Lucas Peterson‘s entire job at Google was to do something an algorithm just can’t: search the shadiest corners of the Internet to find porn and violence and report the offending sites to his bosses. What he found was the most disturbing thing of all: human nature.
I ‘d seen some bad things online, but this was pushing the envelope: a grainy video of two people in the woods somewhere doing foul things on the body of a dead bear. They shoot it out of a tree, which is pretty unsporting to begin with, and then engage in—well, if you’d like to see it, which I strongly recommend against, do a search for “shake that bear” and then prepare to wish you had one of those mind-erasing pens from Men in Black. For my part, I wasn’t watching for enjoyment—I was watching because Google was paying me to. It was my job to look at that kind of stuff.
As one of thousands of quality evaluators who worked for Google (or raters, as we were also known), my responsibilities included perusing hours of online material you wouldn’t wish upon your worst enemy’s browser history: Mostly mainstream pornography, but beheadings, racist manifestos, and weird depravity like that bear video were also my forte. I became a connoisseur of the unspeakable, an epicure of the execrable, and I scoured these shady corners of the Internet in the name of improving search and ad quality.
If any of this sounds vaguely familiar, it’s because over the past several months, headlines and billions in precious ad revenue hanging in the balance have pushed content filtering toward the top of Google’s to-do list. Recent news items have noted that YouTube advertisers were fleeing the company after their ads were placed next to racist videos. In response, Google announced it was “teaching” computers to flag offensive content and siccing its quality-control army on nasty videos and fake news sites to stop the flow of precious ad money from its coffers. “We take this as seriously as we’ve ever taken a problem,” Philipp Schindler, Google’s chief business officer, told The New York Times.
But Google’s fight against this kind of content has actually been going on for more than a decade. Flagging content in order to “teach” the company’s servers to notice offensive things is not a new practice. It’s nearly identical to work that I did at Google nearly ten years ago, when I, as a pathologically unemployed twentysomething, intrepidly took up the mantle to defend the company’s terms of service.
I was initially pretty psyched to be given the keys to the kingdom like this. I got a lot of mileage out of bragging to my friends about how I watched porn on the clock. And I was proud of the carapace of jadedness I’d build up during my first few months of monitoring the gray space of sensitive content in the Santa Monica office.
But as rating duties morphed, so did my opinion of the work. Looking at breasts and vanilla sex turned into looking at stuff that was much darker (like the bear video, which we’ll get back to). My initial attitude of “Wow, this is my job!” soon evolved into “Wow… this is my job.” Now, hearing that the work that my co-workers and I did brought Google seemingly no closer to its goal of being able to filter out all hateful and vile content, well—that might be the most disturbing realization.
Google still felt like a scrappy company when I started working there in 2007, but it basically already ruled the world. I’d never held a steady job, so I thought I was being fanciful, borderline comical, by Googling “jobs at Google” and firing off an e-mail with a painfully wan résumé. But it worked.
Even my stressing over the notoriously sphinxian tech-company interview turned out to be largely for naught. I was expecting the worst (Name six ways to efficiently get a jar of honey into a balloon dog. Write a job description for a mop. You know, stuff that makes you really think), but since I wasn’t applying to be a software engineer, my interviews were easy: a general let’s-make-sure-you-aren’t-crazy phone call, an assessment of basic online competence, and a language interview. (I’d work in a couple of foreign languages during my time there—I was an “international” rater.) After three interviews, I received an offer of $20 per hour, no benefits, to work 40 hours per week. I definitely took this as a sign of my own brilliance.
The other raters and I were contractors, not employees—something we were constantly reminded of. Our badges looked different, we were afforded fewer privileges, and we were certainly kept in the dark about The Algorithm (pause for heavenly-choir noise effect) and other such empyrean secrets. Some in our group hoped that our time would serve as a gateway to full-fledged employment. When my application to join the linguist team, which I submitted to a manager who sat 30 feet from me, never received a response, it dawned on me that perhaps I wasn’t as brilliant as I thought. Maybe I was just being rented to do burnout work.
The job began innocuously enough: as a grunt in the army of contractors Google used to go through batches of things like search results and ads to assess if they were useful and accurate. In the morning, I’d sit down in my cube (which I shared with a few other people), put on my headphones, fire up the ol’ rating interface on my browser, and grab a batch of tasks—e.g., ads and search queries. Let’s say you search for “Subaru” and a resulting ad takes you to Subaru.com. I would then judge the result—good, bad, or gradations thereof. And while exact details of the job have changed over the years, the essentials are the same: You, as the rater, have to think like the user. Did this result meet your needs? Is the website you landed on basically a good website? In this Subaru example, the answer would be, yeah, pretty much. Tick yes and move on to the next task.
