Our understanding of work and workers is significantly enriched by immersive accounts of particular occupations and the people in them. Books like Studs Terkel’s Working, Barbara Ehrenreich’s Nickel and Dimed, and John Bowe’s Nobodies offer powerful narratives of day-to-day hopes, struggles, and indignities of workers in particular industries and milieu. The rhythms of the gig economy are unfamiliar to those in more traditional workplaces, and we are fortunate to have insightful new perspectives on these jobs: Alex Rosenblat on ride-sharing drivers; Karen Levy on truck drivers with electronic logging devices; and Casey Newton on social media content moderators. Add to this list perhaps the most mysterious, hidden form of new labor in our wired economy: the piecemeal “microwork” that facilitates online algorithmic processing.
In Ghost Work, anthropologist Mary Gray and computational social scientist Siddharth Suri—both researchers at Microsoft—have accomplished a deep dive into the world of these online crowdworkers. Although Gray and Suri at times include all sorts of platform workers within their definition, the true heart of the term “ghost work” applies to unseen AI support staffers who provide vital components of human judgment within an overall computational algorithm. As Ghost Work makes clear, key leaps in artificial intelligence capability have been possible only with an army of facilitators who make decisions such as what a camelback couch looks like, whether a face matches an ID picture, or how a slang term is used. Unlike the popular conception of indomitable machines churning through data unaided, most machine learning systems still incorporate significant human decision-making for the “last mile” of AI functionality. These are the people who make those decisions.
Gray and Suri masterfully present these workers, their jobs, the reasons they do them, and their workplace struggles. Drawing from a five-year study with over 200 interviews and tens of thousands of survey responses, the book illuminates the lives of these workers who huddle in front of their screens fielding on-demand requests. The work is performed through application programming interfaces (APIs) that present an onslaught of opportunities, each paying small fees for a completed task. The research team focused on workers at four different machine-learning platforms: Amazon’s Mechanical Turk, Microsoft’s internal Universal Human Relevance System, sales facilitator LeadGenius, and translation site Amara. We meet workers on these platforms, living in the United States and India, who participate with varying degrees of commitment and success in these platforms’ daily churn.
According to Gray and Suri’s research, these on-demand platform workers do not follow a uniform pattern of engagement. Rather, they break down into three categories: new and experimental workers, who are just trying the site and may never return; consistent workers, who regularly participate on the platform for at least some period of time; and the high-performers who work full-time or more and are responsible for most of the work that actually gets done. (P. 103.) And through individualized vignettes, we learn about these workers’ approach to “ghost” work: Zaffar, a young Muslim man in Hyderabad who can deftly ferret out promising opportunities for LeadGenius; Kumuda, a Hindu mother who has become the highest earner in her Indian town through Mechanical Turk; and Joan, who judges whether pictures match their textual description while caring for her ailing mother in Houston.
Ghost Work highlights the advantages of these jobs: flexibility to work at home and on an irregular schedule; the ability to experiment with different types of tasks and skills; and the hidden nature of one’s sex, race, nationality, or disability, so as to avoid discrimination. At the same time, however, the authors demonstrate the precarity of this work, and the potential for the infliction of “algorithmic cruelty.” Ghost workers spend much of their time searching for tasks and vetting the providers without compensation; they train themselves and manage their reputations while hoping not to violate the providers’ unwritten norms and rules; and they may be fired or fail to receive payment through automated decisions that cannot be appealed. Paired with the dehumanization of these isolated and fragmented tasks, the proliferation of ghost work ultimately seems overwhelming and demoralizing.
In the book’s conclusion, Gray and Suri offer a series of reforms designed to render ghost work sustainable. These include: facilitating collaboration and communication among platform workers; allowing workers to take their reputation and experience data to other platforms and positions; providing for a “good work code” across platforms and users; and installing a safety net either through publicly-provided benefits or a universal basic income. Two proposed fixes are of special interest to labor and employment law academics: changing the definition of employment, and using unions or quasi-unions to match workers with jobs and to resolve grievances, especially regarding pay. The authors are vague about the new employment classification, other than specifying it should not focus on full-time work. This superficiality is frustrating. And the book’s history of contingent work has some puzzling assertions: for example, it implies that NLRB v. Hearst Publications and the Taft-Hartley Act related to the definition of employment under the Fair Labor Standards Act. (Pp. 49-50.) This muddle is a shame; the FLSA’s use of “suffer or permit to work” would have provided a nice starting point for a more robust discussion of the legal definition(s) of employment.
Despite Ghost Work’s in-depth descriptions of the travails of this contingent and neglected workforce, Gray and Suri leave room for hope as well. They do not think that AI will engulf human labor; instead, people will always be necessary to produce appropriate and meaningful results. And even within their atomized employment, ghost workers have still found the opportunity to talk with each other about their jobs, collaborate on tasks, and help new workers find their way. The problems of ghost work are neither intractable nor inevitable. By educating us on this largely hidden labor pool, Gray and Suri have opened up a conversation about how best to structure these jobs, and how to promote human flourishing for everyone on these platforms.