What a cruel twist of fate. The more freedom of choice a couch-bound content-devourer has, the more oppressive it feels to actually choose.
It’s death by a thousand decisions. The endlessly scrollable Netflix menu, for instance, should in theory spark joy in every subscriber. All kinds of entertainment, right at your fingertips! But then the reality sets in: ALL kinds of entertainment. Right at your fingertips. Every new choice increasing the likelihood of picking the wrong one and then bailing at minute 15. Netflix users reportedly spend an average of 1.8 seconds considering each title they encounter. And since Netflix has thousands of titles, those 1.8-second sessions add up to hours of analysis paralysis. Do you want to finally watch Swiss Army Man? Or do you want to see a show about botched cosmetic surgery? How sweet it would be to not have to be in charge of deciding.
Thankfully, just as the servers at Cheesecake Factory offer suggestions from the restaurant’s famously mammoth menu, Netflix offers personalized recommendations to help users cut through the clutter. And while a Cheesecake Factory server only knows what he or she likes, or what other customers tend to like, Netflix’s algorithm draws on your entire history of decision-making to determine what you–specifically you–might like, and what you might like right now. Whether it can actually deliver on that promise, however, is another story.
I have long been curious about how Netflix’s recommendation algorithm works in practice. My list of recommendations is total anarchy. The streaming service not only has my viewing habits to draw from, but also the frequent compromise choices I make with my wife, along with the solo viewing choices she makes when she forgets to log in from her own account. The Top Picks for Joe often seem like the work of someone who has no idea who I am, or at least doesn’t know about my intense aversion to Joe Rogan. Is this because the various viewing modes on my account don’t necessarily represent me, or is it because the algorithm itself is inherently flawed?
I decided to find out.
Since my current Netflix account is compromised, I decided to open an entirely new account and, tabula rasa-style, take all of its recommendations for two weeks. I would watch whichever shows and movies Netflix pointed me toward–and watch my recommendations evolve as the platform decided what kind of person was watching. I would be born anew and assessed by the digital sorting hat again. Perhaps this extremely suggestible version of myself would end up triggering more accurate recommendations than my actual account. Or maybe I would bump up against the same kind of randomness that colors my account already. Either way, I hoped it wouldn’t be too boring.
Helping to find the shows I’ll love
Netflix closed out 2018 with 139 million paying users worldwide, 29 million more than it had a year earlier. In the United States alone, Netflix added 1.5 million new members in the fourth quarter of the year. It’s hard to imagine just who is only now getting around to seeing what this whole Netflix thing is all about, but I am re-joining their ranks. Having been inducted over a decade ago, I forgot what this process entails. First, I pick a plan as my alter-ego, Boe Jerkowitz, and then the platform urges me to choose three titles I like.
“It will help us find TV shows & movies you’ll love!” Netflix promises, breathlessly.
There are more than 100 options, but mercifully not the full library. It’s a crowd-pleasing mixture of Netflix Originals like Mowgli: Legend of the Jungle, recent blockbusters like Avengers: Infinity War, and popular network series like Grey’s Anatomy. It seems only appropriate to choose one from each of these unspoken categories, so I pick Stranger Things, Black Panther, and The Good Place, all of which I genuinely enjoy.
A caption appears that reads, “Personalizing for Boe Jerkowitz.” I love to be personalized! Alright, entertainment cyborg, knock my socks off.
It’s interesting to see what Netflix suspects I might want to watch just based on those three initial titles. Autoplaying at the top of the screen is an ad for the just-released Ted Bundy Tapes, a Netflix original that I am indeed interested in, but one that’s a long way from The Good Place. (The smart money says Bundy is in the Bad Place.) The categories these suggestions are sorted into are mostly the usual suspects, but they get more niche the further I scroll. There’s “Conspiracy TV Shows,” “Oddballs & Outcasts,” and most intriguingly, “NX: Multiple Universes, One Home.”
