HF — Your article examines a relatively new kind of online labor market — Amazon’s Mechanical Turk. What is Mechanical Turk, and how does it work?
SN — Mechanical Turk (which is often abbreviated as MTurk) is a very popular crowdsourcing platform. It’s a website where requesters can post tasks and workers can do those tasks for pay. Typical tasks include things like tagging images or transcribing video or audio, which are all tasks that are easy for humans but hard for machines. In fact, MTurk has been absolutely essential for machine learning and artificial intelligence; it is how many scientists and engineers produce large batches of the accurately labeled data that helps artificial intelligence to learn which data goes under which category. For example, if you want to use machine learning to distinguish between photos of cats and photos of dogs, it can be helpful to have a set of labeled photos that help the algorithm figure out what to look for. This is how MTurk workers or “Turkers” have helped fuel the “AI Revolution.”
HF — This sounds as though it’s very close to the perfectly competitive labor markets that some economists hold out as an ideal. Is that what you find?
SN — At first glance, MTurk looks like the archetypical competitive market; after all, here is a market with no legal barriers to entry, many employers and none of the noncompete or no-poaching agreements that have become increasingly common in the regular economy. In a competitive market, however, you would expect that job applicants will care a lot about the money they receive, so that a 10 percent increase in the wage might double or triple the number of Turkers who accept the job. Instead, what our experiments and observational data reveal is that the supply of labor facing each requester is extremely “inelastic,” which means in ordinary language that Turkers don’t respond very much to increased pay — a 10 percent increase in the wage actually increases the number of Turkers by only 1.4 percent. This implies a considerable degree of market power, and we have evidence that “requesters” (the people who hire Turkers) are using this power. But this is about more than just one crowdsourcing market. Economists worry about monopoly power, where suppliers are able to charge higher prices, but they also worry about monopsony, where buyers have sufficient market power to squeeze suppliers and pay them less. If there is monopsony power in this market, arguably the archetype of the laissez-faire market, then it is probably much more pervasive in both online and offline labor markets than previously believed.
HF — What are the most plausible reasons employers have such a striking degree of power in these markets?
SN — Monopsony power may arise because of employer concentration on the platform, because of search frictions that make it harder for Turkers to locate higher-paying tasks, or because of idiosyncratic preferences over task characteristics that make the market less competitive. Research has shown that all three of these reasons are at play in MTurk. First, although concentration alone isn’t a great metric of market power, about 10 percent of all requesters post approximately 98-99 percent of all tasks to the platform. Second, the MTurk search interface is crude, so that workers often have to resort to communicating via off-platform online forums to find good HITs. Third, there is evidence for task specialization and differentiation — some people really like doing some tasks and not others, and so the acceptance rate isn’t that responsive to the wage.
HF — What do these and other findings by you and your colleagues imply for broader labor markets, especially as we move more and more to online ‘matching’ systems and platforms?
SN — Platforms can be designed to favor one side of the market over the other. Specifically, labor markets may be set up to favor employers at the expense of workers, since employers are harder to find than workers in most cases. This is a result of an array of small, seemingly inconsequential design decisions that together make it too hard for workers to find jobs they really want to do, so they settle for the job they can easily find. For example, workers can’t sort by real wage and the platforms don’t provide information about whether the requesters have good or bad reputations. Also, workers have to simply take whatever wage requesters set with no real room for bargaining. There is no obvious reason the last problem couldn’t be fixed pretty easily if the platform designer wants it to.
HF — In an earlier era, workers organized themselves into unions to push back against the power of employers. Do we see anything similar happening here?
SN: Yes, this is quite an interesting arena, with implications far beyond MTurk. There are two different questions. One involves “collective action for data” generally, which is independent of work platforms like Mechanical Turk. There’s an argument to be made that people aren’t being paid for the valuable data that they produce, even though it fuels the profit model of big e-commerce firms. Eric Posner and Glen Weyl have drawn popular attention to this problem in their book on radical markets. I think there is a lot of potential for organizing “data strikes” (maybe even via smart assurance contracts on the bitcoin competitor Ethereum) that would force data to be paid closer to its social value.
The second involves “organizing on labor platforms.” This is being worked on by certain national workers groups, who are concerned with traditionally “gig” work and believe that commercial platforms continue to serve investors and consumers at the expense of workers. Some of these organizations have begun to produce their own “worker-owned” platforms. For example, the National Domestic Workers Alliance has launched Alia, the first portable benefits for gig workers in the country that is essentially owned by the workers. To survive in a market, though, worker-owned platforms need workers to organize collectively; if the supply of labor can’t be restricted, the “pro-employer” platform will always have an advantage. So there is an interesting complementary relationship between worker collective organization (e.g. unions) and worker ownership of labor platforms.
This article is one in a series supported by the MacArthur Foundation Research Network on Opening Governance that seeks to work collaboratively to increase our understanding of how to design more effective and legitimate democratic institutions using new technologies and new methods. Neither the MacArthur Foundation nor the network is responsible for the article’s specific content. Other posts can be found here.