Originally posted on jacobinmag.
Companies devalue them, and consumers rarely know they exist. But the apps and companies that millions of us depend on, like Uber and Amazon, couldn’t function without the invisible, low-wage labor of “ghost workers.”
At random intervals, drivers are required to take a picture of themselves in the app so a facial recognition software can determine whether the image matches the one associated with their account. About once in every 100 pickups, a freshly shaved beard or new pair of sunglasses confounds the software. These pairs of selfies get routed to human workers, who have to determine that the two pictures are of the same person.
In a matter of seconds — no longer than it takes for a rider to find their driver and get in the car — the images are sent to the Indian tech hub of Bangalore, where a worker sitting at her kitchen table quickly evaluates them for a few cents in wages. When driver and rider meet, they have no idea that the ride was only allowed to go forward thanks to the real-time work of a third person across the world.
Tech giants like Amazon, Google, Microsoft, and Uber can only function smoothly thanks to the efforts of these workers, who labor for exceptionally low wages and little security. These workers are the subject of Ghost Work: How to Stop Silicon Valley from Building a New Global Underclass. Authors Mary L. Gray and Siddharth Suri, respectively an anthropologist and a computer scientist, paint a vivid and often devastating picture of the vast global workforce propping up the tech industry behind the scenes. They are called ghost workers because companies devalue and fail to acknowledge their role, while consumers rarely know they exist at all.
The ghost workers whom Gray and Suri profile live in India and the United States, the countries with the biggest on-demand labor pools. They have various (and often quite high) levels of education and seek out on-demand work for different reasons. Some are full-time caretakers who need income on a schedule that will accommodate their obligations at home. Some are women who are expected not to seek work outside their homes. Some are people with disabilities who do not expect to find accommodation in conventional office work. More than anything, however, participation in on-demand labor says less about its supposed benefits than the unavailability of decent, high-paying alternatives.
A Microtask Army
Because online ghost work is global, decentralized, and relatively new, it is not clear exactly how many people do it. According to a 2016 Pew Research Center report, 20 million US adults had earned some money in ghost work in the previous year. Ghost workers are behind every big name in tech and many small ones, but the platforms where all this work takes place keep a lower profile. Every major tech company, from Google to Facebook to Amazon, maintains its own internal platform to fulfill the need for “human computation.”
Microsoft’s internal Universal Human Relevance System, for instance, features prominently in the book’s account. Many companies, big and small, source this work through third party platforms, such as CrowdFlower (now Figure Eight). Platforms like Amazon’s Mechanical Turk let anyone, including individual people, post tasks for workers to complete.
The authors draw intimate portraits of workers on living room couches and bedroom workstations, setups that feel more familiar to many now than when the book was published in 2019. Ghost workers form a standing army of people paid to do “microtasks,” each usually lasting just a few seconds and netting just a few cents, like captioning videos, responding to customer requests, spell-checking, and content moderation — “dollars for dick pics,” as Joan, one worker profiled in the book, described her work filtering content marked as “offensive” on Twitter and Match.com. Joan pieces together an average of ten hours of work per day, during which she earns about $40. Workers like her are the reason that Google searches turn up relevant results and social media feeds are not flooded with profanity.
Machine-learning models are trained by gig workers. Workers tag and sort words and images to create training data to build and improve on artificial intelligence. Other times, as in the case of a particularly challenging Uber driver verification photo, humans jump into an otherwise automated process to complete jobs that are beyond the current capacity of automation.
This second scenario represents what Gray and Suri call the “paradox of automation’s last mile,” a major theme throughout the book. The paradox holds that at the most advanced stage of automation — whatever it happens to be at the time — technology will fall slightly short of its goal and require humans to complete the last mile of the journey. The work of final image taggers, spellcheckers, and moderators tries to improve the technology to the point where such work is no longer needed. But as automation continues to improve, its goals expand. New goals push the last mile of the journey a little further down the road, creating a need for human decision making at the edge.