I go with one from the Top Picks for Boe Jerkowitz section: A Wrinkle In Time. It’s pretty good! Lots of vivid visual flair. When the movie ends, Netflix auto-recommends Lemony Snicket’s A Series of Unfortunate Events, a Netflix original starring Neil Patrick Harris. I hadn’t really been looking for tween-friendly programming to begin with–the pull for A Wrinkle in Time was director Ava DuVernay–but rules are rules.
Lemony Snicket is darker than I expected and not un-fun, but still not quite up my alley. I watch the first and last episode because Netflix only recommends something else after a TV series once you’ve watched through to the end. Next up is the platform’s new take on Carmen Sandiego.
Oh, no. Why did Netflix assume I wanted to go even younger than I’d already gone? Did it think I was a teenager with Benjamin Button disorder? I had been sucked into a YA programming vortex, and I had to escape.
Of course, what might help is if I gave Netflix more to go by other than those original three titles and the two I’d just watched. I soon added 12 more titles to My List, movies and shows from all over the televisual map. I put in Korean zombie movie Train to Busan, Catherine O’Hara’s riches-to-rags sitcom Schitt’s Creek, as well as Schindler’s List, Scott Pilgrim vs. the World, and Black Mirror: Bandersnatch. That should complicate things nicely.
Now that I’d expanded my palate, Boe Jerkowitz’s Top Picks had changed too. Thanks to my recent viewing habits, they now included Coco, Hotel Transylvania 3, and Boss Baby, naturally, but there was also some more adult fare. I definitely needed to pick something with bloodshed or at least colorful language in order to Google Map my way out of Sesame Street.
I choose Chappie, a punchline of a dystopian robot romp from 2015 that would either be secretly great or hilariously awful. It was . . . neither, a bewildering miscalculation that made me wonder how in the world District 9 director Neill Blomkamp convinced a major studio to give him $50M for a movie starring the South African hip-hop outfit Die Antwoord (!) as themselves (!!).
After the movie ends, Netflix recommends Next Gen, an animated feature starring the power trio of John Krasinski, Charlyne Yi, and Jason Sudeikis. I was still stuck in kiddie hell, and in a subsection nobody has ever heard of. (Rotten Tomatoes boasts only a literal handful of reviews for Next Gen.) Why had Chappie led me here? Sure, both films involve robots, but only in Chappie does a man get graphically killed by a b-boy robot doing the A-town stomp on his internal organs.
The answer to this question ends up shaping the rest of my experiment with Netflix’s algorithm. I realized then that the reason the platform pushed me toward Next Gen after Chappie is because Next Gen is a Netflix original that shares a tiny bit of DNA with Chappie. Netflix was steering me from an outside studio’s robot movie to its own movie that involves robots. It was in-house advertising, not a recommendation. Now I wondered just how often this exact scenario happened.
Who does Netflix think I am?
Next Gen be damned, my Top Picks were now finally looking more adult-like. Pulp Fiction, Hellboy, and Shaun of the Dead were there now–all winners in my book. Flipping through the options, I chose Next Friday, which ended up justifying my long-ago prediction that it probably wasn’t anywhere near as good as the first Friday. As the credits rolled, Netflix recommended The Ted Bundy Tapes, a curious pick. Unlike with Chappie and Next Gen, there was zero connective tissue that I could discern between Friday and Bundy. My best guess: Perhaps when there is no direct Netflix analogue for whichever title a user has just watched, the algorithm recommends any random Netflix original from the user’s Top Picks. If that random original happens to be a fresh release the platform is pushing on every subway ad in New York City, so much the better.
Watching a movie that stars Ice Cube after one that stars the members of Die Antwoord had a profound effect on my Top Picks. Netflix now seemed reasonably sure that Boe Jerkowitz was all in on Ice Cube’s entire filmography, and that he wanted to watch other movies starring and/or about rappers, including Eminem’s 8 Mile, Snoop Dogg and Wiz Khalifa’s Mac and Devin Go to High School, and most galling of all, Jamie Kennedy’s Malibu’s Most Wanted.
