"We hate this stupid man versus machine dichotomy."
As conversations about machine learning occurred on the main stage at Gigaom's Structure Data conference in New York, Brian Whitman spoke in a side suite of the Chelsea Piers event complex about how the Echo Nest uses "machine listening" to predict what people what to hear. Billboard caught up with the co-founder and Chief Technology Officer of the Echo Nest to discuss why Pandora "freaks" him out, "thinking deeply about albums" and more.
Billboard: The Echo Nest has been in the position to see how users interact with all sorts of different music services, how will that information influence the decisions you make at Spotify?
Brian Whitman: Obviously, the more we know about the listener, the better everybody is. Any smart service would want to know all sorts of tracking data across everywhere you go. It’s not just for them obviously, it’s for you to know more about your taste. I think the platform will be more of a big issue where soon a lot of your music listening is probably going to come from one service. From my perspective the control that I possibly as my single human can have to track you on youtube or soundcloud is pretty limited right now so I think that is an important thing.
What I'm very interested in, and this is something the Echo Nest has been doing for a while, is the music platform being an underlying service. You are paying your Spofify bill, but it powers all sorts of stuff.
So logging into other services or apps with your Spotify account? Which would help build your taste profile.
Working at Spotify now, you must be well aware of complaints about bad recommendations like, “You listened to Marc Bolan, you must like T. Rex”
On the record, I’m going to get rid of that one. I hate that one, I see it all the time.
In order to have a great personal recommendation tailored to a user you really have to have historical context for that user, right? If you are a brand new user, signing in for the first time, how do you solve that problem?
The way I would solve that is to get whatever contextual clues we can possibly get from you, there are all sorts of things you give away without even noticing, even the fact that you are using an iPhone instead of an Android we know a lot more about your music taste. It’s low level signals that definitely when combined tell you a lot.
I’m not going to guarantee you that your first use of a service that you’ve never told anything about [yourself] is going to give you an amazing recommendation, but I want it to be non-random. I want it to be better than just playing the Katy Perry's "Dark Horse," because thats what they all do now. It's going to be targeted to you because we are going to know something about you.
For example you have Beats, who took a unique approach to the on-boarding system, but I personally found myself disappointed. I told it what I liked and I still got an intro to LCD Soundsystem playlist, which I'm already a fan of, where is the inbetween?
I’m not sure where the inbetween is yet but I’m certainly going to be working on it. I want to find something that people are going to love and your not going to have that boring stuff that you would expect or what we call the WTF results. Its going to take a lot of work. Its already taken the Echo Nest nine years. We are pretty good at it, but we are always going to better and I’m really looking forward to making it really good with these guys.
As algorithms get better and users increasingly depend on them, do you think that leads to a homogenization of taste?
One thing, you call it algorithms but it’s a lot more than that. We are obviously doing a ton of computer stuff but it’s all based on what people are saying and choosing and that stuff. We hate this stupid man versus machine dichotomy.
Well, because man created…
Yes, man created the machine.
Even the stuff we do, yeah it’s a ton of computers but what it’s doing is reading peoples blog posts about bands. I dont want it ever to get to the point where it’s giving the same result to everybody even now with the Echo Nest and all the other services we power, that's not the case. Its going to be unique and tailored to each service.
Is it possible for the Echo Nest to create a really fantastic playlist based on an idea or concept in lyrics?
It is something we are working on quite a lot, I can say that much. It is a huge focus of interest for us.
Especially with playlists, you have Beats or Songza with an editorial team creating playlists, those playlists are static and maybe based on one person’s perception of what that theme means. Can you take that same theme and apply it to just me, my taste profile, my history and what I like and make it an infinite loop, like an artist based radio station? Is that possible?
It will be. Yes, for sure. That is a huge focus of discovery in the future. It comes back to the zero UI [user interface] contextual playlisting, it’s not just that you typed in Aerosmith or Big Star, we just know that at 10 a.m. you are listening to a certain kind of music, or you are using your phone or at work or at home, all of those signals are going to come in. And mood is a big part of it as well.
The ultimate interface for me personally is just, you hit play and you don't have to do anything else. Of course, if you want to keep choosing things as people do, that’s great, but for a lot of people it's still so confusing to use these devices and we want to fix that.
Particularly in the context of this conference which is all about using data and context to track and monitor all aspects of life and then use that to make us more productive or happier or fitter, what do you say to people who, especially in reference to music, call that creepy or a reduction of art?
I could see why they think that and I’ve also felt recommenders, we rely on them too much. I think the people who rely on Pandora, for example, that kind of freaks me out because it’s such a small amount of music they are listening to it’s almost just like top 40 radio all over again. With all the power that technology gives you, I would like to see people doing more with it then just listening to the same 20 songs over and over again.
For me it’s less creepy, it's a way to browse and discover things. It’s like a Google search, we’re not forcing anything on you its a way to explore.
So just like news filtering, if i’m only being recommended news that you already know i'm going to like, am I missing out on the world news story I maybe wouldn't have read otherwise if I hadn’t seen it in a newspaper. With music, am I not going to hear certain time periods or genres?
On both Rdio and Mog, there was the beautiful slider up top that we called adventurousness, it was our pioneering effort to make sure that people didn't just listen to the default stuff. On the average people yanked that slider to adventurous, of course these were early adopters, I would hope that some beautiful music interface of the future still allows people to customize and get themselves out of the bubble.
But only if they chose to.
I'm not going to put implants in their brain! They've got to ask.
Can you give a certain weights to certain artists? Say a label wants to pay Spotify to weight their artists more heavily in a recommendation algorithm?
We would call this editorial. It doesn't have to be a label doing payola and I wouldn't want to do that, but editorial is huge. You’ve got amazing music writers and critics embedded in a lot of services and you want to all those great human curated playlists for example. We already work with a lot of guys who are folding in curatorial input into our results. It’s not just machine or man, its some beautiful happy medium.
Have you worked on self monitoring or album completions? For example, how can I be certain I’ve listened to every single David Bowie song?
We have not worked on that. I would love to see that product. Another thing I really want to see, we’ve all seen similar artists, you go to T Rex and you see David Bowie. It’s a graph. People think that T. Rex and Bowie sound similar. But what if there is someone in the middle you haven't seen yet? I want to make sure you fill in your music education. I’m really into 70’s rock and I want to make sure I get through all of the influencers and all the songs for all of the artists.
You could build a huge web just for album recommendations.
The Echo Nest thinks deeply about albums now. We’ve done a lot of work there. I think we’ll announce that, yeah we have thought deeply about it. Hopefully you will see something there soon.
Tell me more about the Echo nest and Spotify Deal.
Everybody’s happy, that’s all I can say.
Being a co-founder, did you see a situation like this eventually happening? Did you think you’d eventually pick one partner, or that you’d always be a backend to multiple services. You had started to expand internationally recently.
I never really thought too far in the future, I just wanted to keep getting it done. We got lucky in that we chose a really good business plan for the time we were in. We chose to be an API instead of trying to make our own music service which probably wouldn't have ended well. The founders of the company are two scientists from MIT, we’re not hipster looking product guys. We just love music.