Anybody who follows me on social networks knows that I’m constantly listening to music while I work. You know that because I’m not shy about plugging the best of it through tweets, Facebook status updates, etc. I’m a self-appointed musical recommendation engine for all my social-friends. For example, at this very moment, I’m listening to the YouTube playback of Camera Obscura’s great 2006 song “I Need All the Friends I Can Get.”
For those who wonder about this “#kexp” hashtag that accompanies most of my music-recommendation tweets, that refers to a streaming Web public-radio station from Seattle that’s been my primary music-discovery engine since 2002. One of the core reasons I continue to listen is because the DJs pick the music they play. Another reason is because they have a deep, broad, and awesome station library. A third is that they all have great (and diverse) tastes.
I accept KEXP DJs’ “recommendations” by rarely “switching” to another station (or turning it off). I just stream with their stream night and day. And I hear the latest releases across many styles. And most of my music purchases follow from the best of what I hear on KEXP (follow my tweets and you’ll see what I get into).
As a listener ,the KEXP music-discovery experience is addictive because it’s driven by specific people (several of whom I’ve met in the flesh) who share my general tastes. Other people turn to totally or largely automated Web services such as Pandora, Spotify, and Beats Music for a similar reason, and that’s cool. Everybody grooves to a different music-consumption experience.
Though I’m the (remote) devotee of a next-generation “real” radio station, I know that the Pandoras of the world have the dominant business model going forward. Not everybody gets into the eclectic, category-spanning fusion of musical styles that my pet radio station plays. Most music fans in the future will prefer a continuously targeted experience driven by big data, machine learning algorithms, and predictive models. The primary reason for that is its scalability, coupled with its ability to micro-target every niche of music consumer down to the individual (after all, no two people groove to the exact same set of artists, songs, and styles all the time). Personalization is everything when it comes to one’s own “personal” music stash.
This recent New York Times article nicely lays out the algorithmic streaming-music service business model. Data scientists, not DJs in the old-fashioned sense, shape these experiences. Reflecting that trend, Spotify recently acquired Echo Nest, “a company that analyzes music consumption patterns and was founded by computer scientists from MIT.”
At the heart of this deal is a recommendation engine. Specifically, Echo Nest’s technology will be used to drive music-experience optimization and ad-placement optimization, which are the two make-or-break applications in any such service. This all comes down to, as the article states, giving Spotify’s “24 million users better suggestions about what songs to listen to.” Essentially, that’s predictive preference analysis, which depends on continuous historical analysis of “the arcane but valuable science of music data, examining what songs are being listened to by whom, and how.”
But to say these services are entirely “algorithmic” grossly distorts how they actually work. Human recommendations are deep in the mix, but they’ve aggregated, abstracted, and de-personalized to an astonishing degree. As the article notes, ” Music streaming companies like Spotify and Pandora are part of a broad category of online services that rely on technology to crowdsource recommendations. Whether for books on Amazon or films on Netflix, these companies use complex algorithms to comb through their users’ activity to suggest new purchases and products.”
As this discussion indicates, the new music-industry world order can’t be reduced to a simplistic “man vs. machine” struggle. The new order of online “recommendation engines” is one part algorithmic and one part social. Human DJs–a la KEXP’s John Richards, Cheryl Waters, and Kevin Cole–haven’t entirely left the picture–and probably won’t as long as listeners respond to curated streams supervised by named recommenders. But the unnamed human recommenders–the “crowdsources” of the new musical landscape–will play just as big a role in deciding the next song that plays out on our mobile device or in our car. And after a while we’ll notice the human DJs themselves frequently deferring to some algorithmically-ranked recommendation when deciding what next to play.
Or even letting the algorithm itself run the show while they take a bathroom break.