Super Abbreviated History of Music Listening
In the 1980s, mixtapes were recorded, painstakingly, one song at a time. Then we had Now That’s What I Call Music! compilations in the 90s, Napster mp3’s burned onto CD’s at the turn of the century, and iPod shuffles in the mid-aughts. The unifying thread amongst all these methods of playlist construction is that they were human created. It wasn’t until Pandora, Spotify and iTunes Radio hit the mainstream that AI-driven music curation played a significant role in the discovery process for music fans.
Now however, AI curation has become the norm, with BBC reporting that “an algorithm...will decide what you hear–and critically, what you’ll hear next.” People still create customized playlists for special events, and DJs–though they have been forced online recently–still play an important role in music listening and discovery. But in general, how often are we or another human choosing the next song, as opposed to an algorithm?
Hypothesis: People prefer People for Recommendations
We haven’t run the survey yet, but we’re betting that if you asked people whether they want an algorithm or a human creating their playlist, they’d choose the latter. There’s a vital communal element at the heart of music...one that’s often expressed through giving and receiving music recommendations. Artists like Diplo and Annie Mac have built careers on their tastemaking instincts: having an ear positioned slightly ahead of the curve and sharing the best of what they discover.
So Why Do We Need Music Algorithms?
Record labels have been tastemakers for as long as they’ve existed; their business model is predicated on finding the music that people will want to hear, even when the audience doesn’t know it yet. These companies (mostly) succeeded at matchmaking music and listeners for decades, until the internet rendered the recorded history of music available at the click of a button. This shifted the focus of listeners from accessing music to selecting music from the vast sea of options. And while record labels continue to curate rosters to appeal to as many listeners as possible, the tricky task of customizing and personalizing listening–at scale–is mostly falling to streaming companies.
This is where the massive potential of AI comes into play. With AI-driven curation, every playlist is unique to you. By piecing together the manual music selection patterns of hundreds of millions of people, AI can recognize patterns in taste efficiently and accurately in ways no human can. Across a data set of trillions of listens by millions of users, a streaming platform knows that you’ll probably like the next song they recommend. And, this algorithmic curation can certainly lead to “wow” moments of discovery. For example, Apple Music may notice that a subset of subscribers who search for Disclosure also listen to a lot of J Dilla; this will lead to Apple including J Dilla tracks when a Disclosure station is created, even if the gaps in genre, tempo, and subculture render the recommendation a less-than-obvious suggestion.
Robots + Humans = Music Curation
But there’s no personality to an algorithm (yet), and each of the services are even starting to look the same. As an important course correction, the owners of streaming services are making a concerted effort to enliven their algorithms with a human curation filter.
Perhaps the computerization of taste building was unavoidable, but it’s definitely not perfect. Humans don't want to lose out on the human-ness of sharing and seeing the smile on peoples' faces when recommending or playing a good track. Our take is that we all win when we can find a way to leverage the scale and speed of algorithms fused with the nuance and personality of a great DJ.
At Feed.fm, we leverage algorithms and machine learning to help us understand the habits and preference of millions of unique listeners in dozens of different music listening contexts. But we stand by our curation team’s ability to interpret the data, make smart decisions, and find new ways to introduce fresh music to listeners. Rather than an “either/or” situation, we like to think of curation as a “yes/and” opportunity.