Sometimes, another morning listening to Flippa, Lexy, and Bloggo's exploits on commercial radio feels tiresome. And the songs you put on the music app on your phone are great, but they're not quite what you need right now. What to do?
These days, there's a plethora of free or low-cost streaming services that offer radio-like functions without Flippa's "hilarious" anecdotes. And in the age of Big Data, it's no surprise that cold, impersonal mathematical algorithms are crucial to programming what you hear. On services such as Pandora Radio, the algorithm chooses what song you hear next. In contrast, Apple Music or Google Play Music may use algorithms to choose which human-curated playlist you see. And, in all of these services, listening behaviour is fed into the algorithm; if you skip Born to Run, you're less likely to hear Bruce Springsteen.
The role of algorithms in choosing our music can be disconcerting. We like to believe that our connection to music is a little bit mystical; that it's beyond maths and science. We want to believe that when a song makes us a bit teary, it's not because we're predictable robots, but because we're really connecting with the musicians. The thought of having an emotional connection programmed by a mathematical algorithm feels odd – are we really so predictable that a machine can nail our tastes so easily?
But the people behind these algorithms all emphasise that their algorithms are still fundamentally based around human input. Steve Hogan, of Pandora Radio's Music Genome Project, says Pandora's algorithm is based on musicians analysing music. Each of the million-plus songs on Pandora has been analysed by a musician, says Hogan, answering hundreds of questions about the song along the lines of "how much distortion is on this guitar, on a five-point scale?" or "what does the lead vocal sound like?"
"I do think the human ear can identify things in music that it's difficult for a machine to do now," Hogan says. "Even picking the tempo of a piece of music is surprisingly difficult for a machine. It's just not reliable."
After the music has been analysed, Pandora's algorithms figure out which individual songs will suit which playlists. Hogan gives an example: "We might know a particular listener tends to 'thumb up' when they get a breathy voice. So when the algorithms are considering what song to play next, the ones that have a breathy voice have a higher probability of playing for that listener ... we see that people react more favourably when we try and tailor it to their particular musical preferences, which they themselves might not even be able to articulate."
Spotify's Discover Weekly takes a different approach to Pandora: each week, Spotify creates a playlist of new music for you, based on what else appears in the playlists of people with similar "taste profiles". According to Matt Ogle, the developer behind Spotify's Discover Weekly project, itssecret weapon is the ability to analyse "the eight or nine years of playlists that Spotify users have been curating since the service began". Ogle says there are "now well north of 2 billion listens on Discover Weekly since launch [in July 2015]."
Sometimes Discover Weekly can be spookily accurate. Ogle mentions a friend who told him that "the stuff on my Discover Weekly is basically the stack of vinyl sitting next to my lounge". Ogle says, "all the missing tracks that we thought he'd want to listen to – Spotify didn't know he was already listening to them on vinyl."
In contrast to Pandora's focus on a musicological algorithm, Spotify seems intent on experimenting with different algorithms for different purposes. And so Spotify also has an algorithm that uses data from music websites and blogs to generate a playlist (the Fresh Finds playlist), and is working on algorithms that analyse elements of the audio file to judge the "danceability" of a tune.
Matt Ogle dismisses worries that we are all predictable robots. "You picture this sci-fi robot thing" he says, waving his arms around robotically, "but every song on Discover Weekly is there because at least one human being added it to their playlist".
Interestingly, Apple Music head Jimmy Iovine is not a fan of algorithms, having used words such as "sterile", "numbing", and "very limited" to describe their results. Iovine even told Wired magazine last August that "when you listen to a radio station programmed purely by an algorithm, you will go comfortably numb".
Despite Iovine's rhetoric, Apple Music's "For You" section still uses algorithms – the algorithms just determine which human-curated playlist you see, rather than generating the playlists themselves.
Google Play Music takes a similar approach to Apple Music: the "Play Music For Driving" button, for instance, delivers a playlist that was created by a human but selected by an algorithm. Rory Woodbridge, of Google Play Music, says that Google Play Music's approach is to "think about what our users are doing with music, and try and improve that – to help them work out harder, focus on work, or [play something] when they've got friends over". As a result, Google Play Music's algorithms pay more attention to how the time of day and what you're doing might affect the music you want to hear. "You're probably not going to the gym on a Friday night", Woodbridge says.
Regular radio stations, too, use algorithms to determine what song you hear next. "Our music programming jobs are done in part by algorithms," Double J content director Meagan Loader says. "It's no secret that radio uses music-scheduling software." For Double J, the algorithms help make the playlists less predictable, and help keep the correct balance of things such as genre, pacing and female vocals. However, a music director finesses the schedule to make sure that the playlist flows intuitively and that the algorithm gets the timing right. Loader also says the music director "makes sure Fire in the Disco isn't playing straight after the news covering raging bushfires" – something an algorithm might not consider.
In any case, Loader doesn't see streaming services as competition – Triple J has created some of the most popular playlists on Spotify. "When you turn on the radio, it serves a different need or 'job' than when you turn on a streaming service," she says. Radio, Loader argues, is all about a "shared, simultaneous experience", helping you connect to the world around you.
So do these algorithms show us to be predictable robots? Not quite. We humans might be predictable sometimes, but Spotify and Pandora haven't yet decoded why exactly you love Chvrches or Motorhead. In 2016, the algorithms used by streaming services still need to defer to the human judgments of playlist curators, musician analysts, or people who listen to similar music to you. Those still hoping that there's something mystical in our connection to music can rest easy. For the moment, at least.