X

Music and technology: The triumph of the human ear

Great artists deserve your attention. Technology can help.

Tim Westergren The Music Man
Tim Westergren founded Pandora in 2000 with the Music Genome Project, the DNA that makes the radio service's recommendation algorithm work. A musician himself, Westergren guides the company's overall strategy.
Tim Westergren
3 min read

When it comes to the impact of technology, the music industry is an odd bird. On the one hand, it has been utterly reinvented by technology. Recording software has democratized music creation and online retail has replaced in-store distribution. With a laptop, a decent microphone and a $35 annual membership to CDBaby, an artist can do it all themselves. But when it comes to the third, and perhaps most important leg, promotion, it has remained stubbornly resistant to technology.

In the late '90s, many working musicians saw in the Web a chance to finally bring a democratic meritocracy to the industry. With such ease of access and distribution, it seemed so inevitable. The combination of infinite virtual shelf space and the magic of algorithmic recommendation engines were going to allow all music to realize its potential. But that didn't happen. Instead, it was business as usual. Today, if you look at most online services, only a tiny sliver of any catalogue gets meaningful usage. It's still feast or famine and it's still the same concentration of attention on the well-established artists, with the addition of the periodic YouTube breakout artist. In effect, we're right where we started. The Tower Records endcap was replaced by the iTunes homepage.

tim-westergren-inset-6098.jpg

And there's a simple reason for this failure. The vast majority of technology-based solutions have one great big Achilles heel when it comes to promotion: they rely on existing popularity. Algorithms are driven by data. Data yields patterns, and patterns yield recommendations. In a world of data, if you're a new artist and have no data, you can never be part of a pattern. If you're not part of a pattern, you basically don't exist. Algorithms, like robots trying to dance, become clumsy and useless when trying to handle new artists. When was the last time a collaborative filtering engine suggested great new music for you? The Web has thus become a self-perpetuating popularity machine. Sure, there are the occasional come-from-nowhere viral phenoms, but they are the exceptions that prove the rule. For the listener, that means a dearth of discovery. And for most artists, it means still hoping for that lucky break to pull them into the aperture of public attention (where they can start generating data for the pattern recognition machine).

Ironically, at Pandora we have attacked this problem by going back to good old-fashioned human beings. Only a human ear can "hear" new music and know where it belongs. For fifteen years now, we've had trained musicians listening to songs individually, scoring hundreds of attributes per song in order to create a foundation for interconnecting songs based on their intrinsic properties -- obviating the need for data, and creating a genuinely democratic form of discovery. Our Music Genome Project (MGP) is literally unaware of popularity when it matches one song to another. Quite a contrast from an algorithm that literally doesn't know the sound of what it's recommending.

Once the MGP places the new song in the proper neighborhood, then data can do its magic, and instantly amplify that song's presence to an already targeted audience. Pandora creates the best of both worlds. It uses humans as the onramp for hundreds of thousands of songs, then unleashes the power of technology at scale when those songs are played for a gigantic audience of people predisposed to like them.

The result? Over 99 percent of our collection plays EVERY MONTH. There are now nearly 20K artists that have over 10,000 unique station creators on Pandora. That's striking when you consider that over 80 percent of the artists on Pandora have never played on any form of broadcast radio, ever.

In the long run, I think this "human" approach to recommendation will become a standard for all products. It requires an immense, and seemingly irrational investment in people, but it's the only way to create a perfectly, personalized market. Where every product -- every artist, including the lesser known -- finds the audience they truly deserve.

In a world of high-performance algorithms, data science and supercomputers, it's refreshing to think that a human ear holds the key to our own perfect song.

Photo courtesy of Pandora