How the Algorithm Rewards Extremism

Clive Thompson on Big Tech, the Internet, and the Mess We're In

I could say, again, that software is eating the world, though it might be more accurate at this point to say it’s “digesting” it. But what’s noticeable also is the fact that size matters. These days, some of the biggest civic impacts come from the truly titanic, globe-spanning tech companies that sit in the midst of our social and economic life. “Big Tech,” as the journalist Franklin Foer dubs it.

Indeed, there are now a surprisingly small handful of firms that dominate the public sphere. There are the ones that govern how we communicate (like Facebook, Twitter, YouTube, Apple, and Netflix), ones that touch commerce (Amazon, Uber, Airbnb), and the information brokers and toolmakers of our work lives (Google, Microsoft). Big tech is a useful way to think about the particular challenges of software that dominates its area, because it highlights the near monopolies many of these firms enjoy. And they’re mostly extremely young, new companies. Many rose to dominance in barely more than a decade. Their histories are marked by frantic, metastatic growth.

This is not surprising, because it’s in the nature of software itself. A software firm ships code, and code is a historically weird type of product. It’s a machine that does things but which can be replicated globally for little-to-zero additional marginal cost of distribution. It’s as if Chevrolet could design a single Camaro and then instantaneously teleport 200 million copies to the driveways of every household in America. This is a fact that strikes, and occasionally even stuns, the engineers for big firms.

At one point while writing this book, I visited with Ryan Olson, a lead engineer for Instagram, right after his team had just pushed out a massive update (introducing the wildly popular video Stories, cribbed from their rival Snapchat). Olson told me about how, a mere hour or two after their update, he’d been traveling around San Francisco—in bleary, post-crunch exhaustion—and noticing everyday people using his fresh, new code.

“It’s a pretty cool experience,” he said, “to be riding on a train, or last night I was at the climbing gym, and I looked over and someone is using the product. I don’t know if there’s ever been historically any other way where you could reach so many people”—or where “so few people define the experience of so many.” The thrill of overnight growth is vertiginous, powerful, and addictive. It’s why so many coders—particularly those making consumer products—have a holy reverence for scale. They love the idea of creating something that grows at an exponential pace: It’s used by two people, then four, then eight, and soon the entire damn planet. Why, if you can spread your creation around the world so easily, would you ever want to do something small? Isn’t there something kind of sad about a piece of code that doesn’t grow at a frantic, kudzu-like pace?

Indeed, among the reigning kingpins of Silicon Valley there’s a sort of contempt for things that fail to become massive. Smallness seems like weakness. You may recall the story of Jason Ho, the hacker who created a thriving small business by making time-clock code used by companies around the world. It made so much money that he was able to spend much of his twenties with the freedom to travel and invest. If I’d done that, I’d certainly consider it a success myself.

But when I mentioned Ho’s company to the thirty-something founder of a very large tech firm, he scoffed. To him, it was “lifestyle business”—Silicon Valley–speak for an idea that will never scale into the stratosphere. That sort of product is fine, sure, he told me, but Google could do the same thing and put him out of business in a second. If you weren’t aiming to be giant, he asked with a shrug, why bother doing it?

This sentiment is arguably even more pronounced in other software markets like China, which has a famously competitive, winner-take-all tech market. When in 2015 I toured the offices of the e-commerce firm Meituan in Beijing, the company was only five years old but in a frenzy of expansion, hiring young engineers as rapidly as they could roll off the transom of computer science programs. The CEO Wang Xing and I peered out over the sprawling floor of coders, festooned with hundreds of plants to make the scene feel less sterile.

“In China, you either have to become massive or you will get crushed,” Wang told me soberly. (Meituan alone had survived probably a few thousand competitors, as the tech investor Kai-Fu Lee estimated, when I spoke to him.) In the world of high-tech firms, the race to scale is propelled by a carrot (the magical ease of duplicating and running code worldwide) and a stick (the shark-like competition).

If you weren’t aiming to be giant, he asked with a shrug, why bother doing it?

The lust for scale is also fueled by the dictates of venture capitalists. They place their bets on dozens or hundreds of companies, encouraging them all to grow ferociously. The vast majority won’t, but with luck, one or two will break out—making so much money, so quickly, that it makes up for all the other losses. Venture capital is thus perfectly content to accept an ambitious flameout. It adores a sudden, exploding success. But the one thing it finds useless and annoying is a company that’s merely stable, maybe growing a small bit. Even if that firm is making a little profit, who cares? The investor isn’t looking for stability: they want rapid growth that leads to a bigger return on their investment.

The Y Combinator accelerator—which takes in several dozen tech firms each year, to try and help them into the big leagues—ends each cohort’s program with a Demo Day, where the young companies show off their products for a room of handpicked venture capitalists. The start-ups are inevitably desperate to include in their presentation a hockey-stick chart—the one that shows their user base suddenly blasting off into the sky.

One evening, I visited the hacker-house of People.ai, a company that just days earlier had done their Y Combinator demo. They pecked at keyboards and exhaustedly described how they’d spent the three months in Y Combinator frantically registering new clients for their service, in an attempt to produce that hockey stick. “You think about it, the three months it’s all about building the numbers—but you’re going to show them off for only 10 seconds, on your ‘growth’ slide,” Oleg Rogynskyy, the cofounder, said.

Kevin Yang, the lead programmer and cofounder, laughed while remembering the investors sitting there, arms crossed, awaiting the growth figures. “Is that hockey stick not hockey stick enough?” he joked.

“The X axis has to be half the page,” Rogynskyy said.

