Book Review: Ghost Work

I have planned to read Ghost Work by Mary Gray and Siddharth Suri for a long time. I bought the book when I visited a talk by Mary Gray at Data and Society about a year ago. Since then the book still sits on my bookshelf. The general theme of the book is about the necessary unseen human labor behind the seamless automated systems. The core concept of the book is the idea “paradox of automation’s last mile.” It refers to the phenomenon whereby when Artificial Intelligence becomes more advanced, it would create “temporary labor markets” to solve “unforeseen and unpredictable tasks.” Furthermore, the grater paradox of automation is that “the desire to eliminate human labor always generate new tasks for humans.” In other words, automation cannot and would not in a foreseeable future, rather automation reconfigures work, and reorganizes human input in the production process.

In the authors’ own words, on-demand service work is necessary because we do not know when machines need human input:

“As machines get more powerful and algorithms take over more and more problems, we know from past advances in natural language processing and image recognition that industries will continue to identify new problems to tackle. Thus, there is an ever moving frontier between what machines can and cannot solve. We call this the paradox of automation’s last mile: as machine progress, the opportunity to automate something else appears on the horizon. This process constantly repeats, resulting in the expansion of automation through the perpetual creation and destruction of labor markets for new types of human labor. In other words, as machines solve more and more problems, we continue to identify needs for augmenting rather than replacing human effort. This paradox explains why on-demand services – commercial ventures that combine humans and APIs to source, schedule, and deliver access to goods and services – are more likely to dominate the future of work than AI and bots alone (p.176).

The authors interview workers in the United States and India, and bring them to us. They go behind the API curtain, and reveal to us who the workers are, what are their aspirations, and why they work for on demand platforms. This book is comparative on many levels. First, it compares workers in different countries: The US. and India. Second it compares experiences of workers who work for multiple platforms: Amazon Mechanical Turk, Microsoft Internal On-Demand platform, Amara translation service, and LeadGenius. The diversity of platforms, and countries allow us to see a picture of an emerging global on-demand labor market, which performs million of tasks, which vary in complexity. This book therefore enables us to see what is often hidden and abstract.

The authors then provide institutional background on why the on-demand labor market has become necessary in the digital revolution. They also show us how on-demand labor market is not a new system. Before the industrial revolution, many women and households took on the job of sewing the last buttons to clothes before the textile industry figured out how to automate the process, and confined the work within the walls of garment companies. This inclusion of non-institutional labor is important in the process of automation.

As they walk us through lived experiences of their interview subjects, the reader recognizes heterogeneity of human labors in this on-demand market. What becomes apparent is that since the market does not have a clear requirement of educational background, training level, etc, it allows for a diverse labor force with different levels of training, different levels of education, and coming from diverse racial, ethic, and religious backgrounds. However, this heterogeneity of labor supply, and diverse forms of tasks also create inefficiency, and friction because workers have different levels of skills, and that requesters/ employers have to define the tasks themselves. The authors show that this system involves a lot of transaction costs for both workers and employers. The two sides have to put in the time to find the right match, and to explain to each side how to do the task as intended. Workers avoid the problem of looking for tasks by creating social networks outside of the platform. They rely on social media, online forums to find the right tasks. As the author outlines this problem of transaction costs, I wonder whether building a well thought-out communication platforms for on-demand workers and employers would be a potential solution for the various transaction cost problem in this market. This is a technical solution for the current inefficiencies in this market.

One theme that I observe in this book with other Gig economy books that I have read in the past year is that workers in this economy are subject to algorithmic arbitrariness. Workers are suspended, and kicked out of the platforms sometimes randomly, and sometimes according to rules that are not taking their real life situations into account, while workers have no recourse, no where to complain. This shows the power of platforms over workers, and that workers though important to platforms’ profitability are not treated as assets but expandable number that could be eliminated at will.

