When her husband lost his factory job in 2010, Kristy Milland ran through her options.
Until that point, she’d been working at home, earning extra money through odd jobs like selling collectables on eBay. She hadn’t waited on tables, had no experience in fast food, and had not learned any skills that might be particularly useful in a factory. She’d once applied for a job at McDonald’s, but nobody had called her for an interview.
Jobs were more difficult to find in her hometown of Toronto since the beginning of the Great Recession. But there was one place where Milland knew she could get work immediately.
Launched in 2005, Mechanical Turk is an online “crowdsourcing” marketplace run by Amazon. Its clients post work tasks on a dashboard that a “crowd” of independent workers can choose to complete, often for cents apiece.
Amazon had not launched the Mechanical Turk platform with a promise to create work, in the way that Uber would later brag about adding “20,000 new driver jobs” to the economy every month. Rather, it had built the website as a way to integrate human intelligence with code—as a service for programmers. As TechCrunch’s founder put it shortly after the product launched: “[It] is brilliant because it will help application developers overcome certain types of problems (resulting in the possibility for new kinds of applications) and somewhat scary because I can’t get the Matrix-we-are-all-plugged-into-a-machine vision out of my head.” He called the workers who would be plugged into this Matrix “Volunteers.”
Some common tasks on Mechanical Turk included labeling photos that are used to “train” artificial intelligence, filling in spreadsheets with contract information, or writing product descriptions for websites. An entrepreneur once pitched me an app that—through his proprietary system—would provide accurate calorie estimates for meals based only on a photo. Sure enough, shortly later, I found a posting on Mechanical Turk for the company that asked workers to label the food. The technology was humans. But it looked like magic.
About the Author
Sarah Kessler is an editor at Quartz. Previously she was a senior writer at Fast Company, and her reporting has been cited by New York magazine, The Washington Post, and NPR.
Some of Milland’s more remunerative work came from employers who posted hundreds or thousands of tasks at a time that could be completed in rapid succession. Milland would install small software programs that allowed her to complete, say, a simple categorization task by hitting a key on her keyboard (“y” for yellow or “b” for bird) rather than clicking a mouse. Categorizing an item every five seconds for an hour, at $0.03 per image would pay $21.60 per hour. She also took on more complicated tasks. Writing descriptions for product sites, for instance, could pay $1.50 per paragraph. So if she did one every five minutes, she would make $18 an hour. It was a matter of doing the work quickly and sticking with it for a long time.
Turker Nation, a forum where Milland was a moderator, had a place where Turkers alerted each other about these “good work” opportunities, which paid well and could be completed in large batches. To make sure that she didn’t miss any of them, Milland set up an automated system that, when a new “good work” task was posted, would check to see how much it paid and whether she met its qualifications. If she was eligible for a task that paid $0.05, her computer would alert her with a “ping” noise. If she were eligible for a task that paid between $0.05 and $0.25, her computer would sound an alarm that sounded like a laundry machine finishing. If she were eligible for a task that paid more than $0.25, a siren would sound.
No matter where Milland was in her house, if she heard the alarm go off, she would run to her computer. There were thousands of other Mechanical Turk workers competing with each other to grab the high-paying work, which was assigned to whoever could claim it first. Milland would sleep in her office so that she could listen for the alarm to go off at night without waking her husband. When she spotted good tasks, often through her alarm system, she used an automated tool to keep her queue full with the maximum 25 tasks that could be assigned to her at one time, and then worked furiously to finish them and grab more before they were snatched by other people.
She created alerts with different sounds: Tasks paying 5 cents prompted a “ping.” From 5 to 25 cents, a laundry-machine alarm. If a task paid more than 25 cents, a siren would sound.
One of tasks she didn’t like to miss was answering questions from a Q&A service in which people, in the days before mobile internet browsers, could text a number with a burning inquiry such as “Where is the nearest Italian restaurant?” These were posted every 15 minutes, and there were a few different aspects that made them good tasks. The first was that people often asked the same questions, and Milland had compiled a spreadsheet of answers that made these common questions quick to answer. She could get through a batch of several hundred in about five minutes. The second was that to incentivize good work, each month Amazon paid a bonus of a few hundred dollars to the worker whose answers received the highest number of “thumbs up” votes from users. Each question might only pay pennies, but this bonus was significant. It meant that Milland never wanted to miss a batch. Her routine was to listen for the alarm, complete the batch in five minutes, take 10 minutes off, and then get back to work when the next batch of questions dropped.
