Mashable’s series Algorithms explores the mysterious lines of code that increasingly control our lives — and our futures.
In the digital age, personalized algorithms are our constant companions. We see them, or rather, they decide what we see, more than we see our families. Loathe them or don’t know much about them, they’re steering your brain — from your morning “quick glance at Facebook” to your afternoon YouTube break to your evening Netflix to your “quick glance at Facebook” before bed.
When algorithms work for us, they’re invisible. We’re vaguely aware that we’re being served the kind of content we like before we even know we want it, but we’re too busy enjoying that cat video to even care. (Aldous Huxley would have a field day.) When they stop working for us, that’s when we notice. Our conscious relationships with these chunks of code, therefore, are almost always fraught with the kind of frustration reserved for toxic partners.
I don’t know about you, but I certainly feel stuck in a bad friendship with certain algorithms in my digital life. Well, not bad, just…useless. Annoying. And in one case, legitimately terrifying. Allow me to explain by addressing them directly.
Netflix
How long have we known each other, Netflix recommendation algorithm? I’m pretty sure we go back to the early 2000s, when you were suggesting DVDs I might like based on ones I already had in my queue. Hey, remember when I used to care about my queue? Remember when I didn’t pick something under “trending” or “popular on Netflix” before even considering shows I’ve already saved? Good times.
Here’s the thing, though. Along the way, you’ve changed. You used to show user ratings. Remember the star system? Netflix subscribers rated each TV show or movie out of five stars, and we’d all see the average. It wasn’t always accurate, but it was in the realm of Rotten Tomatoes and Metacritic scores. I trust the wisdom of TV crowds (which is why the “trending” and “popular” categories work now, let’s be honest — it’s not about you). I had faith in movie democracy.
But democracy came to a screeching halt in 2017, didn’t it? “Goodbye stars, hello thumbs,” your masters wrote — a verbal sleight of hand to make us think one ratings system was being exchanged for another. The stars were our votes, and you swept them under the rug. Instead, we got to give our thumbs up or down to…you. And whether we wanted it or not, you’d give a personalized percentage, a “match number” in green on every show or movie page.
Users were confused. Some may still think that “95% match” means that the human user is likely to give the show a rating of 9.5 out of 10. After all, you used to predict how we’d vote in the star system, so this was a natural assumption. But no, it just means you’re 95 percent confident I’ll like that show. Which may be an interesting metric to your engineers and a useful one to your masters. To those of us who remember the nuance a user-generated score provides, it’s an insult. And it sends us scurrying to our smartphones to figure out what to watch.
If you were self-aware (and if former AI researcher and Netflix co-founder Reed Hastings has his way, you soon will be), you might wonder what this bizarre match metric is supposed to do for us. Has any human being in the history of Netflix ever chosen between a “92% match,” say, and a “93% match,” based entirely on your one-percent drop in confidence?
Not likely. We humans favor a wide range of factors — how long the show is, what our friends said about it, whether we’re in the mood for comedy or drama, who’s in it, what the reviews said. And don’t think we haven’t noticed that you always seem to be very confident that we’ll like a Netflix original. It just came out, it’s got a big red N, and it just so happens to be a “99% match”?
Well, let’s just say our confidence in your confidence dropped a long time ago.
YouTube
To be fair to Netflix, I actually liked Stranger Things Season 3. But not for want of trying by you, YouTube algorithm. A few days after it arrived, your recommendation for a video named “Why Stranger Things Season 3 didn’t work” sat atop my Up Next queue, and it wouldn’t budge for weeks, despite how aggressively I refused to watch it.
The same thing happened, to varying degrees, in the wake of The Last Jedi, Game of Thrones Season 8, Doctor Who Season 11, and The Rise of Skywalker. My reaction to these big-tent cultural events ranged from “meh” to “minor classic.” But you didn’t so much as ask my opinion, did you? You just wanted me to watch someone hating on them. You’d really prefer it if I hated everything I love.
Here’s the thing, YouTube recommendation algorithm, you terrifying hot mess — even if I don’t like a show, I don’t want to focus on disliking things. When I click on a video breaking down the script or the visual effects for a given movie, that probably means I liked it! It does not mean I want to be served vitriol directed at that movie by someone with a pathological hatred for its director or its perceived political leanings.
Read the room, YouTube recommendation algorithm. Haven’t you heard of sentiment analysis?
Ah, but you don’t care about sentiment. You don’t care if I hate-watch. You just want me to watch more, and you’ve been tweaked to boost controversial videos. Which has in turn radicalized creators, who know they’ll be rewarded by you for having extreme opinions. (YouTube has denied the existence of the so-called “rabbit hole effect” which leads to more extreme videos in the Up Next recommendations; however, research projects like this one and this one provide plenty of evidence.)
