Instagram has announced a proposed change to how its photo feeds will be filtered — algorithmically, instead of chronologically, meaning the latest images won't always appear at the top. The proposal set off an uproar from users fearing that the app would become part of a social media algorithmic apocalypse.
If you're on Instagram you've probably noticed your favorite posters calling on followers to turn on notifications. That's to make sure fans don't miss posts the algorithm may not show.
"There's genuine anger and genuine resentment among users of these services," says Slate senior technology writer Will Oremus. "When these changes happen, they don't like the fact that some engineers in Silicon Valley are deciding for them which friends' posts they see at any given time."
Facebook uses a similar algorithm to rank posts in its users' news feeds. Hilary Mason, a data scientist and the founder and CEO of Fast Forward Labs in New York, explains the thinking behind using an algorithm to sort Instagram.
"An algorithm is really just a recipe," Mason says. "In this case it's a mathematical way of processing the data and what they're doing here is actually not that complicated. They're looking at the things that cause you to engage more, to like photos, to spend more time looking at them. They're using that to train a system that will predict which photos in your feed you're going to like the most and then they're going to show you that set of photos rather than just the chronological set of photos you would have seen otherwise."
Most people, however, seem to prefer Instagram without the algorithm.
Science Friday polled their Twitter followers on the subject, and 92 percent of those who responnded said they preferred a chronological timeline in Instagram. Only 8 percent said they wanted an algorithmic filter.
"People want to be in some sense of control ... over what they see in their social network," Mason says.
"One thing we've lost over here is that the algorithm is not transparent. It is by definition a black box and, you know, if they told us what was going on we'd exploit it so they can't do that. And we can worry about what biases the creators of that algorithm, the people doing the feature engineering might have," she adds. "Like think about if you know a Facebook machine learning engineer really likes dogs and really hates cats. And that gets built into the algorithm. Is that a world we're happy with? It's very unlikely. But still these are questions that are reasonable to ask because we don't know what's happening under the hood."
Still, Oremus argues, social media companies would not be implementing algorithms if they didn't think it would help people enjoy their feeds.
"Facebook has huge teams of data analysts who are looking all the time at this immense array of metrics and every time it changes anything in the algorithm it's going to look at 15 different ways that that's affecting people's engagement. And what the data in aggregate tells Facebook again and again is that, if you use the right data and you tweak the algorithm in the right way, that it will keep people coming back."
So what sorts of data are Facebook and Instagram using to make up their filters? Oremus says they're constantly analyzing people's reactions from dozens of different angles.
"Facebook's algorithm is only as good as the data that goes in. That's why it's always studying your behavior on the site," Oremus says.
"It's looking at what you click on, what you like, what you share. Over time it's been looking at more and more advanced metrics. Like when you click on a post in your feed, how long do you spend reading that post on a different web site before you come back, and do you hit like on it before or after you've read it? ... One of the things they've turned to recently is actually a more personalized form of data. They have now hired thousands of people around the world to go through their Facebook feed every day and say which stories they actually would like to see at the top of their feed."
Mason says the data collected by Facebook and Instagram tend to be incredibly accurate, but only on a broad spectrum.
"One of the really common misconceptions about the sort of machine learning is that it can perfectly model your individual behavior as a unique human being into the future, and that's really not the case," Mason says.
"These things tend to be quite accurate at the population level or when doing an aggregate analysis, but that doesn't mean that it can say that you specifically are going to engage. Just that you may be more likely [to] because you're part of this population."
Readers, what about you? What's your opinion on Instagram's new algorithmic sorting? Leave your comments below and let us know.