Human vs. Artificial Intelligence
A residual question lurking in the background is the interaction between human and artificial intelligence (AI). This is one of the deep philosophical debates of our time and will only become more acute as both technologies develop. The current narrative circulating in the popular press is that robots will rule the Earth and that the sliver of engineers who control them will capture all of the economic surplus. However, technology has historically served as a complement rather than a substitute for human output. As such, AI can help improve the quality of marketplaces coordinating labor in this new online world.
Even if humans are the ones performing actual work, AI technology can assist in coordinating, contracting, assigning, distributing, and evaluating work. This evolution is not inevitable but requires conscious choice by several simultaneous actors. As a society, we must choose to build a future that seeks to expand human opportunity, not shrink it.
At its highest level, the core problem with online outsourcing is: who does the job, and at what price? The academic economics and computer science literatures split the problem into two broad categories: auctions and assignment. Auctions allow platforms to develop prices under highly uncertain environments through a variety of bidding rules, some of which are invisible to users. For example, Google runs the largest ad auction in the world with eBay as its primary bidder. These auctions are designed by humans but implemented through machines. They set prices to reflect changes in demand, supply, and other environmental variables.
Assignment algorithms are an alternate, and more direct, approach to determining who performs the job. In general, auctions work better when there is large uncertainty on price and skill. Auctions reveal this information, but auctions suffer because of their complexity. Assignment algorithms are quicker and simpler than auctions but require more environmental knowledge about the optimal price and the set of workers best qualified for the job. There is variation now in the market, as websites like Thumbtack use auctions to solicit bids from service professionals, whereas others, like TaskRabbit, use direct assignment to allocate tasks. It is impossible to know exactly which method will dominate or what the appropriate mix between the two is. Ultimately, both academic and industrial research will experiment with all kinds of mixtures, and different mechanisms will likely perform better under different circumstances.
New technologies take time to cause changes in behavior. When Marconi first invented the radio in 1895, the prior form of communication was primarily newspapers. Early media companies bought radios, and their radio shows simply involved reading newsprint over the airwaves. Eventually, companies realized that radio opened up new opportunities that were unavailable before, marking the birth of radio talk shows. When television reached a mass audience in the late 1950s, early television shows were videotapes of prior radio shows. (Blanchard, Margaret. History of the Mass Media in the United States: An Encyclopedia. New York City, NY: Routledge, 1998.)
Eventually, media companies again realized television that provided new opportunities unavailable to radio and made full use of the broad spectrum of formats now occupying modern TV (like the Kardashians).
This same shift is occurring through online outsourcing. The early entrant into the online marketplace was oDesk, a website that connected buyers and suppliers of labor through an online directory. The initial technology for oDesk was designed to mimic face-to-face employment. The website monitored the workers through screenshots, and payment was based on hourly wages (The Workforce in the Cloud 2013). oDesk had the greatest revenue share of online labor markets. In 2012 alone, the company reported more than 500,000 hours of work time per week, and the company’s annual earnings are projected to grow from $1 billion in 2012 to $10 billion by 2020. If this trend continues to permeate the entire online labor market industry, the scope of the growth will be enormous (Kokkodis and Ipeirotis 2016). At the same time, Elance used fixed-price contracts, which are a form of output-based pay. These two companies proceeded simultaneously until they merged together to form a single company, Upwork, in 2015 (Pofeldt 2015).2
The existing online labor markets are still in their early innings. Even AMT and Upwork, which offer fixed-price payments, require high touch interaction from the requester/employer. In particular, the employer needs to specify the terms of the agreement of work to be performed, a timeline, and a price. A better longer-term option is to automate all of these choices; software will set the price and eventually negotiate contracts on a somewhat individualized basis with workers. Existing markets also do not make use of worker performance data in a granular or dynamic way. Upwork allows employers to take screenshots of their workers for monitoring, but this information is not explicitly embedded in compensation. However, as the Informativeness Principle of Holmstrom (1979) shows, all relevant performance information should be part of compensation. Otherwise, the employer is effectively leaving money on the table.
