Chances are the last time you ordered something, chatted with customer service, or looked something up online, you didn’t give much thought to who was making sure the information you were looking at was accurate. Welcome to ghost work, the hidden human forces working to ensure that what you get is what you think it is.
Ghost work differs from the typical jobs associated with the gig economy, like Uber and Lyft drivers or Airbnb hosts, in which you do interact with someone. Ghost work refers to the teams worldwide who are captioning photos, flagging and removing inappropriate content, or even writing, designing or coding a project to move it along.
Authors Mary L. Gray and Siddharth Suri are both senior researchers at Microsoft Research, where they came together to write Ghost Work: How to Stop Silicon Valley from Building a New Global Underclass. In the book, they tackle explaining this new and ever-changing workforce, one that will continue to displace the traditional 9-to-5 world with on-demand tasks.
As more jobs and people shift to ghost work, it raises certain dilemmas: How do workers find quality jobs? How will companies find quality workers? How will we move toward this new work ecosystem, one that overwhelmingly takes advantage of its workers, underpaying them and shifting the burden of many parts of employment, such as providing their own hardware, software, training, and more, onto the worker without any established rights to protect them?
I’ll be moderating a discussion with Gray and Suri at Town Hall in Seattle on Wednesday evening. In advance of the event, we spoke about the ins and outs of this new gig economy — tasks being cranked out behind a computer, 24/7, around the world — and a near-future that affects us all.
How did you come to the decision to tackle the issue of “ghost work?”
Siddharth Suri: Back in 2008, while at Yahoo! Research, I wanted to conduct behavioral experiments to understand how the structure of the network people are part of influences their ability to cooperate. At the time, these kinds of experiments were typically done with undergraduate students sitting in a university computer lab. The problem was that at Yahoo! we didn’t have any undergraduates. So I, along with my colleagues, figured out how to use crowdsourcing sites, like Amazon Mechanical Turk to conduct experiments with human subjects. MTurk provided an always available crowd of workers that we could pay to participate in our experiments.
We started off by doing experiments similar to those had previously been done in the lab setting. But quickly thereafter we pushed this new methodology to see what we could study that wasn’t previously possible. We did experiments that took place in a more realistic setting than a computer lab with a more diverse population. We also figured out how to do experiments with hundreds of workers working simultaneously, and we did experiments that took place every day for a month.
Around this time the psychology community heard about this work and moved many of their experiments from the classroom to an online setting. Today it’s almost taken for granted that a researcher conducted, at least part of, their experiments on MTurk. But whenever I would give a presentation someone in the audience would always ask, “Who are the workers?” I would give statistics about the demographics, but I didn’t really know beyond that. So in 2012, shortly after I started at Microsoft Research NYC, Mary [L. Gray] approached me and asked if I’d like to collaborate on doing an ethnography on the workers. I thought it was very cool that an anthropologist and a computer scientist were interested in the same thing. So I said yes, and the project was born.
The downsides of ghost work: In the introduction, you write that “Joan, with years of practice, now knows how to piece together an average 10-hour day that will bring in roughly $40 worth of such tasks.” The online marketplace has actually driven down what professionals can charge for work. Do you see any way of improving the pay scale for workers online?
SS: At first glance these markets might seem very efficient. There is a list of tasks that a worker can search, choose and do. There is a large pool of workers for requesters to choose from. So it would seem there would be very little friction in matching a worker with a task. But when you look closer you see a lot of friction.
The search functionality on these platforms generally doesn’t work very well so workers have to collaborate with one another off the platform to find good work. Workers don’t get paid for this overhead they incur, and it’s time they could be spending working. Many platforms only allow the requesters to see a worker’s reputation but not vice versa. Finally, it is very common on these platforms that only a few requesters that put up the overwhelming majority of the tasks. So the deeper you look you see that the requesters have the majority of the power in these markets. Workers have little choice but to accept whatever wage rate these requesters offer, as there is often no way for workers to bargain. Economists call this monopsony.