*

Scale, of course, brings enormous benefits. It’s certainly financially valuable for the big tech firms! If they grow fast enough, they scare off competitors and develop the lock-in of “network effects.” When a social network like Facebook or WeChat gets big enough, users can’t easily stop using it, because all their friends are there. And certainly, when a tech firm grows rapidly it can be enormously beneficial for users, too. Because of Facebook’s global ubiquity, it’s now the easiest way for people to organize virtually anything, large or small, from family meetups to political fund-raising campaigns to search-and-rescue efforts. The new attention to police abuse of power in recent years? It’s been fueled partly by the commanding size of Facebook and Twitter—which lets users rapidly spread video of horrifying and incontrovertible examples, including livestreamed ones. It is these firms’ huge footprint that permits everyday people to wield it as a broadcasting network.

But the frantic drive for scale also changes software firms. It inexorably pushes them toward tactics that range from dodgy to exploitative. After all, to scale at such a ferocious clip, you can’t charge your users any money up front. The service needs to be “free.” This is particularly true for social networks: They can’t get a million users overnight if every user has to shell out, say, $10 to join. So the only other way to make money is to get as huge as possible, then sell advertising to your audience. Facebook and Twitter and Google have all adopted this free-to-use model—indeed, Facebook boasts on its sign-up page that “It’s free, and always will be.” And the ad market has been deeply lucrative for them: In 2017, Twitter’s revenues were $2.4 billion, Facebook’s were $40.65 billion, and Google dwarfed them both with over $100 billion.

Yet advertising changes the nature of how software firms treat their users—something that many coders and designers, deep inside the bowels of the companies, began to uneasily apprehend.
One such techie was James Williams. A thoughtful, philosophical guy who’d studied English in college before earning a master’s degree in product-design engineering, he’d joined Google in the mid-00s to work as a strategist on the firm’s search advertising systems. He was drawn in by the mission of improving people’s access to information. Googlers talked about that mission all the time in soft-glow terms, and he loved it. “The default view was that ‘more tech is better,’ ‘more information is better,’” he notes.

But Williams eventually began to notice the same side effects that had perturbed Leah Pearlman and Justin Rosenstein, the pair who helped invent Facebook’s Like button. Like them, Williams noticed that any tech firm selling ads inevitably becomes motivated to keep its users staring endlessly at the app. After all, you can only deliver ads to someone while they’re staring at your service. So you quickly begin building as many psychological lures as possible into your code.

The big tech firms would pepper us users with alerts, trying to interrupt us during other tasks, to get us to come back to the mother ship. They’d slap little “quantification” numbers everywhere, to stoke our curiosity and our desire to “clean things up”: You have 14 new items in your feed! What could they be? And they’d make all these alerts bright red, to increase the chance we’d pounce on them. These trends, Williams argued, went into overdrive after the iPhone emerged.

If users click on it, it must be what they want.

“Before mobile, the internet was bounded in a place, because you could step away from it and close the laptop,” he tells me. “But once it was in your pocket, it was a firehose.” It is easy for engineers, Williams realized, to justify these psychological tricks—to argue they’re good. After all, they’d test each new tweak and trick by using A/B tests: Make the alert red, make it yellow, and see which one users click on more often. Red wins, so it must be the right choice! This data-driven form of design can make each psychological trick seem objectively correct: If users click on it, it must be what they want.

To the scale-driven engineering mind, the ethical questions of “What should we be making?” are easily subsumed into the sheerly technical question of “What will help the system grow more and have a bigger throughput?” One anonymous former Facebook employee put it neatly, in a comment to BuzzFeed: “They believe that to the extent that something flourishes or goes viral on Facebook—it’s not a reflection of the company’s role, but a reflection of what people want. And that deeply rational engineer’s view tends to absolve them of some of the responsibility, probably.”

Once advertising and growth become the two pillars of a big-tech firm, then it’s nearly inevitable that they’ll seduce their users into endless, compulsive use—or “engagement,” as it’s euphemistically called. “You’re trying to manage your attention, and they have some of the smartest people in the world trying to distract you,” as Williams says. The end result, he decided, is there’s a fundamentally adversarial relationship between the goals of the coders and designers and those of their users. The former are constantly trying to trick and nudge users into compulsive behavior. It works because the nudges are subconscious, or algorithmically invisible. If they were more obvious, we might reject them. Imagine, Williams says, that GPS worked in a similarly adversarial fashion. You’d ask it to take you home, and it would insert five detours along the way, to bring you past locations that satisfy the needs of advertisers.

Even worse, the dictates of digital advertising have led to a ceaseless tracking of our individual activities online. If a tech firm is offering advertisers the ability to custom target me, they want to know as much as they can about me: what other websites I surf, what neighborhoods I visit, what keywords occur in my emails and public postings. The advent of deep learning makes tech firms even hungrier for more of our personal info, because deep learning works best when it has mammoth amounts of “training” data, the better to predict what ad we’d like to see or what mood we’ll be in on Mondays. This has produced the world where Facebook even collects information on phone calls you’ve made on your smartphone, as the novelist and University of Houston professor Mat Johnson discovered (“cool totally not creepy,” he joked on Twitter.)

Clive Thompson
Clive Thompson
Clive Thompson is a longtime contributing writer for the New York Times Magazine and a columnist for Wired. He is the author of Smarter Than You Think: How Technology is Changing Our Minds for the Better.





More Story
Our Software, Ourselves: Who Really Writes Our Narratives? Never before in history have there been so many ways to recall our lives and the lives that surround us. That’s not exactly...