What it’s really like to be one of the ghost workers on Amazon’s Mechanical Turk
Joan wears her hair in a loose bun skewered by shiny black chopsticks to keep it out of her eyes while she’s working. She’s been living in Houston since 2011, when she returned to care for her 81-year-old mother. Joan cooks, keeps house, and drives her mother to doctor appointments. And for the past three years, she’s made most of her income working on Amazon’s Mechanical Turk.
Before moving back to her hometown, Joan had a full-time job as a technical writer. She drafted and copyedited, among other things, manuals for filing for unemployment insurance in the state of Texas. At first, Joan lived off the money she’d cashed out of her 401(k) plan. But as her mom’s health worsened, Joan looked for work she could do from home. On-demand work seemed like a good fit. Joan turned a spare bedroom into a home office, crowding the small room with a weathered brown chair, computer desk, and large monitor. Then she started searching the internet for work that she could do online.
Joan can’t remember how she first found out about Amazon Mechanical Turk, but she suspects that she learned about it on Reddit. Reddit is one of several online communities where people doing ghost work share tips on how to get started. As a 39-year-old white woman with a master’s degree in communications, Joan is, in some respects, a typical MTurk worker. Almost 70 percent of workers have completed a bachelor’s degree or higher educational attainment. In other ways, she stands out. MTurk workers skew young: 76.9 percent are between the ages of 18 and 37, that bracket of years when people are most actively seeking their first career-defining job.
Though Joan doesn’t recall all the details, the process for setting up a worker account hasn’t changed since she first signed on. She would have gone online, navigated to the main MTurk website, and clicked on the sign-up button. As a newcomer, Joan would have been asked to enter a verifiable name, email address, and password. From that point, she would have been given access to the behind-the-counter side of the site. Visible from Joan’s “dashboard” would have been dozens of tasks. Tasks, or what Amazon refers to as HITs (Human Intelligence Tasks), are jobs for hire. If she had clicked on a task, it would have shown her a short description of what the task required, the deadline, and what it paid. She could click on and complete a task, but, as a new user, she’d need to wait until her account was validated to get paid. Before Amazon pays a worker, it verifies the person’s physical mailing address, national identity, and bank account information. That’s how easy it was for Joan to join the ghost workforce.
To a new worker like Joan, MTurk’s dashboard can look chaotic. One sees multiple expandable menu tabs, including a tab to keep track of one’s account, another to track the individual tasks, and a tab that lists the worker’s “qualifications.” That word doesn’t align with skills. In the world of MTurk, qualifications can be things like a worker’s age, gender, and location. People who post jobs on Amazon use “qualifications” to restrict the type of worker who can accept the job. For instance, if an advertising company is looking for a focus group to give it feedback on a product meant to appeal to women in their forties, it might add qualifications such as gender and age to the job. It can even pay Amazon an extra premium fee for workers who list qualifications like “smoker” (30 cents) or “2016 voter” (10 cents). When Joan first looked at her MTurk dashboard, she remembers, she felt a bit overstimulated, but not deterred. “I thought, Okay, this is not going to pay out at the beginning, but if I do it for a while, it may become a decent source of side income,” she says.
No one knows the exact number of people who use MTurk, but typically about 2,500 workers are actively either searching for tasks or completing tasks on the platform. Because no agency— like a labor union or the Department of Labor—tracks this information, big-picture numbers are equally hard to pin down. Amazon maintains that it has 500,000 registered MTurk workers. According to researchers, anywhere between 100,000 and 200,000 people are registered to work on MTurk. Panos Ipeirotis, a leading researcher most known for his work tracking the ebbs and flows of MTurk worker demographics, estimates that 2,000 to 5,000 workers can be found on the MTurk platform at any given moment. That is roughly the equivalent of a 10,000-to-25,000-person full-time workforce. If we apply this logic to every on-demand platform, there are potentially millions of full-time jobs in the shadows of ghost work. This, of course, assumes that people would want to do this work full-time. However, as will become clear, a sizable percentage of workers stick with on-demand ghost work precisely because it does not demand a full-time commitment.
