer
link |
speaking for me as a person
link |
the answer is quite unequivocally
link |
well we just said about podcasts
link |
and the feeling of being in the middle of a conversation
link |
if you could have an
link |
and we just said that feels like a very personal
link |
setting so if you walk around with these
link |
headphones and this thing, you're speaking with this
link |
thing all of the time that feels like it's in your brain
link |
it's gonna be much easier to fall in love with
link |
than something that would be on your screen
link |
I think that's entirely possible
link |
and then from the, you can probably answer this better than me
link |
but from the concept of
link |
if it's going to be possible
link |
to build a machine
link |
that can achieve that
link |
I think whether you
link |
think of it as, if you can fake it
link |
the philosophical zombie that simulates it enough
link |
or it somehow actually is
link |
it's only a question, if you ask me about time
link |
I'd have to give an answer but if you say I've given
link |
some half infinite time
link |
absolutely, I think it's just
link |
atoms and arrangement
link |
well, I personally think that love
link |
is a lot simpler than people think
link |
so we started with true romance
link |
I don't see a better place to end
link |
Gustav, thanks so much for talking today
link |
thank you so much, it was a lot of fun