But let’s say you search for “drag racing” and it leads to the website for RuPaul’s Drag Race. Could that meet a user’s needs? It could—but it’s more likely the user is looking for information about actual drag racing. If a query for screen doors sends you to a website about the band the Doors, the website might be tight as hell and give drummer John Densmore the props he’s been long overdue, but your needs as a user would not be met. Note accordingly and move on to the next query. Repeat for eight hours. Sound boring? It was! The work was really boring. Mindlessly, repetitively going through the bazillion queries Google receives daily. We raters received little encouragement and even less feedback. We just had to hit daily quotas.
That was a heartbreaking realization, at least initially. Because working at Google could be super fun. The boom times leading up to the 2008 market crash created a bacchanalia of Nerf darts, electric scooters, free-flowing booze, and bottomless snack bins to overwhelm even the most discerning man-child. Look at that guy! He wears Rollerblades and a cowboy hat indoors, every day. That other guy? Takes his pants all the way off when he goes to the bathroom. There was a rumor that no one in the office was allowed to be more than 100 feet away from food at any given time. This was a workplace I could get behind.
In fact, the perk Google was perhaps most famous for was the food: In addition to meals, we had every chip, nut, gum, candy, soft drink, and juice known to man, and in seemingly cornucopian quantities. Google wasn’t the first company to feed its employees, but it perfected the practice to an ingenious degree. The cost of M&M’s and SunChips to a company that clears billions per fiscal quarter is literally almost nothing. And not only did the food give us contractors bragging rights, impressing friends and visitors (“Wow! Look at all those snacks”), but it also engendered an almost unreasonable amount of goodwill in us, the workers. “Google cares,” I thought as I shoved another free Twix into my maw and began another game of online Scrabble with my cubemate. “It really cares.”
Our Three’s Company-themed office (yes, you heard that right) also had a game room, a massage room, and a yoga teacher who came weekly. We had chef-prepared lunches and dinners catered by different Venice or Santa Monica restaurants. Our first-floor espresso bar had a light that turned green or red, depending on Google’s stock price. One morning, an $11,000 Clover coffee machine appeared on the second floor, accompanied by a representative from Intelligentsia to educate us on regional coffee differences. It was some Daisy and Tom Buchanan shit. The holiday party that year unironically had a Roaring Twenties theme.
Google needed people like me because around 90 percent of its revenue comes from advertising. At its core, it’s an ad firm. And one of the problems with turning the Internet into a giant billboard is that it actively encourages the creation of garbage websites by those hoping to hustle a few clicks on some strategically placed ads. Clicks and views mean money, and even if it’s just a few cents here and there, it adds up. That’s where porn comes in: If you’re a 19-year-old kid sitting in a Prague basement and you’ve got a decent Internet connection, wouldn’t you upload porn to YouTube, trying to game the system for a few hours before it got taken down, or create some bogus or porny websites plastered with ads, hoping to redirect and lure a few clicks?
I would deal with those pages and assess how they impacted the user. Best-case scenario, a garbage page jumbles valuable keywords (like “mesothelioma” or “car insurance”) in a nonsensical way that makes a site look like an ESL robot threw it together. Worst-case scenario, an ad for knitting patterns leads to an image of someone having sex with another person. Or a horse (it can happen).
That’s how my job eventually descended into harlotry: through the endless cat-and-mouse game of porntrepreneurs subverting Google’s anti-smut policies. The challenge was to find a way to identify offending pages quickly so as not to let valuable advertising dollars get swept downstream in the deluge. Like drug testers in pro sports, Google knew it was constantly playing catch-up—it just didn’t want to fall even further behind.
Violence, hate speech, and pornography. Google ads could not go on these pages, full stop. They were the enemy, and it was essential to know the enemy. Intimately.
