I like to poke into the furthest reaches of Spotify’s database, in search of songs that have been heard little, if at all. I start with someone popular but a bit weird, like Peter Gabriel, and then I click on the “Related Artists” tab and up pops a set of similar musicians: Sting, Dire Straits, Roxy Music. Next, I select the weirdest group on that list—Roxy Music, obviously—and repeat the process; in this case that’s Japan, a new wave group that peaked early and broke up (the first time) in 1982.

Now I’m properly in the weeds. Blancmange, according to its biography and to Webster’s, is named after a “sweetened and flavored dessert made from gelatinous or starchy ingredients.” Its most popular song on Spotify is 1985’s “What’s Your Problem?” which sounds a little like Erasure and Kajagoogoo, and shows 57,455 listens. Keep going, and I’m in a land of musical serendipity: Landscape, Captain Sensible, and a band called Blanket of Secrecy. (Most-listened song: “Say You Will,” 3,253 listens). An estimated four million songs on Spotify have yet to be heard. There’s a service called Forgotify that helps you find them and play them—leave your footprint on the database.

Spotify picks related artists by watching what you’re listening to; if a lot of people who listen to Peter Gabriel also listen to Sting, then its analytics registers them as “related.” It’s a statistical relationship based on user behavior, nothing more or less. Selena Gomez, Demi Lovato, and Miley Cyrus are all “related.” Cyrus has 1.3 million “followers” on Spotify, and her top ten songs have collectively been streamed nearly a half-billion times. Mozart has 326,244 followers (only eleven million listens to his “top ten”—his followers lack commitment, apparently). Bach has 227,152; Beck, 317,972. Beethoven has 282,126, Camper Van Beethoven has 5,773. These relationships change as people listen to more songs and the algorithm processes more data. The statistics change, too, as more people listen to the songs. In a year, many of these pathways will be different. Keep going, listening to snippets of songs or speech. See what tugs at your ears. It’s one of the more satisfying ways I know to spend an hour. At the end you know something.

Spotify tries very hard to be a social network. You can follow people, send them songs, and engage them in conversations. There is a “Friend Feed” that tells you what songs your friends are listening to, as they listen. It makes me wonder, what the hell is wrong with my friends? One listens to too much classical; another is far too into electronica. There’s the music critic who loves her 1970s metal, the Depeche Mode-obsessive, and a host of other breeds of snob or miscreant. Watching other people listen to music is too much like knowing their sexual proclivities: You start out curious and end up horrified. There’s one exception: A friend who shares her account with her family in India. During the day she listens to synthesizer-noodling, but at night someone else takes over and some great Bollywood earworm will pop up in her feed, and I’ll jump on it. Listening to the favorite songs of distant strangers is a joy.

More interesting than the human interaction is the musical social network revealed by the search for new sounds. Japan, for example, is a bridge between more conventional pop and the treasures of the cut-out bin. Selena Gomez is a different sort of connector; her music is produced via a complex industrial process involving engineers, songwriters, and coaches, under the aegis of the record label Hollywood Records, which is owned by Disney; every unpleasant aural component imaginable is carefully removed until the finger snaps in “The Heart Wants What It Wants” echo around inside your deliriously emptied brain.

Spotify was started in Sweden in 2006 by an Internet entrepreneur named Daniel Ek; it’s now available around the world—which is a peculiar thing to say about an Internet service, which are all presumably global by default. But music services, unlike social networks or blogging platforms, must negotiate rights and permissions for each market. You access Spotify in the same way you search Google or watch Netflix; you don’t “own” anything you download. Spotify has some 30 million songs to be browsed and searched by 60 million users, 15 million of them who pay ($9.99 a month in the United States) to avoid advertisements and to remove limits on how many songs can be skipped. (Free users can only skip six songs an hour.) It’s fun to do some light math and see how those numbers add up. Assuming each song is two minutes long, that’s 114 consecutive years of songs. Now say a song with high-quality audio is ten megabytes in size—a fairly generous allotment. The entire database of music could fit in around 300 terabytes. A one-terabyte hard drive at Best Buy costs around $60, so that’s $18,000 to keep a single, compressed (but high-quality) copy of the 30 million songs. The additional metadata, like the names of the artists, labels, track lengths, dates, and so forth, is only a few gigabytes of data—consider that all of the pages on Wikipedia, sans images, are eleven gigabytes when compressed, so you could fit 80 or 90 copies of that on a single one-terabyte hard drive. Text isn’t your problem, and you can handle band photos pretty efficiently.