As transaction costs fall over time, this information will be incorporated into compensation and boost productivity. My own research shows these productivity gains can be large and help with all manners of decisions, such as when and whether to terminate a worker. When the information from a performance system is coarse or incomplete, low-performing workers stay employed for too long, draining output. On the other hand, when this information is used for compensation and evaluation, it allows the employer to terminate low performers and raise output. In particular, the ones who stay work harder than the ones before (Ray 2007).
Case study: The Philippines
The Philippines, a country of 92.34 million people and widespread English education, has historically served a primary role in business process outsourcing (BPO). Call centers have spread like wildfire across the Philippines in the last decade. In some sense, the first wave of outsourcing focused on call centers in India;now, those centers have shifted to the Philippines. The increase in the connectivity of the Internet will bring online labor markets to even more remote regions of the islands, and I have direct evidence of this.
In 2012, I was seeking cheap dictation transcription services for my notes taken during academic conferences. I located Cheryl Lamorin on oDesk, a nurse living in Cebu, 355 miles from the capital city of Manila. She chose her town because her mother was sick, and she spent half of her time caring for her at home. She needed to cut back on her hours at the hospital in order to do so, but, of course, stay-at-home care does not require full-time attention. Thus, she turned to oDesk to fill her time and earn extra income. I employed Cheryl for two full years until she formed her own medical transcription company with her friends, many of whom were stay-at-home moms.
Critics of online outsourcing have called it the digital sweatshop, effectively holding the workers to low wages and confining them to a life of poverty. However, these arguments fail for the same reasons the basic sweatshop arguments fail: all trade is voluntary, especially in online markets with large amounts of transparency and low costs of switching employers. Moreover, people like Cheryl have a low opportunity cost because they need to stay home for personal reasons. It was impossible for her to take a job outside of her home, and therefore, she was open only to online work. She often thanked me for enabling her to support herself while her mother was sick.
Over time, people like Cheryl will have multiple job opportunities. Their opportunity cost will rise because their options within the online market itself will increase. Over the long term, this will raise the level of wages, which, in general, should follow the long-term rise in productivity once the market develops more fully. The overall macro effect is that the low transaction costs of the Internet will allow the market to spread to remote areas of the world where formal businesses are hard to establish, especially when the population is rural and scattered. This is how these online labor markets will increase the global supply of labor.
Measuring both traditional and online outsourcing poses a challenge because the data is incomplete and highly aggregated. Traditional outsourcing refers to the permanent shifting of jobs to an external provider. Online outsourcing refers to interfacing through a website to contract directly with individuals, some of whom are likely located abroad. Publicly available data is much higher on the former than the latter, primarily due to aggregate government statistics.
At the highest level, some preliminary evidence from the International Monetary Fund (IMF) and the Bureau of Labor Statistics (BLS) shows that traditional outsourcing is increasing. For example, outsourcing in the finance category refers to functions traditionally performed in the finance department of a U.S. manufacturing company. Now, they either contract with a firm abroad or establish a separate division in an overseas office, all within the same company. Aside from financial functions, most of the other services in the chart are steadily increasing over time. This tends to confirm public perception that outsourcing and global integration are increasing.
Precisely measuring the economic magnitudes of this shift is another matter altogether. The two major sources of data on traditional outsourcing are the Mass Layoff Statistics of the BLS and the data from the Bureau of Economic Analysis of the Department of Commerce (Mankiw and Swagel 2006) Levine (2012) conducted a comprehensive study on the BLS data, which tracks firms that lay off more than 50 workers who are out of work for more than 30 days. The BLS survey shows that most of these job losses did not involve any relocation of work. However, there are limitations because the study does not capture work moved outside of layoffs, nor does it measure changes in future hiring as a result of outsourcing. Nonetheless, the survey illustrates that most job losses do not result in relocation of work.