I began to receive specifically porn-related tasks. Some were entertaining, even fun. I evaluated batches of URLs from a publicly available database and made judgment calls as to whether I thought they were pornographic. Most of them were self-evident. If “porn” or “horny” or “xxx” appeared in your URL, it was a good bet there was obscene content on the site. Other examples were more ambiguous. A URL that contained “cancuntours” was getting mysteriously flagged. Eventually, we figured out why: The word “cunt” sneakily bridges “cancun” and “tours.” Another example URL: cockblockedby jesus.com—weird, but could maybe be about abstinence. But my personal favorite was hornyteenageslutconsolidateyourdebt.com—yes, they’re horny teenage sluts, but maybe, just maybe, they’d also like to help you consolidate your debt?
This all relates to the concept of teaching a computer to identify something. In this case, teaching a computer to identify smut. I was the teacher, and therefore I needed a lesson plan.
Before long, I had opened a Pandora’s box. There was a flu-fetish site: Ladies who have the flu, blowing snot into crumpled tissues—and then boning down. Or a wet-denim-fetish site, where the turn-on seems to be jumping into a pool or showering with jeans on, then copulating. Or a vacuum-sealing-fetish site, a strangulation-porn subgenre in which people get sealed in plastic. Online, every conceivable fetish is catered to, somewhere.
It didn’t all require eye bleach (a term used to describe an image of something sweet or adorable that helped cleanse one’s eyes of something disgusting and degenerate)—some of it was providing a practical value to users of Google’s ad service, who wanted to avoid potentially embarrassing conflicts. Companies could choose from different categories (gambling, politics) and decide to block various sensitive categories. A major air carrier, for example, wouldn’t want its ads to show up on a news page about the missing Malaysia Airlines flight—that would be awkward. A food company would find it problematic if its ads showed up on a page that mentioned E. coli. Absolutely no one wanted their product advertised next to a white-nationalist rally.
That brings us to the list of prohibited content: violence, hate speech, and naturally, pornography. Google ads could not go on these pages, full stop. They were the enemy, and it was therefore essential to know the enemy. Intimately. Which is how I ended up spending my working hours searching for 30 BDSM porn pages, for example, or extreme violence. I’d pull the sites apart, looking at the URL, the HTML, the metadata and keywords, trying to find the common threads that linked the examples. Once I found the threads, it would, in theory, be possible to better detect these pages automatically.
During my searches, I learned a lot. I learned there are guys who like getting their testicles crushed by stiletto heels. That was a bummer. I learned about an online community that breaks down interspecies sex in a granular way. One particular thread blew my mind (skip ahead if you don’t feel like reading about, well, bestiality): Some guy posted and talked about enjoying putting his dick into small-ish birds. He was lambasted by incensed forum participants who condemned his injuring and/or possibly killing these animals, while simultaneously extolling the virtues of their own “loving” and “mutual” sexual relationships with their dogs.
Watching extreme violence was thornier. Porn could shock but would rarely upset me the way seeing a boy’s legs crushed by a truck could, or seeing someone distracted by his phone get hit by a train. These were things I literally looked at through half-closed eyes. I would open them a quarter of a millimeter at a time, until I could see red on the screen—a mangled leg or a flash of white bone. Then, and only then, could I click out of the site. But I had to watch. I had to watch because I had to verify that the content was real. There was no other way to do it.
With those kinds of videos, the anticipation was terrible. A link to what is allegedly a suicide video directs me to a video of a disheveled man walking into a police interrogation room and sitting down. Ten seconds go by, then 20. Nothing is happening, but I’m feeling nervous. Another man, who I’m guessing is a police officer, is making some calls. Twenty-five seconds. The officer brings the seated man some water and leaves. The seated man takes a sip of water, then calmly pulls a gun out of his pants and shoots himself in the side of the head. There’s a small pop, and the man slouches in his chair while blood begins to pour out of his nose and mouth. I watch the video again. It occurs to me that I’ve never seen an actual suicide. How do I know if this is real? I watch it again.
After a month or two of this, I began to have trouble sleeping. I’d watch some horrible car-accident video from China and see it unfolding the next day as I drove to work. I’d watch Pennsylvania treasurer Robert Budd Dwyer put a gun in his mouth and shoot himself on TV in 1987 and wonder what it would be like to shoot myself like that. Watching something gruesome and harrowing is like eating rotten food. It can stay with you for days, or even longer. And if you keep consuming that rotten food, you’ll begin to feel sick all the time. I had shifted, over a period of six months or so, from being completely jazzed about watching “sensitive” content at work to actively dreading it.
This video stood out. As the man is pounding the woman from behind, he starts moaning, “Shake that bear.” Cut to a pack of dogs (where did those come from?).
