Research & Policy Briefs From the World Bank Malaysia Hub No. 16, August 2018 The Future of Work: Race with—not against—the Machine Lay Lian Chuah, Norman V. Loayza, and Achim D. Schmillen Will the revolution in digital and information technologies make us obsolete? Will jobs be lost and never replaced? Will wages drop to intolerable levels? History and economic theory and evidence suggest that in the long term, such fears are misplaced. However, in the short and medium term, dislocation can be severe for certain types of work, places, and populations. In the transition period, policies are needed to facilitate labor market flexibility and mobility, introduce and strengthen safety nets and social protection, and improve education and training. The Fear: Are We Running Out of Jobs? countries, the overall evidence on the relationship between employment growth and the skills distribution is both more There is growing fear that recent and emerging breakthroughs in tentative and mixed. While the relationship has been U-shaped technologies such as artificial intelligence (AI) and robotics will for countries as diverse as Malaysia, Poland, and Turkey, patterns lead to the wholesale replacement of human workers by for China and a range of other developing countries have differed machines and an era of mass joblessness and even wider income (figure 1). This diversity is likely to be related to the interaction inequality. The U.S. magazine Mother Jones reports, “Smart between local labor market conditions, including the skills machines probably won’t kill us all—but they’ll definitely take distribution, and the technologies that are adopted. our jobs, and sooner than you think,” while the British newspaper The Guardian argues, “Technology is hollowing out the middle The Past of Work: Have We Been Here Before? class and creating a bifurcated economy.” China’s Global Times notes, “It is not entirely fantastical to suppose that under the rule One way to structure the economic history of developed of the robots, humans would be forced to beg for food since they countries over the last 250 years is to refer to three past Indus- don't have any jobs to do any more.” trial Revolutions that occurred in the 1760s, the 1890s, and the 1970s. In turn, these revolutions can be characterized by the At least since the First Industrial Revolution in the 1750s, technological innovation that propelled them. Thus, the First workers’ jobs and livelihoods have been threatened by machines Industrial Revolution used steam engines and factories to mecha- that can replace them. Facing this threat, the Luddites organized nize production; the Second used electricity, oil, and assembly themselves to destroy weaving machinery in England in the early lines to generate industrial production; and the Third used 1800s. More recently, taxi drivers from Paris to Mexico City to electronics and information technology to automate production. Bogota have blocked streets and at times resorted to violence to protest the advent of technology-enabled ride-sharing services All three past Industrial Revolutions led to large improve- like Uber. Losing our jobs because we have become obsolete as ments in productivity. This in turn raised welfare in developed workers may be one of our greatest fears—and for good reasons: countries to levels previously unimaginable, in terms of both job loss has significant and long-lasting negative effects on future material living standards and leisure (since the 1950s, average employment, earnings, consumption, health, and even life hours per worker have been falling among OECD countries). expectancy. For some individuals, mortality rates in the year after Today, material living standards and leisure in developing a job loss are up to 100 percent higher than would otherwise countries lag far behind those in developed counties. Therefore, have been (Sullivan and von Wachter 2009). the effects of future productivity growth on welfare can be even These concerns have been echoed and studied in economics. more beneficial in developing countries than in developed ones. In his prescient essay on the “Economic Possibilities for Our Yet, productivity gains take time to materialize. In the case of Grandchildren,” Keynes (1930) predicted the decline of employ- electricity, the productivity boom occurred only in the 1920s, ment in the face of modern technologies and labeled it “techno- over 30 years after factory electrification, David (1990) logical unemployment.” Leontief (1983) wondered whether documents. Brynjolfsson, Rock, and Syverson (2018) argue that workers would go “the way of the horses,” replaced by machines. the same has happened with information and communications In the United States and other developed countries, employ- technologies, which started in the 1970s but only in the 2000s ment growth has followed a U-shape in recent decades, increas- have rendered a noticeable increase in productivity. In 1987, ing for low- and high-skilled workers, but declining for middle- Solow famously said, “You can see the computer age everywhere skilled workers, such as factory and clerical workers (Autor, Katz, but in the productivity statistics.” This productivity pause is and Kearney 2006; Goos and Manning 2007; Autor 2015b). This