Gray and Suri explain:

“The worst expression of algorithmic cruelty is disenfranchisement. Under the guise of safety, systems designers make it easy to block or remove an account in case a bad actor tries to cheat the system. This adversarial stance means that good workers are sometimes misinterpreted as shady players. Inevitably, mistakes are made. A worker changes an address, loses her internet connection, or shares an IP address with another worker. Each one of these things is potential red flag. The algorithmic system sees the flag as a possible security threat and, with no one at the helm to distinguish friend from foe, the worker is penalized. The penalty may look like being blocked or suspended, or having an account deactivated. Again, in an ecosystem in which workers are seen as interchangeable, the system automatically eliminates what it deems bad apples. The sad irony is that even the best – intentioned and most seasoned workers can get caught in the dragnet.” 86

Workers are dehumanized through the process of de-identification. Mturkers become lists of numbers. This reminds me of how Jewish prisoners given a number during the Holocaust. Giving a working human being a code to interact with is so dehumanizing for both sides: the requesters, and the Mturkers. The authors though qualify this statement by saying that in case of workers who come from discriminated classes (gender, religion, etc), not being identified by names and gender sometimes giving them advantage.

At the end, I feel that the book presents a good narrative of what is going on in the tech economy. However, as a sociologist of work, one question remains unanswered is the question of “work process” among on-demand gig workers: Why do they work so hard for very little paid, and why don’t they quit? What is the average tenure of an on-demand gig worker working for an on-demand platform? The authors point out the 80/20 Pareto rule to create a typology of three group of workers. However, I want to know among those who make on-demand work their full time career, why do they work so hard for little pay? Another question is why they not call them gig workers? What is then the difference between gig work and on-demand work? Aren’t they the same?

To answer the question what keeps they in the game, the authors provide a partial answer: many of them are in the game for the cognitive benefit of it. They learn new things, keep up their skills (most of these answers come from Indian subjects). However, my sense is that because the book is not an ethnographic research, they can never quite get at the process that workers rationalize the decision to remain in an exploitative labor scheme.

Besides, How about their American counterparts? Why are they working so hard for little pay? The answers are either implicit or not satisfactory. Implicit in the sense that they work for various reasons. One, the workers population are so heterogeneous, they should have different reasons why they work in this sector. Thus, they should also have different reasons why they stay. Is there anything about the on-demand aspect of this that keeps them stay? Is there anything about the brandname (Amazon.com, or Microsoft) that make them stay? These questions remain open.

Finally, as a methodology enthusiast, I feel the book to be not transparent in its methodology. Who were involved in the interviewing process, who was contacted, who was doing the interview, etc. These pieces of information is absent. As mentioned earlier, because the book is not explicit in whether ethnography was involved at all, readers cannot really picture the embodiment aspect of online/on-demand work.

Because I care so much about reproducibility of research, the book does not have a methodological appendix that makes me cringe. I know that it is produced for popular audience, but as a scholar, a researcher, a scientist, I want to know how many people they have interviewed, how did they interview them, how many in person, how many remote. How did they avoid positionality biases being MSR employers, privileged, and at times employers of those ghost workers.

Overall, I agree with the authors that there’s a global ghost work sector that is increasing in size because of the increase in demand for human in the loop tasks from various tech companies. They are working outside of the formal employment structure, and they are subject to the whim of the platforms, and being exploited by requesters because of the platform design. However, I think the book has not answered many questions, and one of which is methodological, and another is theoretical.

Despite many questions that I have, the book is a starting point of a long over-due conversation: who are the human workers who power machines. How can we as society protect them, and enable their creativity for our better future. The book is both practical, and hopeful that we actually will continue to need humans in the loop. The book also provides one practical solution for job training program at the city level that I really like: supporting public education, and letting residents to take college classes that they would want to take in order to benefit their work. This similar program enabled me to audit courses at Humboldt University, Free University and Goettingen University during my stay in Germany. It plugged me into the intellectual environments of those excellent public universities, and through those courses I had also made long lasting friendships. I’m all for investing in public universities and making their courses available to those who pay their taxes to support such excellent public education.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.