New Times,
New Thinking.

  1. Science & Tech
6 September 2019updated 08 Jul 2021 1:32pm

How Netflix’s algorithm can find exactly what I like – despite knowing nothing about me

By Sarah Ditum

Anxieties about online tracking tend to have a flattering element of the personal to spike the creeping fear of being monitored. After all, this is about personalisation. When Google starts serving you ads for things that you’ve been talking about near your Android phone, but are absolutely sure you haven’t searched for, it’s disturbing that it knows so much; it’s also confirmation that there is a you, and you are interesting. But, for some of the services that seem to know the most, there is no “you”. It’s an understandable mistake to make, but a mistake still – at least as far as the Netflix algorithm is concerned.

Giving an accurate recommendation is proof that you really know someone. We give favourite books to friends and lovers, show them our most treasured movies and make them playlists, not just out of generosity but as a kind of test: if you’re who I think you are, who I want you to be, then this will mean as much to you as it does to me. The opposite is true, too. I’ve had relationships that have shrivelled and collapsed like a salted slug following the application of the bad book present: you think I’m this?

In the usual version of ourselves, taste is at the centre. After the obvious, visible qualities such as race and sex; after less easily discerned facts such as class, sexuality, faith and voting affiliation; there, in the tender pulp of our personality, sit the things we love the most. As teenagers, building an identity of our own, we gather culture up like decorator crabs sticking bits of deep-sea detritus to themselves – in the human version, this is not done to camouflage, but to display the essence of our us-ness. I like this; I am this.

Yet this belief that personality equals taste came under attack long before the arrival of algorithms. Reading the sociologist Pierre Bourdieu at university, and his analysis of cultural habits as a marker of demographics, I recoiled: I liked what I liked because it was good and I was discerning, not because I was the child of white, middle-class parents. All the same, I could remember with a smart those family conversations where the world cleaved hard around what we did and didn’t like. I knew, without ever really being told, that it would be utterly embarrassing to like Andrew Lloyd Webber; and I have never liked Andrew Lloyd Webber.

Still, at least learning that my taste was a property of my background left some sense of selfhood behind. When it comes to Netflix, I simply don’t exist. There’s a general assumption that a service such as Netflix must be profiling you – figuring out all those Bourdieu-ish categories that will tell it what you enjoy.

But that’s not how Netflix works. All it knows is what you watch, and what other people who watched those things also watched. Even the word “people” in that sentence is arguably out of place: there are no people in the Netflix algorithm, only relationships between shows and movies.

It’s quite hard to let go of the idea that you exist, of course, and users are still routinely convinced that Netflix must be profiling them. The ever-more-detailed way that Netflix “A/B tests” its interface – testing two versions of a single variant – certainly doesn’t help with the paranoia. When black viewers realised Netflix was serving them thumbnails that highlighted black actors who were bit-players rather than the stars of a movie, they understandably felt they were being exploited. “There was 30 minutes of a romcom I ended up watching last week because I thought it was about the black couple I was shown on the poster,” Tolani Shoneye, a host on the Receipts Podcast, told a newspaper in 2018. “I want to see those stories. They know I want to see those stories. Why don’t they just make more of them?”

The interpretation reached by Shoneye and others was that Netflix knew their race, and was effectively blackwashing movies with majority white casts rather than creating new vehicles for black talent. And that is what Netflix was doing – but not because it had any clue who the users were, only because it knew that they had previously watched shows advertised with thumbnails of black actors. Part of Netflix’s success in getting people to click on its content is thanks to its facility for repackaging everything it offers for every possible sector of viewers. There’s a twofold vacancy here: first of all the user doesn’t exist to the company, and then the company eliminates itself from responsibility for its own choices. It was just the algorithm doing its thing.

This might be elegant, but it is ultimately conservative. Netflix’s system solved the problem of predicting what people want to watch, but it couldn’t solve the politics of the entertainment industry which determine what gets made. And it has the potential to create a closed world where everything you encounter resembles something you already like.

The system works, though. I know that I’m the kind of person who would like to watch improving documentaries and subtitled dramas on BBC Four and I dutifully add them to my dusty queue on iPlayer. Then Netflix serves me campy reality shows and true-crime series, and they get watched. You didn’t need to gaze into my soul to understand my personal aesthetic. You just needed to start from the position that I’m as basic as everyone else.

Right now, for example, Netflix is convinced that I should watch a thriller called Velvet Buzzsaw. It looks tedious and try-hard from the trailer; but I’m keen on its stars, who include Jake Gyllenhaal and Zawe Ashton, and it riffs on bits of pop culture I’m obsessed with.

I pressed play. It was irredeemably terrible. I felt the surge of revulsion that accompanies the misjudged gift: you think I’m this? Unfortunately, I watched it anyway. Netflix doesn’t care who I am, but that just makes it all the better at pushing the buttons for what I am.

Content from our partners
The Circular Economy: Green growth, jobs and resilience
Water security: is it a government priority?
Defend, deter, protect: the critical capabilities we rely on

This article appears in the 03 Apr 2019 issue of the New Statesman, The Brexit wreckers