common to most technologies but is particularly pronounced for has resulted in both employment and wage polarization. While general-purpose technologies such as the steam engine, electric- other trends like climate change, demographic change, and ity, computers, and internet. Using them effectively requires a globalization have also affected jobs, a study of the United States transformation of the production process that can take years, as found that those counties (jurisdictions below the state level) well as substantial investment with no immediate payoff. more “exposed” to robots have lost more employment than others (Acemoglu and Restrepo 2017). All Industrial Revolutions have also led to economic transfor- mation and threatened employment. In the past 250 years, There is also some evidence of U-shaped employment however, technological innovation has not produced mass unem- growth for many developing countries. However, for this group of ployment (Gordon 2016). A specific good, type of work, or even Affiliation: Chuah and Loayza: Development Research Group, the World Bank. Schmillen: Social Protection and Jobs Global Practice, the World Bank. Acknowledgement: Gabriel Demombynes, David McKenzie, Rong Qian, Indhira Santos, and Luis Serven contributed with insights, comments, and suggestions to this brief. Objective and disclaimer: Research & Policy Briefs synthesize existing research and data to shed light on a useful and interesting question for policy debate. Research & Policy Briefs carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions are entirely those of the authors. They do not necessarily represent the views of the World Bank Group, its Executive Directors, or the governments they represent. Global Knowledge & Research Hub in Malaysia The Future of Work: Race with—not against—the Machine Figure 1. Annual Average Change in Employment Share, circa 1995–2012 200 150 100 50 Percent 0 -50 -100 -150 -200 -250 Panama Guatemala Turkey South Africa Malaysia Honduras Mexico Ukraine Poland Mauritius Tanzania Croatia Uganda Serbia Bolivia India Jamaica Sri Lanka Egypt Uruguay Peru Macedonia, FYR Philippines Lithuania Dominican Rep. Thailand Russian Fed. Bhutan Costa Rica Kazakhstan Namibia Ghana Pakistan Barbados Nicaragua Argentina Mongolia Botswana Ethiopia China High-skilled occupations (intensive in non-routine cognitive and interpersonal skills) Middle-skilled occupations (intensive in routine cognitive and manual skills) Low-skilled occupations (intensive in non-routine manual skills) Source: World Bank 2016. a sector in the economy can dwindle and even disappear in the information to provide a wide array of goods and services, from advent of new technologies. However, what is true for one automated manufacturing and transportation to bookkeeping sector, product, or job has not been true for the economy overall and judicial decisions (Brynjolfsson and McAfee 2011, 2014). (Autor 2015b). The disruption caused by the Fourth Industrial Revolution The example of farming in developed countries is instructive. appears particularly palpable in developed countries, but there In the United States between 1900 and 2000, farming went from are also growing signs of it in the developing world. In the Philip- being the main employer in the economy, with 41 percent of all pines in recent years, for example, the business process jobs, to employing only 2 percent of workers, according to data outsourcing industry has become a major sector of economic from the U.S. Department of Agriculture. Over this century, activity and source of well-paying jobs, employing more than 1 productivity gains allowed agriculture to feed a growing popula- million people. However, some companies in the industry have tion with fewer workers, while the rise of new economic activities recently invested heavily in technology and, for instance, begun created better-paying jobs and opportunities in cities for all replacing call center agents by chatbots powered by artificial workers. In developing countries, farming still plays a relatively intelligence systems. While the impact of technological change is more important role. Yet, even within this country group, its for the moment mostly evident on relatively low-skilled share among overall employment has been in a slow but secular “process-driven” business outsourcing, there are widespread decline. Among low and middle-income countries, employment fears of more general impacts in the medium term. in agriculture as a share of total employment fell from 53 percent to 32 percent from 1991 to 2016, according to the World Bank’s This does not mean that machines will replace all labor or World Development Indicators. that wages will plummet across the board. Computers based on AI are remarkably effective in conducting specific tasks rather Although the positive labor effects of the past three Industrial than replicating human intelligence. The early attempts to Revolutions did materialize in the long run, there was a long imitate humans in the 1970s derailed AI for decades. By contrast, period of time when wages and employment fell or remained the recent success of AI has been based on an algorithmic stagnant even though new technologies were adopted and approach that uses neural networks and