Adapted from the introduction and conclusion of Ghost Work: How to Stop Silicon Valley from Building a New Global Underclass by Mary L. Gray and Siddharth Suri. Copyright © 2019 by Mary L. Gray and Siddharth Suri. Reprinted by permission of Harper Business Books, an imprint of HarperCollins Publishers. All rights reserved.
The OECD Forum Network is a space for experts and thought leaders—from around the world and all parts of society— to discuss and develop solutions now and for the future. Aiming to foster the fruitful exchange of expertise and perspectives across fields to help us rise to this critical challenge, opinions expressed do not necessarily represent the views of the OECD.
The human labour powering many mobile phone apps, websites, and artificial intelligence systems can be hard to see—in fact, it's often intentionally hidden. We call this opaque world of employment ghost work. Think about the last time you searched for something on the web. Maybe you were looking for trending news topic, an update on your favorite team, or fresh celebrity gossip. Ever wonder why the images and links that the search engine returned didn't contain adult content or completely random results? After all, every business, illicit or legitimate, advertising online would love to have its site ranked higher in your web search. Or think about the last time you scrolled through your Facebook, Instagram, or Twitter feed. How do those sites enforce their no-graphic-violence and no-hate-speech policies? On the internet, anyone can say anything, and, given the chance, people certainly will. So how do we get such a sanitized view? The answer is people and software working together to deliver seemingly automated services to customers like you and me. […]
Anyone who scrutinizes the shadows of artificial intelligence (AI), as we have done, will find a new world of work in which software manages people doing jobs that computers can't do. As builders create systems to transfer tasks from humans to machines, they surface new problems to solve through automation. For example, it was only after the web became mainstream that companies like Facebook, Twitter, and Instagram faced growing demand to moderate their online content, outstripping the limited capacity of automated moderation tools. At the same time, as novel systems are brought online, they typically face unanticipated problems and fall short of their promise […]. The inevitable glitches that automated processes encounter along the way to perfection generate temporary work for people. Once they have successfully trained artificial intelligence to perform like humans, workers move on to the next tasks engineers assign them that push the boundaries of automation. Since the finish line moves as people dream of new applications for AI, we can't be sure if the "last mile" of the journey toward full automation will ever be completed. We call this the “paradox of automation's last mile”. […] On this frontier, the peaks and valleys of temporary work shift constantly, redefining relationships between humans and machines in the process.
Delivering services and jobs on demand could be an integral part of the future of work. It could also have unintended, potentially disastrous consequences if not designed and managed with care.
The rise of on-demand labour platforms signals the allure of using application programming interfaces (APIs) to organize, route, and schedule work. [This] reorientation to use contingent labor to develop new technologies fueled the recent "AI revolution”. When an Al system that powers a phone app or online service isn’t confident about what to do next for a customer, it needs human help, and it needs it fast. End users expect software running search engines and social media to respond in milliseconds. Traditional methods of hiring won't do here. So if an AI needs a human in the loop to make sense of a spike in search terms tied to, say, a sudden natural disaster, it needs to get human input immediately. The disaster will fade into history. The software will have learned what it needed from the momentary food of human input. That is exactly what an always-on labour pool, plugged into APIs, provides. Software developers can write code that automatically hires someone to solve an immediate problem, checks their work, and pays them for doing the job. Similarly, scientists and researchers using modern machine learning systems depend on training data that is clear and error-free. They need an automated method to get help generating and cleaning up that data, and they rely on many people around the world to do it. On-demand labor platforms offer today's online businesses a combination of human labor and AI, creating a massive, hidden pool of people available for ghost work. Delivering services and jobs on demand could be an integral part of the future of work. It could also have unintended, potentially disastrous consequences if not designed and managed with care and attention to how it is restructuring the experience and meaning that people attach to their daily jobs. […]
To find out more, see also our dedicated website to access a wide range of resources and reports on the impact of AI on the labour market:
[S]oftware is, by design, rigid. And its inability to make judgment calls results in three types of algorithmic cruelty. First, requesters can upload a large number of tasks at any time automatically using the API, which disappear once they are done. Collaborating to find good work can help smooth out the "bursty" nature of when work arrives on the platform, helping workers offset the costs of the instability and insecurity that often come with this line of work. Nonetheless, collaboration does not fully solve the problem, as workers often feel the need to be always on call or hyper vigilant to get good work before it's gone, which is the first form of algorithmic cruelty. Second, since the API governs the interaction between the worker and the requester, should the worker need guidance or assistance, they have no one to turn to. As a result, they often tap their work networks for help. Finally, platforms can unilaterally decide who has access and who doesn't, which implicitly means they decide who can earn and who can't. They use their own internal software tools to render a verdict on who stays or goes. Workers often have no avenue for recourse. These three forms of algorithmic cruelty are brought on by the ability to hire workers and give them access to a platform via APIs.
Workers persevere, in spite of these difficulties, because they [can] mold on-demand work to fit the constraints that they face. For example, cultural norms may dictate who can and can't work outside the home due to race, gender, religion, or disability status. Since on-demand work can be done anonymously from anywhere, including from home, workers can use it to push against this barrier. Similarly, workers may have family obligations that limit their work hours. Since on-demand work can be available at any time, it can be molded around these responsibilities as well. Finally, if workers are constrained because they don't have the training for a job they seek, they can use on-demand work to build up a résumé of experience showing that they have what it takes to do a specific job.
Building better work is neither a social nor a technical challenge. It will always be both.
To understand how workers navigate the sometimes sterile on-demand environment, we circumvented the API. We interviewed workers in person, we studied how they go about their day, and we measured their activity in aggregate. Our key finding is that workers are putting the humanity and meaning back into their jobs. First and foremost, they are adding collaboration back into their work. They are re-creating the office watercooler, albeit in an online or virtual environment, to provide and receive social support. Workers clearly value peer collaboration they are spending their own time and money building online forums, away from the platforms they work on, when they could be spending it working and earning. Workers also collaborate to find good work and reduce the overheads they experience in doing ghost work, which sheds light on some of the transaction costs that workers face. They don't get paid to search for tasks, to learn how to do them, or even how to learn how to use the on-demand labour platform in the first place. Ghost work shifts many of the transaction costs to the workers and requesters hiring them. So workers collaborate to reduce the toll of the unpaid work they have to do to make it in the on-demand economy. […]
[In our book Ghost Work], we offer a list of our top recommendations for turning ghost work into sustainable, on-demand employment opportunities. As we hope these arguments make clear, building better work is neither a social nor a technical challenge. It will always be both. Folding technology into our work lives requires the engineers and businesses building technologies to “think socially” about what people need in order to do good work in equal measure, and crafting the future of work requires that policy makers deeply understand the technologies introduced in the workplace. As more people turn to ghost work—or have their formal employment turned into ghost work—we have a chance to learn from labour history and people's experiences of ghost work today to tackle its technical and social malfunctions. There is still time to bring jobs out of the shadow of Al and make them equitable and dignified employment of which all parties involved will feel proud.
Find out more about Ghost Work by Mary L. Gray and Siddharth Suri
(May 2019, © Harper Business Books)
|Digital Inclusion||Artificial Intelligence||Future of Work|
Please sign in or register for FREE
If you are a registered user on The OECD Forum Network, please sign in