So their big cost isn’t maintaining an archive of music. So what does Spotify, which is valued at nearly $6 billion, actually do? Well, it organizes those songs, backs them up, and then deals with the frustrations of making them available over the Internet to 60 million human beings. Obviously one copy of the whole database can’t serve all those people, but it doesn’t make sense to keep 60 million copies, either. Finding the middle ground is an ongoing engineering challenge. And they need to account for bandwidth: Sending data over the Internet is still one of the more expensive things a digital business can do. Which leads to another back-of-napkin estimate: If compressed audio weighs in at between 2.4 and three megabytes per minute, and you have 60 million users listen to ten hours of music a month, that’s around 86 petabytes of data per month. Using Amazon Web Services’ calculator to estimate the costs, that might run you $4.5 million a month.

Someone has to get that music off the disk, track who’s listening to it, verify that he or she is an authentic user, make sure no one’s logging in at multiple places. Spotify has different services for this—it’s not just one big program called Spotify running, but more than 100 programs running all over the place. Music comes in via the digital equivalent of a truck delivery: Spotify ingests big piles of audio and data that are dumped on it by the music labels, and its robot librarians go to work making sense of it all.

Spotify does not own the vast majority of the assets that it makes available; it does not control the relationships with the artists; does not own the networks over which its music is distributed; and many of its computers are leased. What it does is create and manage the relationships with the music labels, so that it can lease all that music, then provide users with access to that music; and then it does an accounting around these things. It takes money from paid users and advertisers, then sends royalties back to the labels, who, by many accounts, hand out dolefully small portions of what they receive to the artists themselves.

One of the most entertaining searches you can do on Spotify is for Hitler: There are tons of songs. A band called The Buttplugs wrote one about Hitler’s nipples. There’s one by Antony & the Johnsons (“Hitler in My Heart”), and one by Faith No More (“Crack Hitler”). There’s the obligatory Mel Brooks number, plenty of punk, and a track by Bob Newhart. There’s a Churchill speech and a testimonial from an RAF Bomber, and the announcement of the Führer’s death on German radio. Under related artists, where you’d expect to find Hideki Tojo, Benito Mussolini, or maybe Himmler, you find Neville Chamberlain, Edward Kennedy, John Glenn, and Charles Lindbergh. Statistics aren’t the same as historians.

Related Artists is actually a social network for people with extremely eccentric friends: You can get from Nazis to an album of Kurt Vonnegut reading Slaughterhouse-Five in a few clicks. Here’s how: Start with Hitler, and then go to Charles Lindbergh. Take a left at Franklin D. Roosevelt, a hard left at Studs Terkel, and an even harder left at Ward Churchill. Veer slightly right (but you’re really still going left) to Howard Zinn, then Angela Davis. Enter a tunnel until you hit Lawrence Ferlinghetti. Next you’re at Gertrude Stein, who is unexpectedly close to Dorothy Parker. Head right until you see Rudyard Kipling, and after that you can’t miss Vonnegut.

People lament the ascendance of pop culture. Just because Katy Perry is in charge doesn’t mean that Charles Ives is out in the cold. He has 57,643 listens for his 1906 composition “The Unanswered Question.” Honestly, I don’t care what people like, only who they were and what they thought. I want a way for people to mark their paths through all this sound, so that I might follow.

Paul Ford's Related Artists Playlist