deep learning for productivity increased. Allen (2009) dubbed this period “Engels’ well-defined and limited tasks. Human contribution is likely to pause,” after Friedrich Engels’ essays on the British working class. remain the crucial ingredient—the “O-ring,” as Autor (2015b) “Engels’ pause” lasted almost 80 years after the onset of the First calls it. Through this illustration and his reflections on Polanyi’s Industrial Revolution and about 40 years after the Second. It paradox (“our tacit knowledge of how the world works often caused labor disruption and social unrest (as insightfully exceeds our explicit understanding”), Autor (2015a, 2015b) has illustrated in Charles Dickens’ stories), and, arguably, even stressed the strong complementary between machines and political revolutions, such as those sweeping through Europe in humans. the 1840s. The replacement of labor by machines takes time and The Future of Work: Is This Time Different? depends on circumstances specific to a given context. Techno- logical innovations tend to occur in developed countries, and No Industrial Revolution has exactly the same labor market their adoption in developing countries usually occurs with a effects as the preceding ones. Breakthroughs in artificial intelli- time lag. Generally, labor is also much cheaper in developing gence, robotics, and other technologies have led to claims that countries than in developed ones. This further slows down we are on the cusp of a new machine age that will dwarf previous the relative pace of adoption of new technologies in waves of automation in terms of the scale, speed, and scope of developing countries, which implies that in many of them the disruption it causes. A defining characteristic of the Fourth concerns about the implica-tions of the Third Industrial Industrial Revolution seems to be that while, previously, technol- Revolution still appear more urgent than those about the ogy was increasingly able to perform routine manual and cogni- Fourth. However, even low labor costs do not stop tive tasks, in the current digital and computing revolution, technology adoption completely. For instance, Malaysia’s machines can also perform some nonroutine tasks that had been Top Glove is one of the world’s largest manufacturer of rubber hitherto reserved to humans: the application of logic and gloves, with about one-quarter of global market share. As 2 wages in Malaysia have gradually risen over the last 25 years, Research & Policy Brief No.16 the firm has remained competitive by gradually substituting in rigid labor markets (regulatory, search-and-match, behavioral foreign for domestic labor. However, as various factors have frictions) can guide policy reforms. further increased the relative cost of labor, the company is now increasingly looking to automate. What Policies Are Needed? What Can Countries Do? A Framework to Assess the Impact of Technological Today, more people are employed than ever before (figure 2). In Innovation on Jobs and Wages the long run, new tasks and new jobs will be created that are difficult to envision now (in the same way that even the most Acemoglu and Autor (2011) and Acemoglu and Restrepo (2018) knowledgeable and imaginative observer at the beginning of the provide a helpful framework for assessing the employment and 1900s would not have guessed how workers leaving agriculture wage effects of technological innovation. According to this would be employed in the following decades). At the same time, framework, there are broadly speaking two types of innovations: many of the current technological advances widen inequality. enabling technologies and replacing technologies. Enabling The returns to tasks complementing new technologies have technologies expand the productivity of labor and lead to higher grown dramatically, but many low- and mid-skilled jobs are at risk employment and wages. Modern examples are computer-aided of being replaced by automation. The prospect of an “Engels’ design (CAT) and statistical software for economic and social pause” is, moreover, looming in the horizon. This raises the analysis. Replacing technologies, in contrast, substitute for labor, question how to mitigate, if not avoid, the negative effects of making workers less useful and lowering their wages. Modern technological change. examples are industrial robots for car manufacturing and software for accounting and tax reporting. Technological change promises tremendous gains in produc- tivity and welfare. Therefore, “neo-Luddite” policies that aim to The direct effect of replacing technologies is negative on stop or delay the Fourth Industrial Revolution appear misguided. wages and employment. However, these technologies can still Instead, the main policy question is how to maximize the poten- have a positive effect in two main ways. First, the new technolo- tial social gains from technological change. This calls for policies gies can generate complementary tasks. In the United States, for that facilitate labor market flexibility and mobility, introduce and example, after automatic teller machines (ATMs) were strengthen safety nets and social protection, and improve educa- introduced 40 years ago, the number of bank tellers, far from tion and training. dwindling, doubled; tellers’ function became more service- and information-oriented (Bessen 2015). Second, the productivity Policies that make labor unduly expensive induce the effects can be sufficiently large to create wealth and generate adoption of labor-replacing technologies. Labor market reform demand for other jobs (for instance, in tourism and hospitality). should be directed at facilitating labor flexibility and mobility, including international migration. Recent evidence for the United The characterization of enabling and replacing technologies States, for instance, suggests that immigration reduces the depends not only on the technical properties of the innovations negative effects of technological change on the employment of but also on the workers’ abilities and labor market conditions native workers at the lower end of the wage distribution. This is where they are implemented. The same technology can replace because an inflow of immigrants specialized in manual tasks workers in some instances and enable workers in others: those attenuates the downgrading of native workers’ jobs and wages well prepared with complementary skills would benefit the most induced by technological change (Basso, Peri, and Rahman from technological innovations. Therefore, an important 2017). Getting the basic business environment right for firms to challenge for policy makers, educational institutions, and house- invest and hire workers and reducing market failures hindering holds is identifying these complementary skills for future work. startups can similarly help capture the gains of technological Labor market conditions, on their part, can affect how innova- change. The policy principle should not be to protect jobs that tions impact employment and wages. Rigid labor markets would are becoming outdated and unproductive due to technological tend to adjust by shedding labor, while more flexible labor change but to protect people (as the Danish flexicurity approach markets would adjust through wage reductions. Flexible labor to labor market exemplifies; World Bank 2013). markets can also induce workers’ reallocation and mobility in the face of technological shocks, mitigating negative effects on both A more dynamic labor market requires better social protec- employment and wages. Identifying the main sources of friction tion to be both feasible and desirable. Safety nets–including cash Figure 2 . Total Employment, 1991–2016 (1991=100) 220 200 180 160 140 120 100 14 15 16 91 92 93 94 95 96 97 98 99 00 01 02 03 04 05 06 07 08 09 10 11 12 13 20 20 20 20 20 19 19 19 19 19 19 19 19 19 20 20 20 20 20 20 20 20 20 20 20 20 High-income countries Upper-middle-income countries Lower-middle-income countries Low-income countries Source: Authors’ calculations based on World Bank World Development Indicators. 3 The Future of Work: Race with—not against—the Machine transfers to the poor and unemployed—are essential to support share of China’s labor force with at least upper secondary educa- workers (and their families) who may become displaced or tion increased from 6 percent to 29 percent from 1980 to 2015. replaced when new technologies are implemented. Evidence In parallel, the share of the country’s labor force with tertiary from around the world shows that well-targeted and education increased from 1 percent to more than 12 percent, well-designed safety nets make a substantial contribution to the while the share of employment in the private sector jumped fight against poverty and inequality, both in the long run and in from virtually zero in 1978 to more than 83 percent in 2014. The the adjustment to technological and other large shocks (World resulting improvements in human capital and more efficient Bank 2013, 2018). allocation of labor facilitated effective technological adaptation and economic transformation. It can be considered a key factor In the long run, broader redistribution policies—such as behind China’s economic success since 1978 (Li et al. 2017). better and more inclusive public goods, social insurance at least partly decoupled from traditional wage employment, redistribu- The main principle underlying these policies is that technolo- tion of capital market shares, earned-income tax credits, and gies and markets do not produce outcomes; people and institu- even a universal basic income—may be desirable to make sure tions do. The comparison with natural resource wealth is that the technological dividends are spread around the popula- informative: depending on public institutions, it can lead to tion, making everyone an “owner” of the current and potential substantial increase of social welfare or to waste and plutocratic technologies (Freeman 2015). gains. Not less important, educational reform—emphasizing scientific, mathematical, and communicational abilities, as well Conclusion: Race with—not against—the Machine as softer skills such as perseverance, flexibility, creativity, adapt- Keynes’ essay on the “Economic Possibilities for Our Grandchil- ability, and team work—is crucial to develop the complementary dren” was ultimately optimistic, a voice of hope, as the world skills that workers need to benefit from all types of machines and economy was about to plunge into the Great Depression. He technologies. Complementing fundamental