Netflix is giving you bad taste Aesthetic value can't be decided by algorithms

The end of holiday season is upon us, and a chance to apply a new perspective to life as it slowly returns to boring normality. For a few short weeks, if you were lucky enough, you were able to immerse yourself in a different culture: eating new food; meeting new people; gazing, fascinated, upon new and startling sights as you encountered your Airbnb landlord’s Netflix algorithms.

In my own case, a week in Scotland led me to a lot of American college football documentaries. I now know how Jonny Manziel wrecked his career and exactly how Manti Te’o got catfished; what a “fake punt” and a “pancake block” is; and how pivotal the spread offence was to Urban Meyer’s tenure at the Florida Gators. Don’t tell me travel doesn’t broaden the mind.

Safely back home in the comforting arms of my own viewing profile, I’ve started to seek out more documentaries about sports I previously knew nothing about. (My god, the cobble stage of the Tour de France is brutal, isn’t it?) The algorithms generating my homepage, normally invisible to me, have gone haywire — a fact prompting me to look into exactly how they were constructed in the first place.

To work out what you might want to watch next, Netflix first classifies its programmes into at least 76,897 “microgenres”. These group each product under one or more relatively detailed headings, so that combinations of features which apparently appeal to a viewer in one instance can be quickly found in other cases, and then recommended to her. These include such categories as: “Goofy Comedies about Royalty”, “Violent Foreign Mother Son Relationship Movies”, “Understated Australian Movies”, and “Award Winning Underdog Movies”. It is unrecorded whether any particular programme counts as all four of these at once.

It’s interesting to consider what microgenres do to genre criticism in the traditional sense. Critics often assess the success or failure of an artwork against intended genre membership — if it’s a western or a romcom, for instance. Is it a good example of the genre or a bad one? Is it intended to subvert the genre in question, even? Such questions assume that the creators know which larger categories they are working within, or against.

Microgenres aren’t like this, though. You can hardly blame a director for not knowing, all along, that she was competing against other Dark Dramas for Hopeless Romantics Based on Books. And I don’t see how you could “subvert” microgenres at this level of specificity — you’d just automatically enter a different one.

In any case: after humans classify all the available products into microgenres, machine learning takes over. It cross-references information about what you already watched, at what time of day, and for exactly how long, with what other people watch. This places you into one of at least 2,000 worldwide “taste communities” or “taste clusters”. Based on your membership of a given community, Netflix can now start recommending to you the films and programmes also watched by what a company vice-president has referred to as your “taste doppelgangers”. Unlikely as it may appear, it seems I’m probably not the only viewer in the world to have simultaneous recommendations for Joan Didion: The Centre Will Not Hold and At Home With The Furys.

Doing my research, I’ve been unable to find out whether micro-generic adjectives like “Goofy”, “Violent” and “Understated” are supposed by the company to refer to some absolute standard, or whether they are better understood as relative to local standards in the product’s country of origin. It seems to me that an Understated Australian Movie is probably quite different from an Understated French One.

But what does seem clear is that — officially according to Netflix, at least — membership of taste communities is not influenced by geographical location. Nor, allegedly, is it influenced by age, nor even sex. “We have seen that where you live, gender, age and other demographics are not significantly indicative of the content you will enjoy,” a company spokesperson has claimed. “Time after time, we see that what members actually watch and do on the service transcends the predictions of stereotypical demographics.”

If this really is true, it goes against traditional thinking about aesthetic response. For centuries, it’s been assumed that your taste is heavily influenced by local context. Take David Hume, for instance, writing in 1757 about books: “We are more pleased, in the course of our reading, with pictures and characters, that resemble objects which are found in our own age or country, than with those which describe a different set of customs”. For this reason, he also claimed that “comedy is not easily transferred from one age or nation to another”. (In his defence, this was years before the international popularity of The Benny Hill Show complicated the picture.)

So, are national or regional sensibilities irrelevant to aesthetic taste after all? Or is the truth more depressing: that, thanks to companies like Netflix, each of us now swims in a pleasingly benign, individually personalised yet globalised televisual soup, so that nothing as discerning as aesthetic judgement can get going in the first place? I fear it’s the latter.  Choosing your next title from a range preselected for you — and especially where you know pretty much nothing about the titles otherwise — is more like randomly selecting a chocolate from the box than an exercise in careful discrimination.

When people tend to fret about algorithms governing streaming, they usually worry about what it does to our morals — potentially distorting our preferences towards the violent and exploitative on the basis of a few random watches — but not what it might do to aesthetic sensibilities. Yet maybe they should. It’s not just that failure to achieve instant popularity with your assigned taste community means a show can drop out of visibility on your homepage even just one day after launch, meaning that slow-burning series get no real chance for ignition. It’s also that, in terms of guidance about what to engage with, algorithms are replacing recommendations by humans, but with none of the surrounding benefits.

It’s clear we all need help in selecting items from the tsunami of streaming content available — decision-paralysis is definitely a thing. But algorithms speak straight to your limbic system, as it were, bypassing self-conscious reasons for picking this thing over that thing. They move choice out of what philosophers call “the space of reasons” into the less cerebral space of impulse-driven reaction. Critics and reviewers, in contrast, can tell you not just what you should like, but why you should like it.

Before streaming arrived, the timing of watching was not up to the viewer. It wasn’t like music where you could instantly buy a CD. You had to wait for your opportunity to watch new releases, along with everybody else in the country. Not only did this make possible more of a communal experience, but it also gave the professional critic a power that is now fast disappearing. She could, with some authoritative pull on her readership’s attention, issue them with a bold challenge: watch this programme, and either agree with my verdict about it, or explain to yourself — and to each other — why.

These days though, when a reviewer writes critically about a show on a streaming service, she’s already playing catch up with existing public responses. There is less urgency, and a lot of her authority to command attention has been lost. If she praises something that’s already popular, she sounds obsequious and timid. If she dislikes it, she sounds querulous and out-of-touch. She can still write about terrestrial channels of course; but here, too, there’s no longer a shared, simultaneous experience of the programme to ground public discussion. Everybody is off in private corners of the internet, doing their own thing.

Some may respond that even so, the world is still awash with people writing critically about film and TV. This is true but is scant consolation. With most internet criticism being fired aimlessly into the global void in the vague hope of an audience, there is little sense of a dialogue being initiated, with a defined readership in mind, in order to experience something all together — good, bad or indifferent — then argue about it.

Another problem is that, just as decision-paralysis is generated by the sheer number of programmes out there unless we have help of some kind, the same goes for the huge volume of online reviews too. Interesting voices or views tend to get lost in the crowd. Soon enough, machine algorithms will probably work out what kinds of review an individual might like to read: Withering Scandinavian; Largely Effusive With A Couple of Caveats; Sixties British Sexist; and so on. For all I know, maybe they do already.

To bypass this volume problem, sites such as Metacritic and Rotten Tomatoes sample highlights from reviewers and aggregate their ratings for a film or show, effectively treating individual critics as members of large panels whose joint verdict counts more than individual ones. Hume would approve of this. He also thought that joint verdicts of “true judges” were more reliable than ones from single individuals, since they allowed private prejudices, biases and blind spots to more readily cancel each other out.

But I think that, with criticism, we shouldn’t necessarily seek reliability. Choosing what programme to consume for an evening is not like buying a used car or choosing a spouse. The stakes are really not that high. What I want from a critic, rather, is erudition and provocation, the better to form my own judgements and not just blindly follow those of others. No matter how well the algorithms anticipate what I want to watch this evening, they are not going to give me that.

---

Kathleen Stock is Contributing Editor at UnHerd.

Docstockk