Automation and Inequality

By Daron Acemoglu and Pascual Restrepo

Posted on May 2, 2022


Why have the US and many industrialized countries seen rising wage inequality go hand in hand with modest productivity gains? In this article, we consider why economists have often failed to offer compelling answers to this conundrum. In contrast, our recent research offers robust empirical evidence that automation has been the key driver of change in the US wage structure since 1980.


Labor market inequality has risen significantly in many industrialized economies over the past four decades. This trend is particularly clear in the US, where the rise in real wages of workers with a postgraduate degree has been accompanied by a significant decline in the real wages of low-education workers. The real earnings of men without a high-school degree are now 15% lower than in 1980. Simultaneously, many of these economies have experienced a decline in the labor share in national income.


Despite a voluminous literature on both topics, these trends remain imperfectly understood. Leading explanations relate to the changing nature of technological progress. For example, computers are argued to be skill-biased technologies that have raised the productivity of skilled workers, especially those with college or postgraduate degrees, more than those of less skilled workers.


Why do economists struggle to explain rising wage inequality?


In the most canonical approach, technological progress increases the productivity capital and labor inputs, which are combined to produce output in the economy. Skill-biased technical change (SBTC) in this framework corresponds to new technologies directly augmenting and increasing the productivity of skilled-workers, which then increases inequality because it increases skilled wages more than those of lower-skill workers. Likewise, in this framework, improvements and machinery can also increase the productivity of capital inputs. In both cases, however, technologies tend to increase wages for all worker types, because they are either increasing worker productivity or the productivity of the equipment that is complementary to workers.


These popular and influential frameworks are problematic, however. First, they lack descriptive realism and clear empirical support. Most technologies improve the productivity of a factor in some tasks (for example, a better paintbrush makes a worker better at painting, but not necessarily other tasks), improve the productivity of some industry, reallocate some tasks from one factor to another (as with the spinning and weaving technologies that started the British Industrial Revolution in the 18th century), create new tasks, invent new goods, or introduce new ways of combining existing tasks or intermediates. None of these easily fit into the factor-augmenting framework.


Second, and perhaps more importantly, these frameworks make a range of counterfactual predictions. For example, skill-biased technical change can benefit college graduates more than high school graduates, but should never reduce the real wages of high school graduates. But declining real wages of low-education men has been a persistent trend in the US labor market over the past four decades. Similarly, with just factor-augmenting technological changes, it is difficult to have instances in which new technologies reduce labor demand, employment and wages. Yet again, we have plenty of examples of new technologies, especially automation technologies such as industrial robots, that have been associated with lower wages and employment.


“Much of the change in US wage structure is driven by the automation of tasks performed by certain types of workers in some industries, such as those in manufacturing replaced by industrial robots.”


Third, and relatedly, a framework based on factor-augmenting technologies does not generate meaningful changes in the labor share. For realistic values of the elasticity of substitution between factors, generating the changes in labor share experienced in US manufacturing would require huge changes in technology, which should be associated with very large increases in productivity. We do not observe these increases in the data. Likewise, to match the observed changes in the skill premium, the standard SBTC model would need unrealistically large changes in productivity.


Task displacement and falling wages


In recent work, we propose an alternative approach to wage inequality. We argue that much of the change in US wage structure is driven by the automation of tasks previously performed by certain types of workers in some industries, such as the blue-collar workers in manufacturing replaced by numerically controlled machinery or industrial robots. Workers who have not been displaced from the tasks in which they have a comparative advantage, such as those with a postgraduate degree or women with a college degree, enjoyed real wage gains; low-education men, and those who used to specialize in tasks and industries undergoing rapid automation, have experienced stagnant or even declining real wages.


The striking empirical finding of our work is that a simple measure of task displacement explains much of the recent changes in the US wage structure. Specifically, for 500 worker types (distinguished by gender, age, education, race and native/immigrant status), we construct measures of their specialization in different occupations and industries in 1980, emphasizing routine tasks that can be automated. We then construct a measure which captures whether workers specialized in routine tasks in industries that experienced subsequent labor share declines – a telltale sign of automation – saw their relative wages fall between 1980 and 2016. This task displacement measure explains 70% of changes in wage structure. Put simply, rising inequality in the labor market is largely accounted for by declining relative wages for routine tasks at industries susceptible to automation.


“Our model can explain a sizable fraction of the real wage declines observed in the data. It also explains how automation can transform the wage structure while having a tiny impact on productivity growth.”


The main fact we document is very robust. In particular, the link between our measure of task displacement and real wages is unaffected when we control for changes in industry markups, deunionization, import competition from China, and other non-automation technological developments (and these competing variables do not appear to be important for explaining changes in the US wage structure). Moreover, when we control for various forms of SBTC (for example, allowing productivity to change over time according to workers’ educational levels), our measure of task displacement still explains 50%-70% of observed changes in wage structure, while traditional SBTC proxies account for about 8%.


Why do some workers suffer, while others benefit?


To explain why workers specialized in automated tasks suffer wage declines, we start with a model in which each industry performs a range of tasks, some of which are “routine” and susceptible to automation. There are several groups of workers, each with a different comparative advantage across tasks and industries. While we allow technology to directly complement/augment different types of workers, the innovation of our model is to allow for automation technologies that increase the productivity of capital in certain routine tasks that used to be performed by workers.