education with predicted that technological unemployment would be a tempo- active labor market policies, workforce training, and other rary phenomenon. In the long run, technological innovation opportunities for lifelong learning can encourage workers to stay would bring about higher incomes and quality of life, including engaged and continue to participate in changing labor markets more leisure. Even in light of the challenges brought about by the (Card, Kluve, and Weber 2018; OECD 2017). Fourth Industrial Revolution, this prediction is attainable for the Having the right skills can transform “replacing” technologies entire population and not only for a privileged few—but only if into “enabling” technologies for workers. The “high school move- public institutions promote equality of opportunities, generate ment” in the United States in the early 1900s (which mandated an educational system that favors flexible skills and creativity, and and facilitated children’s stay in school until 16 years of age) was use redistribution policies to share the proceeds of technological a large investment that prepared several generations to benefit gains. With proper public institutions, instead of raging or racing from the structural transformation away from farming and the against the machine, we can race with the machines toward a concomitant Second Industrial Revolution. More recently, the better future. References Acemoglu, Daron, and David H. Autor. 2011. “Skills, Tasks and Technologies: Card, David, Jochen Kluve, and Andrea Weber. 2018. “What Works? A Implications for Employment and Earnings.” In Handbook of Labor Economics, Meta-Analysis of Recent Active Labor Market Program Evaluations.” Journal Volume 4, edited by Orley Ashenfelter and David E. Card. Amsterdam: of the European Economic Association 16 (3): 894–931. Elsevier. David, Paul A. 1990. “The Dynamo and the Computer: An Historical Perspective Acemoglu, Daron, and Pascual Restrepo. 2017. “Robots and Jobs: Evidence from on the Modern Productivity Paradox.” The American Economic Review 80 US Labor Markets.” NBER Working Paper 23285, National Bureau of (2): 355–61. Economic Research, Cambridge, MA. Freeman, Richard B. 2015. “Who Owns the Robots Owns the World.” IZA World of ---------. 2018. “The Race between Man and Machine: Implications of Technology Labor 2015: 5. doi: 10.15185/izawol.5. for Growth, Factor Shares, and Employment” American Economic Review 108 (6): 1488–1542. Goos, Maarte, and Alan Manning. 2007. “Lousy and Lovely Jobs: The Rising Polarization of Work in Britain.” Review of Economics and Statistics 89 (1): Allen, Robert C. 2009. “Engels’ Pause: Technical Change, Capital Accumulation, 118–33. and Inequality in the British Industrial Revolution.” Explorations in Economic History 46 (4): 418–35. Gordon, Robert J. 2016. The Rise and Fall of American Growth. Princeton: University Press. Autor, David H. 2015a. “Polanyi’s Paradox and the Shape of Employment Growth.” In Re-Evaluating Labor Market Dynamics. Proceedings-Economic Policy Keynes, John Maynard. 1930. “Economic Possibilities for Our Grandchildren.” Symposium-Jackson Hole, 2014. Federal Reserve Bank of Kansas City. Nation and Athenaeum 11 and 18 (October). ---------. 2015b. “Why Are There Still So Many Jobs? The History and Future of Leontief, Wassily. 1983. “National Perspective: The Definition of Problems and Workplace Automation.” Journal of Economic Perspectives 29 (3): 3–30. Opportunities.” In The Long-Term Impact of Technology on Employment and Unemployment. Washington, DC: The National Academies Press. Autor, David H., Lawrence F. Katz, and Melissa S. Kearney. 2006. “The Polarization of the U.S. Labor Market.” The American Economic Review 96 (2): 189–94. Li, Hongbin, Prashant Loyalka, Scott Rozelle, and Binzhen Wu. 2017. “Human Capital and China’s Future Growth.” Journal of Economic Perspectives 31 Basso, Gaetano, Giovanni Peri, and Ahmed Rahman. 2017. “Immigration (1): 1–26. Responses to Technological Shocks: Theory and Evidence from the United States.” Unpublished working paper. OECD. 2017. “The Future of Work and Skills.” Paper Presented at the Second Meeting of the G20 Employment Working Group. Bessen, James. 2015. “Toil and Technology.” Finance and Development 52 (1): 16–19. Sullivan, Daniel, and Till von Wachter. 2009. “Job Displacement and Mortality: An Brynjolfsson, Erik, and Andrew McAfee. 2011. Race Against the Machine. Analysis Using Administrative Data.” The Quarterly Journal of Economics Lexington: Digital Frontier Press. 124 (3): 1265–1306. ---------. 2014. The Second Machine Age: Work Progress, and Prosperity in a Time of World Bank. 2013. Risk and Opportunity – World Development Report 2014. Brilliant Technologies. New York: W. W. Norton & Company. Washington, DC: World Bank. Brynjolfsson, Erik, Daniel Rock, and Chad Syverson. 2018. “Artificial Intelligence ---------. 2016. Digital Dividends – World Development Report 2016. Washington, and the Modern Productivity Paradox: A Clash of Expectations and DC: World Bank. Statistics.” In The Economics of Artificial Intelligence: An Agenda. Cambridge: MA: National Bureau of Economic Research. ---------. 2018. The State of Social Safety Nets 2018. Washington, DC: World Bank. 4 Global Knowledge & Research Hub in Malaysia