Here at Mind Matters, we have often covered the way that humans are used to supplement Artificial Intelligence. Artificial intelligence has generally been misunderstood as replacing human effort in society, while, in reality, it is usually leveraging it, instead. Whether using humans to find good training data, mining content for intentionality, or even using humans directly within machine learning algorithms, today’s most prominent “AI” systems are actually strange hybrids of humans and computers. As a matter of fact, the market for human supplementation of AI is so large that Amazon has an entire service built around it.
While much of this work is done either for free (oftentimes through games on the Internet) or through traditional paid office work, a growing amount is being done through “microtasks,” through systems such as CloudFlower and Amazon’s Mechanical Turk. These services take tiny jobs that require a human touch (i.e., “what keywords best describe this image”) and pay people small amounts per job to perform the task. Because this is behind the scenes, and oftentimes people don’t even know there is a human there, it is often termed “ghost work.” But who are the workers behind this new class of vocation?
The book Ghost Work: How to Stop Silicon Valley from Building a New Global Underclass attempts to provide a sociological profile for this new kind of work and worker. This book looks into the lives and work of the people on these platforms, what they do on a daily basis, why they do it, and the problems they face. On the whole, the book does a good job of helping people understand this new class of work and why and how it is done, but seems to have some fundamental misunderstandings about choices, tradeoffs, and economics, which make them often misunderstand their own subjects and present “solutions” that would probably be more problematic than the problems they look to solve.
On the good side, this book is the first real book to describe the role of humans in artificial intelligence. Most people assume that artificial intelligence means that humans are being replaced with computers, but this isn’t the case at all. In fact, the authors show that this is a long misunderstood part of technological development, which they call the “last mile paradox.” In nearly every automation from the industrial revolution to the present, there is always some amount of “finishing work” that needs to be done by hand. In the textile industry, this is known as “piece work,” where difficult-to-automate aspects of clothing continue to be done by hand even after the large-scale takeover of industrial machinery. It is called “piece work” because the people that this is outsourced to are paid by the “piece” instead of by the hour.
In the modern day, computer automation requires what can be essentially described as “algorithmic piece work,” where things that are not adequately handled by algorithms can be doled out to people who work by the job instead of by the hour. Most technology users are wholly ignorant that this works even goes on, and assumes that there is automation, not humans, behind all of the tasks that get performed on computers. How do computers tell if an image is pornographic? How can they tell if a “trending hashtag” is a legitimate phenomenon or some kind of scam or hack? How can you tell if someone who cut their hair differently still matches their photo on file? These are all things that computers do not have the context or programming to handle, but still need to be done on algorithmic platforms. These often get farmed out to systems such as Mechanical Turk for a human to process. The book does a great job describing this process and the people behind it.
The book, however, also aims to correct what the authors perceive as injustices towards the workers. However, most of the time the injustices that the authors perceive are either (a) contradicted by the authors’ own reporting, (b) misunderstanding the choices that lay behind the perceived injustice, or (c) something that has nothing to do with the subject (i.e., “ghost work”), but is simply endemic to being underprivileged.
As an example, the authors seem to think that ghost workers should be given the protections of full-time employees, and indicate that it is problematic that they don’t have the same protections. However, why are there different protections for full-time workers than contractors? The authors note that because these workers are not in offices and can work their own hours they therefore aren’t qualified to consider themselves full-time workers. However, the entire reason for considering protections for employees is that, when you work on-site, your employer is responsible for the conditions of your workspace. That is, whatever safety measures are or aren’t there is entirely in the hands of the employer. Additionally, to the extent that an employer asserts their own authority over your life (requiring specific hours, requiring overtime, etc.) also makes them responsible for how they do so. In the case of ghost work, the employee is working from their own home and they can work at any time, day or night. Thus, the employer exhibits no influence whatsoever on the working conditions of the worker and asserts no authority over the worker’s life. Therefore, there is no moral reason why they should treat the worker as someone over which they have a large amount of responsibility.
Additionally, the authors complained about the tough competition for jobs while simultaneously complaining about how difficult it is to join the workforce, specifying the many hurdles people in underdeveloped countries face. However, making it easier to join actually exacerbates the problem of competition. The authors seem fully ignorant of this basic aspect of markets.
The authors also seem to be unaware of the conditions required for the very existence of this type of employment. They spend a good amount of time critiquing the anti-fraud measures of the system, while completely ignoring the fact that the Internet is rampant with fraudsters trying to play such systems. If the systems can’t eliminate fraud, and don’t have the flexibility to do so economically, then this entire area of the economy disappears. Like most people today, the authors seem fully unaware that there are preconditions for the existence of certain types of work.
There is one criticism, however, that I found helpful: Companies need to have a better way to appeal automated decisions. This is true everywhere. Companies are looking to automate, which is great. But having a fallback, a way to address grievances (or even communication) by a human in the company that has authority to resolve or escalate issues, is extremely important. I think that whether technology becomes a tool of freedom or oppression may largely hinge on this one question, not just for ghost work, but across all aspects of the economy.
In all, Ghost Work does a good job of helping people understand how the underbelly of our technological systems work as well as the people making it happen. However, its prescriptions for modifying these work arrangements are incoherent and problematic for the very people they want to benefit.