This setting delivers three key results. First, by allowing technological change to reduce the wages of displaced workers, our framework can explain the puzzles of other SBTC models such as the association of rapid automation with slow productivity growth. Second, we derive a simple equation linking the wage changes of a demographic group to the task displacement it experiences. Third, task displacement can be measured by the group’s employment share in routine tasks at industries undergoing automation. In turn, industry-level automation is tightly connected to changes in labor share in that industry.

Fig. 1 – Evolution of US wage inequality

Fig. 1 – Evolution of US wage inequality


Although our analysis provides evidence of a strong negative relationship between task displacement and relative wage changes across worker groups, it misses three indirect effects affecting real wages. First, our results are not informative about real wage level changes. Second, our evidence does not account for ripple effects, which result from displaced workers competing against others for some of their tasks and bidding down their wages. Third, because automation and task displacement are concentrated in a handful of industries, they can change the sectoral composition of the economy, shifting the demand for different types of workers.


To account for these issues, we explore the implications of task displacement for the wage structure, real wage levels, productivity, output, and sectoral composition of the economy. Our conceptual framework provides explicit formulas to calculate the impact of all these effects in combination with our measure of task displacement, as well as cost-saving gains from automation, product demand elasticities, and the ripple effects between different groups of workers.


We find that task displacement accounts for about 68% of the observed change in relative wages during this period and explains 85% of the observed increase in the college premium. Finally, task displacement alone closes the gender gap by about 12%. Importantly, these sizable distributional effects are accompanied by modest increases in the average wage level, GDP and productivity (which increases by a mere 2% from 1980 to 2016). As a result, our model is capable of explaining a sizable fraction of the real wage declines observed in the data and displayed in Figure 1. In doing so, our model also explains how automation can transform the wage structure while having a tiny impact on productivity growth.


Key Takeaways

  • Up to 70% of changes in the US wage structure since 1980 are accounted for by the relative wage declines of workers specialized in routine tasks in industries experiencing rapid automation.
  • We find robust evidence of a simple relationship between the wage changes of a demographic group and the task displacement it experiences.
  • Our task displacement measure captures the effects of automation rather than rising market power, markups, deunionization, or import competition, which themselves do not appear to play a major role in US wage inequality.
  • Our evaluation of the full effects of task displacement explains how major changes in wage inequality can go hand-in-hand with modest productivity gains.


Illustration: robots painting a car

Acemoglu, Daron and David Autor (2011) “Skills, tasks and technologies: Implications for employment and earnings,” Handbook of Labor Economics, 4: 1043-1171.


Acemoglu, Daron and Pascual Restrepo (2018) “The Race Between Machine and Man: Implications of Technology for Growth, Factor Shares and EmploymentAmerican Economic Review, 108(6): 1488-1542.


Acemoglu, Daron and Pascual Restrepo (2019) “Automation and New Tasks: How Technology Displaces and Reinstates LaborJournal of Economic Perspectives, 33(2): 3-30.


Acemoglu, Daron and Pascual Restrepo (2020) “Robots and Jobs: Evidence from US Labor MarketsJournal of Political Economy, 128(6)


Acemoglu, Daron and Pascual Restrepo (2021) “Automation, Tasks and the Rise in US Wage Inequality” mimeo.


Autor, David H., Frank Levy and Richard J. Murnane (2003) “The Skill Content of Recent Technological Change: An Empirical Exploration," The Quarterly Journal of Economics, 118(4): 1279-1333.


Goldin, Claudia and Lawrence Katz (2007) The Race between Education and Technology, Belknap Press.


Graetz, Georg and Guy Michaels (2018) “Robots at Work,” Review of Economics and Statistics.


Katz, Lawrence, and Kevin Murphy (1992) “Changes in Relative Wages: Supply and Demand Factors,” Quarterly Journal of Economics 107(1): 35-78.


Lin, Jeffrey (2011) “Technological Adaptation, Cities, and New WorkReview of Economics and Statistics 93(2): 554-574.


Tinbergen, Jan (1974) “Substitution of Graduate by Other Labor,” Kyklos 27(2): 217-226.


Zeira, Joseph (1998) “Workers, Machines, and Economic Growth,” Quarterly Journal of Economics, 113(4): 1091-1117.


The preceding is republished on TAP with permission by its author, Professor Daron Acemoglu and by the Toulouse Network for Information Technology (TNIT). “Automation and Inequality” was originally published in TNIT’s February 2022 newsletter.


Daron Acemoglu is MIT’s Institute Professor in the Department of Economics. Professor Acemoglu is a leading thinker on the labor market implications of artificial intelligence, robotics, automation, and new technologies. His innovative work challenges the way people think about how these technologies intersect with the world of work. Professor Acemoglu’s recent research focuses on the political, economic and social causes of differences in economic development across societies; the factors affecting the institutional and political evolution of nations; and how technology impacts growth and distribution of resources and is itself determined by economic and social incentives.


Pascual Restrepo is Assistant Professor of Economics at Boston University. His research focuses on the impact of technology on inequality, labor markets and productivity. His current research examines the impact of technology, and in particular of automation, on labor markets, employment, wages, inequality, the distribution of income, and growth.