We posted a task on the MTurk platform to understand how workers are distributed around the world. Upon accepting the task, workers were shown a Bing map of the world and told, “Just double click your location and submit the HIT — it’s that simple.” Over ten weeks, 8,763 workers across the globe self-reported their locations. Workers are distributed throughout the United States in both highly and sparsely populated regions, but Indian workers are concentrated in the southern part of the country, a point we’ll return to in the next chapter (see figures 1A and 1B). Like most workers we met, Joan starts her day looking for tasks. One of the tasks she does the most is text categorization. She might read a snippet of text, perhaps a sentence or two from a news story, and either create a category for it or pick “politics” or “sport” from a list of options presented to her. The first time we spoke with Joan, she was doing one such task. For every data point she categorized, she made two cents. She classifies tens of thousands of pieces of text every week.
Joan spent the first six months on MTurk finding her footing. In time, she learned that the trick to making decent money was to quickly find doable work and to evaluate the requester offering the job. She noticed that on MTurk, every second counted; a slow internet connection, time spent finding work, or any unplanned downtime was the equivalent of lost income. In her first year on MTurk, she made $4,400. Some people might see that number as insignificant, she says, but “$4,400 is a meaningful amount when your previous income was zero.” Two years later, her MTurk earnings had almost quadrupled, to $16,000. Joan is now among the 4 percent of MTurk workers who are skilled, practiced, and lucky enough to earn more than $7.25 an hour completing tasks.
Hypervigilance is a necessity for top earners. Those doing ghost work who make the most money spend hours monitoring their dashboards and scrolling through pages upon pages of job postings. Joan, like so many others who are trying to make MTurk a core source of income, turns to free software tools and workers’ online forums to reduce some of the search costs that are an unpaid part of the job.18 They must be ready to snap up a well-paying or fast-and-easy task the second it pops on their screen, lest another worker click on the link and accept it first. “I’ve worked harder at this than I ever did at any office job,” she says. To enhance her speed, Joan arranged her web browser’s display of the MTurk dashboard to show 25 tasks or HITs at a time, and she uses keyboard shortcuts she created to flip through the pages quickly.
When Joan is in the zone, she can complete about 1,100 HITs an hour, netting roughly $22 an hour. She knows that people are likely to assume that the work is mind-numbing, but she finds the variety of tasks intellectually stimulating. She especially enjoys work that involves editing, which plays to her strength as someone with a background in technical writing. “I’m good at it and it’s easy to do,” she says. When the work does feel mundane or repetitive, she stays alert by listening to techno music or watching television. When we spoke, she was working her way through several seasons of Top Gear, a show for car lovers. “People talk about ‘Netflix and chill,’” she says, “but I watch Netflix and MTurk.”
MTurk set a minimum fee for work at one cent per task and, from there, requesters decide how much to offer MTurk workers for each assignment. On average, requesters price tasks to offer the equivalent of $11 an hour, but low-paying requesters flood the market with minimum-fee work, which drags down the overall earning potential of workers, who must wade through lists of poor-paying tasks to find decent work. “It’s a constant race to the bottom,” says Joan. By some estimates, the total revenue of the requesters on MTurk and similar sites like CrowdFlower adds up to $120 million per year. Workers keep what requesters pay out, but Amazon charges the requesters 20 percent of what MTurk calls “the reward”—a worker’s paycheck, including any bonus amount (the equivalent of a tip)—as its fee for operating the platform. Amazon charges an additional 20 percent for HITs that require ten or more workers.20
Unlike a traditional employer-employee relationship, MTurk workers are largely anonymous and mostly autonomous, meaning that a requester cannot specify the people who will carry out the work nor dictate exactly how the task is completed once it’s been accepted by a worker. Workers alone are responsible for the taxes on their MTurk income. They are expected to file as independent contractors, the 1099 forms familiar to anyone in the freelance consulting world. The trade-off for the requester is that the work is done fast and without the associated costs of officially hiring an employee. The trade-off for the worker is that they don’t have to stick with the same job any longer than it takes to complete the task. They can fit work around the demands of their lives rather than hand their lives over to the long commutes or hostile environments that come with some nine-to-five jobs. And they can stop working the second they’ve made the money they need to make. But the completion of a task does not always equal a payday. The work submitted by MTurk workers is reviewed by a human being or an algorithm that either deems the work satisfactory or rejects it. If the work is rejected, the worker isn’t paid. Each worker’s approval rating, the fraction of tasks they have had accepted, serves as a reputation score on the site. Many tasks on MTurk require workers to have approval ratings of over 95 percent, so even one rejection can seriously affect a worker’s ability to earn money by limiting their access to future tasks.