So to improve the pay scale for workers we need to decrease these frictions and balance the power in the market. We would need all platforms to let both sides of the market, workers and requesters, see the reputation of the other party. We would need platforms to seriously invest in their search functionality. We would need more requesters putting up more tasks and more types of tasks. Finally, for those platforms who don’t allow workers to bargain for wages, that would help balance the power as well.
Were you surprised by the other downsides that come with ghost work, like hyper-vigilance and a lack of direction on projects?
SS: I was surprised by how much of the transaction costs got pushed onto the workers. The specific transaction cost I was most surprised by was hyper-vigilance. When you first look at these platforms it appears there are many requesters posting many jobs, so it looks like there is a lot of choice for the workers. In reality, only a few requesters post the majority of the work, and the best, most lucrative work gets snapped up quickly so workers have to constantly be on the lookout. Requesters also reported seeing the consequences of hyper-vigilance. They said whenever they posted a task there would be a flood of workers trying to do it which made it hard for the requesters to choose the most qualified worker. So both sides of the market were losing due to this inefficiency.
This finding showed a nice interplay between anthropology and computer science. Mary’s interviews found out that workers feel the need to constantly be on call for good work. Then we conducted behavioral experiments to measure how much do workers value, in terms of dollars, some flexibility in when they do a task.
The upsides of ghost work: There were a lot of pleasant surprises here, like camaraderie and creativity. What surprised you when you started asking people what they liked about the work?
SS: I had been doing work in crowdsourcing for over five years, and I had no idea that workers collaborate and communicate. It sounds obvious in hindsight, but it was definitely not in foresight. When I think of the term “crowd” I think of independent individuals. I would bet that most computer scientists thought that as well.
But it turns out the crowd is actually a network. It’s just that the API requesters use to hire workers hides the connections between them. The API lets a requester hire one worker, or a collection of workers, but it doesn’t show you who they might be connected to. It turns out that workers use these connections to re-create the office watercooler. They congregate in online forums to offer each other social support, blow off steam, and share best practices.
The finding that workers form their own social networks and how we mapped the network is one of my favorite findings in the book.
As more work transitions online, what are some full-time fields you see transitioning to ghost work, or more “macro-tasks,” in the near future?
Mary L. Gray: If history is any indicator, most knowledge work and information service jobs that can be, at least in part, sourced, scheduled, routed, managed, shipped and billed through a mix of Application Programming Interfaces (APIs), the internet, and a sprinkle of AI are up for grabs. Ghost work is not a niche job so much as a mechanism for dismantling of full-time jobs and transforming them into task-driven work done on contract. If trends continue at the current rate, economists estimate that by the early 2030s, tech innovation could dismantle and semi-automate roughly 38 percent of jobs in the U.S. alone (John Hawksworth et al., UK Economic Outlook: Prospects for the Housing Market and the Impact of AI on Jobs. London: PricewaterhouseCoopers, 2017).
The most obvious fields in the thick of transition are healthcare and retail. For example, it’s now possible to route all the requests for following up with a healthcare professional and turning the callbacks, patient guidance, appointment scheduling, and other bits of office work into tasks that can be distributed to a host of people able to answer a quick customer query, compare calendars, and decipher clinician’s written notes so that they stay attached to a patient record. None of those tasks need to happen in front of a patient or in the medical office.
Or, in the case of retail, U-Haul now allows customer service texts to be routed to on-demand workers who can pick up queries about road closures or store hours and reply on-the-spot, no matter where the customer generating the question might be. Law and financial services are also easy targets because so much of the filing rules and legal guidelines are spelled out. Any work that has some rote directions to it can be structured in a way that a person can be brought in to double-check how the software took the available information, like a tax law, and applied it to something a customer added, like their annual income. In essence, a person is threaded into the loop to check the math before the filing heads to the IRS. That’s exactly the type of service that Intuit and LegalZoom offer now.
Can you talk more about how the Microsoft permatemp case was a miss for establishing worker rights in this arena?
