In 1960, scientists established that atmospheric concentrations of CO2 were rising. Despite this advance in knowledge, the understanding of global warming later remained quite muddled even a decade later. Historian of science Spencer Weart summarizes the situation. “In the early 1970s, the rise of environmentalism raised public doubts about the benefits of human activity for the planet. Curiosity about climate turned into anxious concern. Alongside the greenhouse effect, some scientists pointed out that human activity was putting dust and smog particles into the atmosphere, where they could block sunlight and cool the world. Moreover, analysis of Northern Hemisphere weather statistics showed that a cooling trend had begun in the 1940s. The mass media (to the limited extent they covered the issue) were confused, sometimes predicting a balmy globe with coastal areas flooded as the ice caps melted, sometimes warning of the prospect of a catastrophic new ice age.”a
Our current understanding of the Robot Revolution is equivalent to the understanding of global warming circa the 1970s. We know some things with certainty. Computers are eliminating jobs involving structured tasks in manufacturing, clerical work, and some other “mid-skill” occupations. Computers are creating jobs in some occupations, particularly for the technically skilled. Computerized work, unlike global warming, should increase GDP. Beyond these facts lies a broad landscape of speculation and spin. We do not know much about the Robot Revolution’s net effect on employment and wages just as we do not know much about the speed at which the revolution is proceeding. As Weart might say, the mass media (and the rest of us) are confused.
One cause of our confusion is the absence of sustained research of the type that eventually clarified our understanding of global warming. Given the potential disruption of rapidly advancing computerized work, this research program is needed badly. What follows are some thoughts on how such a research program could be advanced including obstacles that must be overcome. Our observations draw in part from our experience in 10 lunch meetings at MIT over the past two years that have brought economists and computer scientists together for the purpose of mutual education.
Just as global warming research has required multiple disciplines, researching the Robot Revolution will require cross-disciplinary learning beginning with computer scientists and economists. Current writing on the revolution is weak in part because most economists know less than they should about artificial intelligence while computer scientists have knowledge deficits in economics.
Two examples make the point. Early in the MIT lunch series, an economist asked what computer scientists meant when they said “computers lack common sense.” A member of the computer science faculty responded with a lucid explanation of the common sense problem—the thousands of apparently trivial facts that humans use to perform tasks and that are difficult to capture in software. At that lunch were six economists, five of whom (including us) had written extensively on the impact of computers on work. A reasonable estimate is that no more than one of the economists had previously known of the common sense problem. None had used it in their writings.
The main obstacle facing cross-disciplinary research is an asymmetric incentive structure.
Similarly, in a recent Communications Viewpoint, software developer Martin Ford discusses one possible impact of the Robot Revolution: “Entry-level positions are especially vulnerable, and this may have something to do with the fact that wages for new college graduates have actually been declining over the past decade, while up to 50% of new graduates are forced to take jobs that do not require a college degree.”3
An economist would argue that inferring causality is not so simple. Over the last decade, computers have been spreading through the economy and, as Ford stated, wages for recent college graduates have been declining. But other things have happened over the last decade including the 2008 financial collapse and subsequent deep recession. History shows that recessions that follow a financial collapse last an average of five years4 and all recessions are particularly hard on new labor market entrants. A rigorous estimate of computers’ effects on recent college graduates would have to first control for the recession’s effects.
The main obstacle facing cross-disciplinary research is an asymmetric incentive structure. Economists are paid to study the factors (including computers) that affect labor markets. They know who funds such research, the seminars and conferences where they can present their results, and the journals that will publish the papers they write. To our knowledge, a computer scientist who studies the labor market effects of the Robot Revolution has no such infrastructure.
To the contrary, time taken to study the labor market impacts of computerized work is time taken away from career-advancing activities. Because of this asymmetry, computer scientists who are interested in the impacts of technological advances on employment are likely to be dispersed across institutions rather than concentrated in the way that machine vision groups or natural language processing groups exist in many computer science departments. We return to this point later.
One might wonder whether a second obstacle is unwillingness among computer scientists to consider the potentially negative employment consequences of their work. Economists can offer some comfort on this issue. In the wake of the financial collapse, economists took significant criticism for promoting the deregulation and financial instruments that led to the collapse not to mention failing to predict the collapse in the first place. In practice, the profession shrugged off the criticismsb and continued to pursue problems they found interesting.
It is easy to imagine a collaborative research program beginning with conversations among researchers leading to case studies that generate hypotheses, moving on to econometric analysis and simulations—research designed to get a clearer sense of the future. The question is how to start. More precisely, the question is how to identify and connect people who want to do the work.
Identifying interested economists is straightforward. Virtually all economists who work on the Robot Revolution belong to one of three program groups in the National Bureau of Economic Research (NBER)—Labor Studies, Economics of Education, and Productivity, Innovation and Entre-preneurship—and could be reached quickly through the NBER structure. Identifying interested computer scientists is more difficult since, as we argued, interested individuals are likely to be geographically dispersed and none of the ACM Special Interest groups is directly in this area. We need to send up a flare so that interested computer scientists can respond and be put into contact with each other and with like-minded economists.
A logical candidate for this “flare” would be an ACM-sponsored symposium on the employment effects—positive and negative—of computerized work. The call for abstracts would be disseminated to computer scientists through both Communications and other ACM channels. The call would be disseminated to economists through the program group email lists within the NBER.
Computer scientists who are interested in the impacts of technological advances on employment are likely to be dispersed across institutions.
The symposium would have two goals: to serve as a point of contact for interested researchers and to begin to set standards in a new but important field. As such, the symposium’s goal should be to advance research rather than raw speculation. It is currently too early at this stage to expect collaborative research between computer scientists and economists—we are still at the introduction phase—but at least it is possible to require that submissions be grounded in the authors’ expertise and focus on near term (or past) developments rather than project far into the future.
For an economist, an example of grounded relevant research is an econometric study estimating the employment or wage effects of one or more technological advances.1 This research is historical rather than forward looking but it would lay out the economist’s techniques of model building and data handling in ways that would help inform future collaborative work. A different example is a case study of the implementation of a particular technology2 that can be used to generate hypotheses for future work.
For the computer scientist, an example of relevant research would be the dissection of a technology that, if fully developed, could have significant impact on labor markets. Dissection would include discussion of the obstacles to full development and the author’s assessment of future progress on those obstacles. What navigation and recognition problems must be solved to achieve a fully autonomous vehicle?c What improvements in natural language processing and information retrieval are needed to outperform foreign call centers in answering customer service calls for particular domains?
With these examples in mind, a selection committee of computer scientists and economists would be chosen to write a call for abstracts and to select among the submissions. Presented papers would be discussed by both a computer scientist and an economist. Assuming the revised papers were of sufficient quality, short versions could be published in Communications while full versions could be posted on an associated website or published in another venue. Both the symposium and website could also serve as a meeting point at which interested computer scientists and economists could post contact information and interests and begin correspondence pointing toward collaborative work. Based on informal conversations, there appears to be significant foundation interest in funding a conference of this kind.
Over the next quarter-century, global warming and the diffusion of computerized work each has a potential to profoundly change society. It is past time that research efforts start to reflect their equally disruptive powers.
Join the Discussion (0)
Become a Member or Sign In to Post a Comment