Another task that she prioritized was one that asked customers to take photos of products in order to find the same product on Amazon. It was how the company encouraged customers to comparison shop, but not everyone used this feature of Amazon’s app as intended. When they sent pictures of genitals, Milland sent back a link to a book called I’m Calling the Police. It was worth dealing with rude inquiries like these because she had found a way to earn extra money when Amazon’s users sent photos of actual products: She sent them her affiliate marketing links and earned a percentage of purchases they made after clicking on them.
Milland also started proactively asking requesters if they needed help designing their requests, contacting them through the Mechanical Turk site. Sometimes she collected consulting fees for teaching them how to improve their results.
The paradigm on both the employer and worker side of Mechanical Turk was less of a relationship between two colleagues than it was two people trying to beat a system. In one common example, companies posted the same work three different times on Mechanical Turk in order to check the accuracy of responses. If one worker submitted a different answer than the other two who completed the task, the company assumed that worker had given a bad answer and rejected her work (which meant she wouldn’t be paid). To beat this system on the worker side, all a Mechanical Turk worker needed were two accounts to agree with each other. Some Turkers built automated bots to submit arbitrary (but matching) work results. The bots collected payment because they agreed, while the person who had earnestly done the work didn’t get paid. All Turkers could do when someone rigged a task this way was to tell each other to avoid it.
She opened a task to find a slide show of still shots taken from ISIS videos. People kneeling next to an explosive wire, preparing to die.
Milland didn’t feel like she could leave her apartment, or even her computer, lest she miss out on an opportunity to work on good tasks. Unlike an employee at a fast-food restaurant or a cleaning company, she didn’t get paid for downtime, and she could earn more money by working smarter and faster. The psychology was that of a game that required her to be constantly on alert. In a way, that psychology kept her going: She’d set a goal for $100 per day, and, cent by cent, she often met it.
What It Costs
Making $100 a day on Mechanical Turk was possible, but not always pleasant. There were times when Milland stumbled into emotionally taxing work that in a regular workplace would have come with preparation and consent. On one such occasion, she opened a task to find a slide show of still shots taken from ISIS videos. People kneeling next to an explosive wire, preparing to die. A wicker basket full of human heads. It came with instructions similar to any other photo tagging job. Another slideshow contained photos of animal abuse so graphic that years later Milland had trouble taking her dogs to the vet without crying.
The only indication that something exceptionally graphic could be found inside a task was often an “adult only” qualification. Employers used this designation on any jobs that involved user-generated content that they couldn’t control. Such jobs could pay well, and they most often didn’t contain anything disturbing. Milland considered them worth taking. And so she accepted that psychological stress would be part of the job.
So, too, would physical stress. She’d ignored the small, hard bump that had developed on her wrist until it started to grow. It got a little bigger every day, until eventually it was the size of a marble and interfering with the way she held her mouse. When she finally went to the doctor, she learned that it was a ganglion cyst. He recommended surgery, but that would mean post-surgery prescriptions not covered by Canada’s state health insurance. Another traditional treatment for “Bible bumps,” as they are sometimes called, is to hit them with a heavy book. And so one day, when she couldn’t stand it anymore, Milland gave the bump a good bashing.
Eventually the pain went away. Then a new pain appeared, up her wrist, toward her elbow. A neurologist told her that it was carpal tunnel and a repetitive strain injury. The ideal response would have been rest. But there is no workers’ compensation in the gig economy. There is no paid sick leave in the gig economy. And among US workers who rely on sites like Mechanical Turk for their entire income, almost 40 percent don’t have health insurance. Milland lived in Canada, with universal healthcare, but she couldn’t afford the break. She wore a wrist brace and an elbow brace and kept on clicking.
Mechanical Turk was the only option when Milland’s family needed money quickly. It allowed her husband to go back to high school to get his diploma. And it allowed her to earn an income from home without a college degree or a thick resume. She made it work.
But when, two years after she’d been working on Mechanical Turk, Milland’s husband finally landed a job, as a forklift driver at a printer company, she told him she never wanted to depend on Mechanical Turk again.
Shortly later, she applied for university.
Adapted from Gigged: The End of the Job and the Future of Work by Sarah Kessler. Copyright © 2018 by the author and reprinted with permission of St. Martin’s Press, LLC.