Like all people doing ghost work, Joan must weather fluctuating income streams. Requesters can bring booming business one day and disappear the next. Not long after she’d signed up with MTurk, Joan got a string of decent-paying tasks posted by Taste of the World, a pseudonym widely rumored among workers to belong to the popular travel site Trip Advisor. Taste of the World posted hundreds of thousands of tasks on MTurk, jobs like removing duplicate hotel listings, validating website links, writing descriptions of top travel destinations, creating city-specific lists of the best places to eat, and cleaning up typos. The average Taste of the World task could net an experienced worker the equivalent of $10 an hour, and working for the requester had other perks, too. “The work was available just about every day . . . and it was posted hours at a time,” says Joan, meaning she didn’t have to jump on it before it disappeared. She could step away from her computer to make dinner and, when she returned, the Taste of the World tasks were still available, because the sheer volume of work was so large. But just as abruptly as the jobs arrived, they dried up. Joan told us that less than a year into using MTurk, Taste of the World posted to MTurk Forum that “we have enough people.” Joan flatly added, “And that was the end.”
When it comes to paying for ghost work, each platform operates a bit differently. Amazon, in some ways, operates like both an ATM and a company store. New workers on MTurk must wait out the initial ten-day holding period before they can claim any money made doing tasks. After successfully submitting ten days of requester-approved work on the platform, U.S. workers have a choice: they can receive the full value of their earnings in the form of an Amazon.com gift card or they can transfer their paycheck into an Amazon Pay account. From an Amazon Pay account, workers can then transfer their earnings to a personal bank account, but they have to pay a transfer fee to Amazon for the privilege. International workers, with the exception of citizens of India, can only convert their earnings into an Amazon.com gift card.
Ghost Work: How To Stop Silicon Valley from Building a New Global Underclass by Mary L. Gray and Sidharth Suri
Indian citizens are the only international workers who can also earn cash for their MTurk ghost work. There’s no reason for this other than the fact that Amazon’s multinational holdings allow it to operate and transfer money between its U.S. and India office locations. India’s MTurk workers could opt to fill up an Amazon gift card, though the company doesn’t reliably deliver to many of India’s sprawling, informal neighborhoods. If a worker in India wants to move their money to a personal bank account, they must first hand over their birth date and a scanned copy of their permanent account number (PAN) card, the equivalent of a U.S. social security number. It takes Amazon a week or more to verify PAN card information. Once that is done, workers in India have one more bar to clear: they must send their bank account information to Amazon for verification. Once it’s verified, Amazon can, for an additional fee, cut paper checks or offer direct deposits to India’s MTurk workers.
For her part, Joan didn’t plan to turn MTurk into a full-time job. It just happened. Now she has settled into the life of an independent worker, and her long-term goal is to create financial stability by knitting together several sources of income. This was a common theme among workers we met. And, indeed, 75 percent of workers on MTurk report having at least one other source of income. In Joan’s case, in addition to her work on MTurk, she spins her own wool and sells knitted crafts at a local market. She’s also ramping up her technical writing skills, with the goal of having a more competitive freelancing profile on the macro-task site Upwork. And she thinks about getting a part-time telecommuting job, like online customer service work, but she hasn’t figured out how to do that while taking care of her mom. And, like 75 percent of MTurk workers, Joan does ghost work on other platforms, including Microsoft’s UHRS, even though she notes that her primary source of income, for more than a year, was MTurk.
Excerpted from GHOST WORK: How to Stop Silicon Valley from Building a New Global Underclass by Mary L. Gary and Siddharth Suri. Copyright © 2019 by Mary L. Gary and Siddharth Suri. Published and reprinted by permission of Houghton Mifflin Harcourt. All rights reserved.