Avoiding The Pigeonhole In Music Discovery

Cymbal started as a way to make sense of the complex music ecosystem. With so many songs out there on the internet, Cymbal puts a spotlight on what really matters: The songs your friends can’t stop listening to — the songs the people you care about deeply identify with. Today, I want to talk more about the future, where music seems to be heading, and how I see Cymbal changing the way we listen.

There’s a battle going on in music technology right now: How should streaming services get you the right song, out of the tens of millions that exist? Some services use fine-tuned algorithms to recommend users songs that others with similar tastes liked; Others give huge, one-way platforms in the form of curated playlists or radio stations for influencers to share their respected tastes. Algorithmic recommendations, in large part, work by learning your tastes and trying to give you more of the same. This solution rests on the presumptions that 1. users want a lot of the same and that 2. streaming services are good at deciding what that same is. All of this is up for debate.

Ryan Dombal of Pitchfork recently discussed this topic with Ben Ratliff, author of Every Song Ever: Twenty Ways to Listen in an Age of Musical Plenty. Ryan writes,

“Though human beings now have instant access to an endless amount of music in every style imaginable, it can often feel like streaming services are hell bent on narrowing our perspective instead of expanding it. Whether their recommendations come from algorithms or actual people, the results merely flatter our tastes, leading us to what Ben Ratliff calls ‘bottomless comfort zones.’”

Ratliff goes on to discuss the problem more in-depth. “When I listen along the comfortable contours of what I know and who I’ve been, I don’t learn anything,” he states. “I don’t grow.” Think about the times when you came to love a new artist that sounds nothing like what you normally listen to. Aren’t these the most rewarding musical experiences? Current music recommendation services, Ratliff argues, seem to limit these opportunities by associating listeners with a single genre.

Ratliff and Dombal are hitting a key point, yet there are other issues with music recommendation algorithms. Since music is so deeply connected to the context in which it’s discovered, and because recommendation algorithms are by nature devoid of context, they can often feel empty. With a lack of diversity and personal connection, they become stale. Songs aren’t just validators of our tastes; They’re banks for memories, places, relationships.

In a scene from Almost Famous, Zoey Deschanel character leaves a box of records for her little brother, hoping to guide him through a musical awakening. P.

We came from the experience that so much of what makes a song or a recommendation meaningful is who it comes from. So we built Cymbal to not only be a means for people to share their favorite songs, but also to see what those who matter to them are enjoying. Follow the curators you trust the most and, in return, get the songs they feel are the best of the best at this moment. This information is really meaningful. With it, we can draw conclusions and make predictions that aren’t just based on what people listen to, but what they love. We’re set up to not only get the best possible profile of your musical taste, but also to create a system that solves Ratliff’s issue and create this “magical” listening experience he searches for.

Cymbal is built on deliberate sharing. When you decide to share a song on Cymbal, you are telling the world that this song, more than others, is the one you feel most connected to right now. Think about it this way: The songs in your Spotify/SoundCloud libraries are like all the pictures you have been tagged in on Facebook. Some you love, some you just like, some you might not even like, but are there because they are in an album or by an artist you do really like. The songs you decide to share on Cymbal, on the other hand, are your profile pictures. You chose them in the moment to best represent you as a listener. This is the difference between passive listening and active listening. Making judgments on musical taste must be made from data on active listening. Otherwise the data is muddy.

Jack Black’s character in High Fidelity helps a customer navigate a sea of records by recommending only the ones that are most important to him. In this scene, he holds up one of his all time favorite Dylan records, “Blonde On Blonde.”

Collecting data on deliberate sharing only solves half of the problem. It also really matters what you do with that data. How do you use it to guide the user into a listening experience that feels real? One that feels like natural growth and avoids Ratliff’s feared “bottomless comfort zone”. To this end, Ratliff explains:

“What I mean about listening with purpose is following trails that might lead us to magical places that we haven’t noticed before.”

Meaning: Recommendations should come from a new kind of algorithm that doesn’t place listeners in a box they’ll eventually grow out of. And what if those recommendations came from intentional, self-defining data, a type that many streaming services don’t have?

To illustrate how much better this could get with more contextual data, let’s look at a theoretical example. Imagine there is is a Cymbal user (let’s call him “Hugo”) who loves listening to old-school rap. When looking at Hugo’s profile, you see A Tribe Called Quest, Wu-Tang, Ice Cube, etc. A “bottomless comfort zone” recommendation service would recommend similar artists. Maybe Outkast, De La Soul, or Tupac. Chances are, Hugo would really like these artists. Now imagine we find another user (let’s call her “Alison”) who also began by posting the same subset of artists. We find that Hugo and Alison are on similar musical paths. However, as time passes, Alison’s taste begins to diverge from Hugo’s. Alison begins listening to other types of music. Maybe she begins listening to more experimental hip-hop like Shabazz Palaces, Deah Grips, or Odd Future. Maybe recommending these artists to Hugo is the best idea, regardless of the fact that they lie outside his normal listening habits. Maybe this is the path that Hugo wants to take and just doesn’t know it yet.

Part of what this discussion tells us is that the future of music discovery is about guiding a user down new paths that will help them naturally grow as a listener. The way to find these paths comes back to the core engine of Cymbal: People. Music is a powerful representation of personality, and personalities change over time. There is a certain path you took, both as a person and as a listener, to arrive at your current preferences. The magical moments Ratliff speaks of are when you made the big leaps. Even more importantly, as you are changing, there is someone out there who is changing in a very similar way.

This idea is not new, but it’s possible that it has not been applied to music yet. Stat god Nate Silver wrote about a similar theory in respect to basketball. Their algorithm is called CARMELO. “CARMELO identifies similar players throughout modern NBA history and uses their careers to forecast the current player’s future,” he writes. He identified three steps in the process,

1. Define the player’s skills.
2. Identify comparable players
3. Make a projection.

Maybe it’s time we apply this theory to music discovery. Maybe it’s time we start focusing on steps 2 and 3 and not just step 1. With deliberate sharing, and social data, Cymbal is in a position to make the comparisons between people mean more, and to make the projections that substantiate our recommendations grow as you do. Recommendations that connect people who are growing and changing in a similar fashion. Recommendations that help people “listen with purpose.”

Good tech will always rely on good algorithms, but I believe it is important to think about these algorithms as a way to not only figure out what users want, but also figure out what they are going to want. That is when they really become magical.

By Gabe Jacobs, Co-founder of Cymbal