MLG: In the late 1980s, Microsoft was thrust into the spotlight not so much for its status in the growing tech industry—it was already a big name in software—as for its approach to meeting the intense staffing needed to constantly test out new products and prototypes. Microsoft was assigning temporary workers tasks that were virtually identical to what their permanent staff did. These “permatemps” spent years on software projects with the same responsibilities, reporting to the same management, and on full-time hours as their full-time counterparts. In many cases, everyone was making the same amount of money and had the same benefits, except for stock options extended to full-time employees only.
By 1989, the IRS had grown wary of this arrangement and audited Microsoft’s staffing procedures. The IRS ended up deciding that only about 600 of Microsoft’s independent contractors should be reclassified as permanent employees, because their work was entirely under Microsoft’s control.
In 1992, a group of temporary workers filed a class-action suit (Vizcaino v. Microsoft) against Microsoft claiming that they were common-law employees and should receive the same benefits as permanent staff. In 2000, after nearly eight years of litigation, roughly 8,000 Microsoft permatemps received a settlement of $97 million.
Without a court ruling, the question of what kinds of workers these permatemps were and what kinds of benefits and protections they deserved was never resolved. Keep in mind the timing. The case settled just as the dot.com bubble popped. Post-permatemp, Big Tech could lean heavily on a contingent workforce to rebuild. The permatemp settlement left new start-ups and tech companies that had weathered the economic downturn of the early ’00s in charge of interpreting 1940s-era labor laws written for assembly-line workers to set the terms for hiring coders, designers and a host of other newly unemployed creatives on contract.
Small companies rising from the ashes of the crash started everyone on-contract until they proved too valuable to lose once their contract ran out. This became the default mode of hiring in Silicon Valley. The cadre of mostly white, middle-class, college-educated 20 and 30-somethings hired by tech companies to work on projects neither shared a professional identity nor a particular interest in organizing for their rights.
Sociologist Gina Neff writes about this moment beautifully in her book Venture Labor. These workers were happy to have decent paychecks and some semblance of upward mobility (or least some relief from feeling knocked off that trajectory). Competition for some workers, particularly freelance software developers, heated up. Some individuals moved around to get better compensation, but the churn made it even harder to organize workers’ interests.
The settlement is the great irony of the tech industry: the tech industry could’ve never cycled through so many failed projects—moved so fast, broke so many things—if it had not been able to rely on a contingent workforce, hired to tackle 18-month build cycles and released when the software update shipped. Add to the mix the instability introduced by the start-up acquisitions and company mergers that meant a quarterly reinvention of what jobs were “mission critical” and you have the perfect storm for undoing worker solidarity and shared vision for a long-term future that had been the main organizing tactics for most of the history of labor in the United States.
Tech companies didn’t know it at the time, but they would become increasingly dependent on contract staffing for almost every business operation. Now they, like the contract workers, are stuck with an odd set of rules about how long someone can contract with a single company and stay in a specific role there; the limits on their access to workplace amenities, like gyms, commuter buses, or after-work pizza parties; and whether they have a right to organize or collectively bargain at their workplace.
The opportunity lost was two-fold: Had the case gone to court, we would have had an answer to what are the corporate responsibilities to workers who are invaluable for a specifically defined project that may last several months or, if things go well, a couple of years? Are these workers any less valuable to a company’s “return on investment” in some measurable way and, if not, shouldn’t they get the same benefits as someone recruited for a less-defined project or term of service?
And, bigger than that, had the case continued, we could have had a robust, public debate about the rights of every worker, no matter where they work or their hours on the job to basics, like healthcare, retirement, paid leave, continuing education, and the other protections that, today, are reserved for those who land full-time employment.
Your suggestions for a brighter ghost work future: better healthcare options, hours, projects, teamwork, etc. How likely is that more companies will adopt the creativity/collaboration mindset like LeadGenius or Amara? Or will the trend be to continue to push the burdens of the work life (software, admin costs, benefits, learning curves, etc.) all on the workers?
MLG: The answer to this question really depends on consumers and citizens. If history tells us anything it’s that the collective fate of workers is a moral and political question that should not be determined by market forces, even if they come from the good intentions of fair-minded businesses like LeadGenius and Amara…
Our global economy is driven by information and service work. Businesses make their money on finding new consumer markets dependent on the exchange of an ever-evolving set of service experiences and information. Yet, our labor laws still operate as though most of us will learn one set of skills, enter the workforce, and hold a 9-5 job with the same employer for most of our work lives. This is an outdated snapshot of how our global economy works.