Microtasks come to workers through application program interfaces (APIs), which allow tech developers to communicate with these workers much in the way they would with any other piece of software. The design is purposefully obfuscating and dehumanizing. To the people and companies requesting microtasks, the workers completing them appear only as a string of numbers and letters. Such abstraction exacerbates the exploitation of ghost workers. Ghost work platforms deliver services to developers and consumers under the pretense of one of big tech’s most pernicious lies: that technology can replace workers entirely.
In this context, workers become numbers, and companies or people that hire them become requesters. But major platform operators remain the boss, clearly aligned with requesters over workers — not a neutral party. As bosses hide, ghost workers likewise do not get the distinction of being employees. Because they are not considered employees, workers are not owed any minimum wage. They don’t get any employment benefits and don’t get compensated for time spent learning how to navigate platforms or searching for tasks.
Ghosts of Ghost Workers Past
Ghost work platforms as they exist today began with Amazon in the early 2000s. In 2005, it launched Amazon Mechanical Turk, called MTurk by users, which would allow anyone to make an account and spend some time cleaning up its listings for a small amount of pay. Other employers, now called “requesters,” could also post tasks on the platform and use it to make payments. Thus, Amazon turned the massive contingent labor force it had created into its own product to sell. Gray and Suri report a rumor that MTurk was a personal project of Jeff Bezos himself, who wanted to make labor into an immediately available resource — just like all the site’s other products.
Gray and Suri situate ghost workers in historical context, connecting those workers today with piece workers making bows outside of clothing mills in nineteenth-century New England and the contingent labor force of human “computers” that made space exploration possible in the 1960s. In both cases, proximity to the most advanced form of automation at the time devalued workers, mostly white women and women of color, respectively. They were seen as primed for replacement by machines, even though the machines could not work without them for many years.
Early contingent workers — precursors to the ghost workers of today — were also left out of the early US labor movement. Gray and Suri develop an uneasy relationship with organized labor in this book, as they place some responsibility for ghost workers’ predicament at the feet of the US labor movement. Landmark labor bargains like the 1950 “Treaty of Detroit,” in which auto workers won major wage, retirement, and health care benefits, not only defanged unions by giving up the right to strike, but was also instrumental in tying all benefits to full-time employment across industries. This tempered victory took place against the backdrop of the Taft-Hartley Act, the reactionary 1947 law backed by the National Association of Manufacturers and passed over president Harry Truman’s veto, meant to curb the growing power of organized labor after World War II. Taft-Hartley, still largely in place today, subjected workers to a strict litmus test of full-time employee status in order to fall within the protection of the Fair Labor Standards Act.
Well before the 1950s, organized labor lost an opportunity by neglecting workers who could or would not leave home-based piecework, perhaps understandably seeing off-site work as a threat to the bargaining unit, and by focusing efforts on men in traditional heavy industry. In spite of this history and because of it, the path to a more secure future for ghost workers, whom Gray and Suri cast as pieceworkers of today, lies with the labor movement.
The legal distinction between employee and contractor still harms workers today, including ghost workers. Gray and Suri point out that a select set of “always-on” dedicated workers perform 80 percent of the available tasks, managing to cobble together something suspiciously close to full-time employment. Because they are not employees of on-demand labor platforms nor of job requesters, none of these workers get any protection.
Gray and Suri’s research shows that a full 30 percent of people doing ghost work report not getting paid in full for their work. This is no surprise given that many platforms, including Amazon’s MTurk, allow requesters to decline to pay workers if they are dissatisfied with the job in any way. Workers have no way to challenge this decision. The workers that Gray and Suri visit in their research all report having their accounts shut down or temporarily suspended for reasons that they cannot discern, much less contest — cutting them off from their income with no recourse. Gray and Suri refer to this unthinking, unfeeling processing of human labor as “algorithmic cruelty.”
