Just as computers cannot yet create powerful and imaginative art or prose, they cannot truly appreciate music. And arranging a poignant or compelling music playlist takes a type of insight they don’t have—the ability to find similarities in musical elements and to get the emotional resonance and cultural context of songs. For all the progress being made in artificial intelligence, machines are still hopelessly unimaginative and predictable. This is why Apple has hired hundreds of people to serve as DJs and playlist makers, in addition to the algorithmic recommendations it still offers.
More recently, algorithms have begun producing playlists that can feel a lot more nuanced and tailor-made. The world’s biggest streaming service, Spotify, which has more than 75 million users, is pushing the state of the art, using vast amounts of data to make personalized recommendations.
Spotify’s deep-learning system still has to be trained using millions of example songs, and it would be perplexed by a bold new style of music. What’s more, such algorithms cannot arrange songs in a creative way. Nor can they distinguish between a truly original piece and yet another me-too imitation of a popular sound. (Spotify’s Chris) Johnson acknowledges this limitation, and he says human expertise will remain a key part of Spotify’s algorithms for the foreseeable future.
Though some consider human curation to be elitist, I feel music fans and listeners welcome and crave it. Who doesn’t enjoy a trusted source giving suggestions of cool new music to discover? It’s been the secret of success for certain radio shows, record store clerks, magazine music reviewers, and music blogs. The whole mixtape phenomenon is built on it. My SoundCloud stream is built on it. Basically, if you’re into discovery, you’re into the trusted recommendation … or, in modern industry-speak, “curation”.
Despite my love of human recommendations, I am genuinely curious about Spotify’s algorithmic ‘Discovery’ playlist and want to dig more into it. (There are some technical issues I have with Spotify’s OS X app that keep me from using it more which I won’t go into here.) The team at Spotify seem very confident in what the technology is able to do, and anything that encourages listeners to check out new music is all right by me. But I can’t help but wonder if a human / tastemaker guided algorithm – a mixture of computer recommendation and ‘music fan’ supervision – might be the way to go. From this article, it sounds like this is where we are headed.
An area that I find frustratingly overlooked is the realm of the Pandora-like ‘sounds like’ radio stations. These don’t work for me, not on any of the services, and this ‘radio’ would be my most accessed feature if they did. Pandora drove me crazy because (as an example) I’d program a Joy Division station, and then would hear “Love Will Tear Us Apart” every single time I chose it, but didn’t necessarily want to give it a ‘thumbs down’ and banish the song from its repertoire. I might want to hear it now and then … but not every single time. Of course, I’m not picking solely on Pandora here as none of the services get this radio feature right. If I create a ‘station’ based off Brian Eno’s “Lizard Point” – a beatless, droning composition opening his album Ambient 4: On Land – I’m sure all the services will give me large doses of ’70s art rock instead of the ambient music I’m looking for.
Apple Radio – the one pre-dating Apple Music – tried to address this to a degree with a slider allowing the user to choose if he/she wanted to hear more of the ‘hits’ or, with the slider all the way to the right, to choose an adventurous discovery-oriented path. It wasn’t perfect, but the concept was solid and I would have loved for Apple to fine tune it rather than ditching it altogether in Apple Music at present. Regardless, we’re a long way from computers getting it right (and potentially achieving music snob A.I.) but it’s a fascinating study to see the aims and attempts to move us closer. And, once again, all this effort going into helping listeners find new independent artists is a terrific thing.
Relying on algorithms to serve up music can be just as pointless — and even, anachronistic — as automated FM programming from two decades ago. The current metadata is too broad yet not refined enough to make those musical connections that would be obvious to a human DJ. Tastemaker wins.