Our research offers solid, empirical evidence that when platform companies recognize and address the needs of workers, they get better information services. Does this increase operational costs? Yes. In the short-term. But should businesses pay more rather than be allowed to shift business costs to independent workers? If you were an independent worker, what would you answer? Right now, consumers are getting inexpensive platform services at the expense of workers. Are we willing to pay more to see workers’ conditions improve? And are we, as citizens, motivated to lobby for changes to existing labor laws so that every worker, no matter their employment status, has access to the basics to make work life sustainable? Shining a spotlight on ghost work conditions—a labor market defined by projects done on-contract where there is no promise of steady employment or career advancement—is an opportunity to debate what we want the future of work to look like.
Ghost workers have pulled together to stand for their rights, like the letter-writing campaign to Jeff Bezos you cited in the book. What are the likelihoods that more workers will mobilize to fight for better rights in the online workplace?
MLG: Lyft and Uber drivers recently held a strike in solidarity to demonstrate their frustration with both companies’ astronomical IPO valuation. These drivers are the visible tip of an iceberg of work below the surface of APIs. These drivers took aim at the larger industry of ride-sharing services rather than a specific platform.
As more and more knowledge workers come to depend on platform services to pick up projects distributed as task-based work for hire, there will be more chances to organize workers to mobilize for their rights to improve work conditions, no matter what platform they use to find work. Because any one online workplace is dependent on the larger ecosystem of platform companies, it will, likely take fighting to redefine what it means to be gainfully employed as an independent worker to meaningfully convert ghost work conditions to a sustainable version of on-demand employment.
We are at the very beginning of a labor movement among independent workers finding their common cause in calling on a right to fair treatment and control over their lives. It’s likely they’ll succeed if we collectively see how this is about fighting for all our employment rights, not just the rights of a poor few, doing a niche job.
Do you eventually see large established platforms, like Microsoft and Amazon, implementing any of the suggested changes to improve their services for both requesters and workers?
MLG: Large tech companies, notably Microsoft and Google, have already put in place requirements that their labor contracting services provide benefits, like paid leave and healthcare. This is a step in the right direction. And, arguably, the biggest brand names in tech have the most to gain by making the AI labor supply chain fueling tech advancement as transparent and equitable as possible.
Large companies, like Nike, learned this as they came to grips with the horrific labor conditions that went into producing their products. Consumers, particularly younger ones, are statistically likely to care more about where what they buy or eat comes from and how it was produced. There’s no reason to believe that tomorrow’s consumers will be any less scrutinizing about the sourcing and labor conditions that went into their favorite information services.
Odds are good that the future of work will be augmented through technologies rather than automated outright. Companies that equip both consumers and workers for this future have the most to gain from leading the industry in setting the bar for creating dignity and sustainability as they build technologies for future workers.
You have strong conclusion that the future of work will include an ever-changing relationship between AI and people, dispelling the popular notion that AI will completely displace people in the workplace. Did you know this going into the project, or did the research help you come to this conclusion?
SS: The research most definitely helped us come to this conclusion. As an anthropologist, Mary believes that computers will never be able to emulate human intellect. As a computer scientist, I believe machines are currently very far from human intelligence, but I don’t know if we’ll ever get there or not. I don’t think anyone knows, and I’m not even sure anyone can know right now given our current understanding of cognition.
But I do believe that we’re many decades from knowing the answer to this question and I respect the creativity and ingenuity of scientists and engineers. In discussing our different viewpoints we realized that whatever we believe about the end point ignores how we get there, and that’s a huge part of what this book is about. For the foreseeable future, AI systems will require human labor, and our goal was to shine a light on those humans.
Ghost Work: How to Stop Silicon Valley from Building a New Global Underclass by Mary L. Gray and Siddharth Suri is out today. The authors speak with GeekWire book reviewer Molly Brown at Town Hall on Wednesday, May 15, in Seattle.