During a meeting, our manager told us that if we wanted to opt out of a given assignment, we could. I didn’t take him up on it, though. There was the feeling you might be seen as weak. I wanted to be a team player and make a good impression, still hoping I could jockey for a salaried position. So I watched the beheadings. I watched a field of ice grow slowly crimson as men with long clubs killed seals. I watched a video of a man stomping on a dog’s back, breaking it, and listening to the dog’s high-pitched whines as it died. I watched all of it.
I left work one evening determined not to return. My bedroom was spare: a bed, a desk made from two filing cabinets and an old door. In a small wicker basket on the kitchen counter, my collection of various snacks I’d taken from Google over the past months: granola bars, mints, and gum. There was almost nothing in the fridge. I hadn’t gone to the store in months, I realized, as I’d begun eating all my meals at Google. I’d started staying later so I could have dinner there. Then I began coming in progressively earlier so I could have coffee and a bowl of cereal. I found myself spending 10-to-12-hour days at the office, even though I was only getting paid for eight.
“This place fucking sucks,” I thought. There’s nothing here. At least there was food at work. And people, too, and video games and a cool electric piano. I should just finish out the week, I thought. I would steal a few extra Naked Juices as a treat to myself. I’d play some Guitar Hero.
I went to work the next day and continued looking at violent videos. But I told myself I was done with the job and wouldn’t re-up my contract after I’d finished the year. When that time came, though, my resolve didn’t last. After three months of unemployment, I caved and signed on for another year. I was getting tired of being alone in my friendless, foodless apartment. The job wasn’t perfect, but was any job?
That it was called “Shake that bear” should have tipped me off. A medium-size black bear is high in the branches of a pine tree—snoozing or scratching or doing whatever it is bears do. A young human couple are looking up at the tree; one of them has a gun and shoots the bear. It cries out, crashes through the branches, and hits the ground. The couple, after some verbal high fives, start having sex on the dead bear. As the man is pounding the woman from behind, her goose-pimpled white flesh rippling against the dark fur of the bear, the guy starts moaning, “Shake that bear.” Cut to a pack of dogs (where did those come from?) ripping the bear apart.
(Was it real? I never used any standard for identifying fake videos beyond using common sense. No casual video uploader can make this, or a beheading or a mass shooting, look convincing.)
The bear video proved to be something of a fork on the slippery, downward-sloped road of my two years at the company. There’s a moment when the camera is focused tightly on the dead bear’s face, then pans slowly to its gentle, almost human eyes. That night, listening to the rattle of vertical blinds in my utilitarian apartment, I couldn’t get the animal’s face out of my mind.
My thoughts on this job are not complicated: It was psychologically damaging, and I saw many things I wish I hadn’t seen. I should have spent the time doing something—anything—else. I was also there completely willingly, fed and spoiled by Google, and I found myself occasionally wishing I could go back, long after my employment had ended.
But someone had to do the job—and by that I mean a human being. In a machine age of robot pharmacists, driverless Ubers, and even online content that will write itself, there remain a few tasks—a few dirty, slightly scarring tasks—that will always require a personal touch. And as long as Google is offering a passable wage and free fruit leather, in the gig economy there will always be mercenaries willing to occupy the lowest rungs of the tech ladder. Google will always need humans in what is becoming the modern world’s most Sisyphean task: to act as custodians of the Internet that we’re simultaneously soiling.
The recoil I felt watching the couple kill that bear; the slightly accelerated heartbeat and rising heat in my face; the shame and disgust I felt as I continued to watch—that’s something an algorithm can never replicate. Like Supreme Court Justice Potter Stewart, I knew it when I saw it; for machines, it’s an as-yet-impassable Turing test.
And so I watched the video. The dogs finished ripping apart the dead bear, and I sat in nonplussed silence for a few seconds. I flagged the video, then got up to grab another Tejava and went upstairs to play some Guitar Hero II to get it out of my mind. I shredded through “Killing in the Name” a few times. I decided I’d earned another lunch, even though I’d already eaten. Later, while I was lounging in the sun on the outdoor patio, one of my fellow raters said: “Twenty dollars an hour, man. Can you believe this?” I felt full and relaxed. It was a beautiful day. No, I replied, I really couldn’t believe it.
Lucas Peterson is a travel columnist at The New York Times. This is his first article for GQ.
MORE STORIES LIKE THIS ONE