Ghost work platforms claim to eliminate the costs and risks of hiring full-time employees for companies. This is largely true. They do so by shifting those costs and risks onto the workers themselves. Anyone who manages to earn appreciable income through this kind of work achieves it with the help of active forums where workers build relationships and show each other the ropes of each platform, guiding one another toward higher-paying tasks and away from scams. APIs do not include any built-in way for workers to communicate with each other, underscoring the way that ghost work tries to strip work down to the bare elements of assignment and pay. But that couldn’t be further from reality. Instead, workers on these forums fill the training and collaboration role, at the cost of their own time and effort, that would have been provided to them in a traditional job.
When workers collaborate and build friendships, they recreate the social side of work that is sorely missing from their atomized online work environment. It also creates the potential for organizing collective action. Sadly, Gray and Suri ultimately land on a tepid, if somewhat hopeful, note about ghost workers’ organizing potential. After all, they don’t share a worksite, common hours, or a professional identity — three typically essential ingredients for successful collective action. Nonetheless, ghost workers have tried to organize before.
In 2014, workers rallied around a letter to Jeff Bezos calling on him to stop selling them as cheap labor and provide them with tools to represent themselves. Though widely publicized at the time, the campaign has since fizzled. Written in 2019, this book could not account for the ways in which COVID-19 would amplify the urgency of its message. The implications of remote work for organizing is one of them.
Ghost Work ends with prescriptions for how to better serve ghost workers, and all workers, as the nature of employment changes. The authors clearly understand the barriers to collective action that ghost workers face; perhaps because of this, their proposals fall back on somewhat technical policy solutions. Gray and Suri’s suggestions range from legal grievance procedures to demanding accountability from platforms for worker rights abuses to a basic income for all working-age adults.
The authors’ core premise that we must separate the basic security of health care, paid leave, and education from the status of full-time employment rings urgently true. The authors seem reluctant to call for something so radical or expansive as policies explicitly designed to empower workers at the expense of their bosses; they couch their recommendations as mutually beneficial for employers and workers. After all, they argue, employers will ultimately be better off with a workforce that is less stressed, more available, better educated, and healthier. They look to the success of more positive models of on-demand work platforms, such as startup LeadGenius and the mission-driven captioning and translation nonprofit Amara.org, to underscore this point. LeadGenius guarantees workers a minimum wage, and both companies acknowledge and actively facilitate collaboration among workers.
But any optimism that the likes of Amazon would find it in their own interest to support broad policy change in support of workers is misguided, creating a strange dissonance between the authors’ optimistic policy suggestions and the realities of work and bosses’ power under capitalism that their ethnographic research lays bare. As they write, workers’ willingness to weather the low pay and algorithmic cruelty of ghost work says less about how good that work is than how awful every other kind of work is.
Gray and Suri emphasize that ghost workers don’t work for money alone — some derive a sense of purpose from seeing their time as productive, while others see the work as a way to build skills, and many value the relationships they form while navigating the world of ghost work. Even so, earning an income remains the primary drive for most. Without the constant insecurity created by capitalism, on-demand work platforms might have no workers at all.
Gray and Suri are clear about one thing: Ghost work is coming for more of us. COVID-19 has accelerated this process. Many white-collar workers prefer working from home for many of the same reasons that some ghost workers do — they can avoid wasting hours of their day commuting and exert a bit more control over the time they do spend working. But bosses are already trying to claw back some of those benefits in the form of reduced pay or heightened demands. Workers newly atomized in their homes are at major risk for this kind of creep — more duties for the same pay, less support and accountability, more surveillance, and no coworkers to compare notes and build solidarity with.
Still, Ghost Work brings to light the experience of workers who need recognition and on-the-job protections. Ghost workers’ experience is a warning, to everyone who still enjoys some legal protection in their jobs, of how corporations will treat workers if they can get away with it. Millions of workers are toiling in their homes, purposely hidden from view, keeping the machines up and running.
Ghost work falls outside of the formal definitions of labor law, and that has real consequences for workers. But it falls squarely within the basic dynamics of work under capitalism. Their insecurity is a reflection of the kind of insecurity all workers face to varying degrees. Gray and Suri propose a cultural shift toward recognizing ghost workers and providing them more autonomy and a safety net. As with any other workers under capitalism, those changes must start with organizing.