What are the responses of companies to these macroeconomic trends? Again, the data is mixed. Some studies find U.S. parent companies use more labor as outsourcing increases. This is consistent with the logic that outsourcing reduces the cost for the firm, freeing up resources for investment elsewhere. These investments can lead the firm to hire more labor in new areas. Outsourcing increases job losses for domestic workers without a high school degree, but there is little evidence significantly related to job loss for workers with college degrees (Kemeny et al. 2013). This supports the intuitive notion that education is an antidote to outsourcing-related job loss.
The only data available for AMT comes from disclosures from Amazon itself. As of 2011, AMT claimed that over 500,000 workers were registered on their website from over 190 countries (Aws 2011). Online forums, like Turk Nation, have speculated that it is difficult for workers to earn a full-time wage on AMT given the low prices of the jobs (Guarino 2015). An NYU Professor calculated the average compensation per hour for an AMT worker in 2009 as $2.30, well below the U.S. minimum wage, though still above the income for much of the developing world (Guarino 2015).
Some preliminary data published through white papers on their website shows that, at least for Elance, the total amount of outsourcing has increased over time, ranging from one million tasks outsourced in 2011 to 3.5 million tasks in 2013 (Elance 2013). These facts are undoubtedly part of the company’s marketing apparatus because they show their success as a platform in matching buyers and sellers. However, it also shows that online outsourcing is on the rise.
In my view, these numbers are just the tip of the iceberg. The existing evidence on online outsourcing is more prospective than retrospective. It shows that online outsourcing passes a basic “proof of concept” test and can ultimately lead to massive growth. This is especially apparent when placed in the context of the large resources Facebook and Google are deploying to bring people online, as discussed earlier. The current level of outsourcing is just a sliver of what is possible, since connectivity in the world still reaches only half the globe. 4 Once this connectivity expands, it is entirely possible that outsourcing will explode in both quantity and quality, especially because these new people brought online will primarily contribute their labor to the global economy rather than consumer purchases.
Much of the growth in IT outsourcing has occurred through services and businesses located on the coasts of America, which makes sense given the high concentration of service businesses in California and along the eastern seaboard. BPO refers to all tasks traditionally relegated to the business back office, such as customer service, accounting, tech support, performance evaluation, and legal services. Gartner finds that global spending for IT services was approximately $932 billion in 2013 and is expected to grow to $967 billion in 2014, a 3.8% . growth from 2013. The market for BPO itself was predicted to grow by 6.2% in 2013. This suggests a global increase in BPO as a whole—and given the increasing levels of outsourcing, the relative growth of outsourcing in Asian countries captures a disproportionate share of the growth. Finally, an Ernst & Young research publication found the five-year compound growth rate for worldwide BPO would be 5.3% through 2016.
(As of 2015, nearly 94% of the world’s population receives a mobile phone signal, 48% are covered by mobile internet, and nearly 28% have subscribed to a data package (Kende 2015). As of 2014, there were nearly 711 million broadband connections worldwide (The State Broadband 2014)
Trade in services
Several policy questions arise in this new world of online outsourcing. Should the government subsidize or discourage online outsourcing? How should the government handle assistance to those hurt by outsourcing, if at all? Can the government invest in education for Americans to compete in, or help develop, these online labor markets themselves?
At its highest level, online outsourcing is another form of international trade. A company or individual can turn to a website and locate a worker who performs a service for a fee. Because the transaction is voluntary, both parties are better off. The buyer is better off because the value of his/her time is higher than the price paid for the work, while the seller is better off because his/her income from work exceeds the opportunity cost. This elemental transaction between a buyer and seller lies at the heart of all economics.
Economists such as Adam Smith and David Ricardo have recognized the benefits of trade at the international level for over two centuries. Ricardo referred to this as comparative advantage; in a world with two countries and two industries, even when one country is more productive than the other in both industries, both nations are better off if they trade. Each nation specializes in their highest comparative advantage.
Online outsourcing is the modern manifestation of international trade and should receive the same broad support from economists as free trade does.