I had never felt less understood in my entire life–and that includes the time I went to Paris by myself without an iPhone or even a rudimentary grasp of the French language.
Who does Netflix think I am? Furthermore, who am I? Suddenly, I was curious to see whether Netflix knew the me on my actual account any better than Boe. When I logged in, I noticed some differences between the tiles on some films and shows on this account versus the other one. The Fyre Festival documentary here featured a closeup on a bikini butt with crossed fingers in front of it. On Boe’s account, the tile showed a logo for the film with a fishhook through an Instagram heart. Did Netflix think Boe hated butts? Black Panther here featured the hero in costume, ready for combat; Boe’s version sported a shirtless Killmonger instead. What about my viewing habits suggests that I side with Killmonger (I mean, maybe a little?) or that I’m attracted to Michael B. Jordan’s physique? (Who isn’t?) My home account’s Good Place tile was a yellow and blue bowtie, whereas Boe’s offered a fro-yo swirl of those two colors. I wasn’t philosophically equipped to unpack the implications of all these differences.
At this point in the project, I began to surrender my hopes that Netflix would get to know me better with my clean, newborn account. All I wanted to know now was whether it ever led its users from the end of a movie or show to another movie or show that wasn’t a Netflix original. I spent the remainder of my second week hoping that, just once, it would slip up and recommend other studios’ films and shows. You know, like the original purpose of Netflix?
It never did.
I watched Seth Rogen’s North Korea-agitating comedy, The Interview, which led me to a Netflix original standup special by Gabriel Iglesias.
I watched the Gabriel Iglesias special, and it led me to another Netflix original standup special, this one by Sebastian Maniscalco. The only thing less funny to me than either of their specials, two spectacularly unmemorable flurries of joke-shaped words and gestures, was their rankings on the Highest-Paid Comedians of 2018 list.
I watched a no-budget horror movie called Malicious, which led me to Malevolent, a low-budget Netflix original horror movie with a name so similar to the one I’d just watched that it felt like a joke, possibly one from Sebastian Maniscalco.
I watched the Vince Vaughn/Jennifer Aniston rom-com, The Break Up, which led to the Netflix original rom-com When We First Met, starring Adam Devine and Alexandra Daddario. It was at this point I knew that the experiment had officially concluded. What more did I need to know?
Netflix doesn’t want to model itself after any one studio or TV network; it wants to usurp all studios and all TV networks, to be all things to all people. It used to be a platform for giving viewers what they want, but now it’s mainly a platform for giving viewers its own Netflixian version of what Netflix’s carefully kept data says its viewers want.
The company spent $12 billion on content last year and it expects to spend $15 billion in 2019, so this strategy will inevitably sometimes pay off with your next authentic TV obsession. (Hello, Russian Doll, which I discovered during this experiment.) You don’t need an algorithm to tell you the math is on their side in that regard. But in many other instances, Netflix will be content to function like a glitched-out vending machine that hears “Vince Vaughn and Jennifer Aniston breaking up” and serves you “Adam Devine and Alexandra Daddario getting together.”
I am now no closer to knowing what I want, but at least I know what Netflix wants.
[Photo Illustration: Samir Abady; Velvet Buzzsaw: Claudette Barius/Netflix; Narcos: Mexico: Carlos Somonte/Netflix; Maniac: Michele K. Short/Netflix; To All the Boys I’ve Loved Before: Masha Weisberg/Netflix; Gabriel “Fluffy” Iglesias: One Show Fits All: Anthony Nunez/Netflix: Big Mouth: courtesy of Netflix; Sex Education: Jon Hall/Netflix; Nappily Ever After: Tina Rowden/Netflix; Dumplin’: courtesy of Netflix]