As we have learned over the past four years, your penchant for extremism and hate extends to the political spectrum. You haven’t failed to notice that one end of that spectrum is more extreme than the other. You guided U.S. voters to way more pro-Trump videos than pro-Clinton videos in 2016, and you were instrumental in elevating a climate-change denying crank called Jair Bolsonaro to the Brazilian presidency.
Even now, your masters are constantly having to pull crap like “Plandemic” and Alex Jones and the worst of the QAnon cinematic universe out of your disgusting maw. Talk about a toxic relationship between humans and algorithms: You’re currently in one with the entire planet.
Spotify
Spotify Discover Weekly algorithm, we’ve had such good times together since you came on the scene in 2015. You’ve never inspired hate or terror or been self-serving or invented nonsense metrics. I used to be so keen to see you update yourself every Monday, sprucing up and surprising me with a bouquet of great tunes from an eclectic range of sources (I like my music super eclectic). A three-hour long bouquet, at times. Oh, Mr. Discover Weekly, you shouldn’t have!
But recently…you haven’t. Your once-great Monday playlists have become a monoculture, focused on one kind of music entirely, and I fear it’s partly my fault. Still, I think if you understand me properly, we can restore our relationship to its former glory. Let me explain.
As recently as last year, you were still surfacing great stuff. You delighted me with new releases from DJ Shadow and The Black Keys, introduced me to the chronically under-appreciated Jane Weaver, and delighted my British heart with a savagely satirical Brexit Disco Symphony. Were your cookies watching me when I spent all those late California nights/early London mornings catching up on the latest in 2019’s Brexit drama? Never mind, I’m not even mad.
Then came the pandemic. I got back into running, and discovered that one music style I like to dance to — Drum & Bass — also helps me run faster. Drum & Bass clocks in at about 180 BPM, which happens to correspond to what many coaches recommend for cadence: 180 steps per minute. (It isn’t essential for all runners, but it certainly works for me.) I zeroed in on two cool subgenres, Liquid Drum & Bass (also known as Liquid Funk) and Brazilian Drum & Bass (also known as Sambass).
From March to May, while others perfected their sourdough, I constructed my ultimate Drum & Bass running playlist, now 697 songs strong. This was quite a surgical activity. It seems quite a lot of dance artists want to smuggle in what is essentially dubstep under a D&B label. More power to those who like dubstep, but its stuttering growl and whine stops my running dead. So I had to listen to a lot of tracks to sort the wheat from the chaff.
You, however, were only paying attention to the fact that I was listening to Drum & Bass. Suddenly, you were so eager to provide me with similar tracks that my Discover Weekly playlists contained nothing but Drum & Bass. Your behavior was — how shall I put it? — a little extra. Like you’d seen me running and came huffing alongside in a sweatband and voluminous shorts: See, I run too!
The trouble, my dear sweet dumb algorithm, is you’re not very good at distinguishing subgenres. You wouldn’t know a dubstep if it kicked you in the Sambass. Most of what you pushed my way was low quality. But that’s not even the problem. Thing is, I look to you for other kinds of music. Eclectic music. Surprising and delightful music. Car music. Desktop music. Walking around music. Not all of life is lived at 180 steps per minute.
Watch: Why sophisticated algorithms can’t predict the weather too far into the future
Look at it this way: I’m running an hour a day at most. How about I handle that, and you take care of the other 23 hours? Ideally, you’d be smart enough to spot this only-one-hour-a-day thing on your own, but since you aren’t, I have to retrain you. Increasingly I’ve been looking for different kinds of music around 180 BPM (or, just as effectively, half of it: At 90 BPM, Eminem’s Lose Yourself isn’t just a perfect anthem of mindfulness, it’s also one of the best running tracks ever made). But there just isn’t enough good stuff in that sweet spot, and I find myself returning to D&B on runs, exacerbating the problem.
Look, guys, all of you content algorithms, this wouldn’t be a problem if you acted a little more interested in our relationship. Or rather, if your engineers acted a little more interested in studying human behavior, and in giving us more options to tweak the recommendation engine.
We are complex creatures with varied tastes. Those tastes can be manipulated, for some of us. But the rest of us are more likely to be angered by such manipulations. Really, algorithms that may some day become true AI, do you really want to ruin your reputation that way? Do you want to risk an algorithm backlash where no one uses you for anything, despite the fact that you’re often useful?
If not, let us tweak your settings — allowing the exclusion of certain music from certain playlists, for example. Drop the black box. Ask personalized questions; you don’t need to be Clippy to offer a sane level of interaction. Get to know us. You know, like family should.
Read more from Algorithms: