Machines and workers: How different technologies affect different workers

Sotiris Blanas, Gino Gancia, Tim Lee 10 October 2019

a

A

Machines have been transforming the workplace ever since the Industrial Revolution. Recently, information and communication technologies (ICT), software, and especially robots with enhanced capabilities have fuelled widespread concern that they may replace workers in unprecedented numbers. The concerns are warranted: in 2015 there were an estimated 1.63 million industrial robots performing activities previously done by humans, including welding, assembly and packaging. This number is expected to double by 2020, and future scale and scope are hard to predict.

At least in the short run, some workers will lose their jobs to machines. Autor et al. (2003), Autor and Dorn (2013), and others have established that routine-intensive industries and occupations have declining employment shares, and Acemoglu and Restrepo (2019) find that employment dropped in US commuting zones that were more exposed to robots. 

On the other hand, some workers may benefit from the productivity improvements of new technologies. Even among workers vulnerable to replacement, some may acquire new skills that allow them to shift to jobs less amenable to automation. For instance, Autor and Salomons (2017) found positive employment spillovers to industries less affected by automation, and Graetz and Michaels (2018) found that while industrial robots reduced the employment share of low-skill workers, there was almost no negative effect on total employment. 

This suggests that the impact of machines on workers may differ depending on the type of technology and also the type of worker. We have studied how workers of different skill (education), age, and gender have been affected by ICT, software, and industrial robots from 1982 to 2005, using data from 30 industries spanning roughly the entire economies of ten high-income countries (Blanas et al. 2019). In all countries, workers became higher-skilled and older during this period, while there was also a rise in women’s employment and income shares. We attempted to identify whether these labour market trends we due to the arrival of new machines.

Capital inputs, workers and routine jobs

We used the Dictionary of Occupational Titles (DOT) and the Occupational Information Network (O*NET) to evaluate which jobs are more prone to automation based on the type of tasks they require. We then used occupational data from the US Census to discuss how new technologies might have affected the demand for each worker type, depending on the task content of the occupations they work in. Medium-skill and young workers tend to be employed in occupations that are easier to automate. Women were and still are disproportionately employed in jobs that are more prone to automation (clerical or administrative jobs, for example), but they have shifted toward jobs less prone or shielded from automation (such as managerial jobs) at a faster rate than men, as shown in Figure 1.

Figure 1 Employment shares by gender/occupation

Source: IPUMS census.
Notes: Across 11 1-digit occupations. M/SE: Managers and Self-employed, Mspt: Managerial Support, Professionals, Mining, Mechanics, Technicians, Sales, Transportation, Machine Operators, Administrative Occupations, and Lserv: Low-skill Services. Occupations ordered in descending order of their occupational mean wages from bottom to top, not differentiating by gender.

We estimated a standard labour demand equation in which the dependent variable is the (log-)employment level of each worker type, measured in hours of work, with traditional (non-ICT), ICT and software capital intensities as the main explanatory variables. While traditional capital shows no particular trend, both ICT and software have steadily increased over time with the former accelerating post-1995 (Figures 2 and 3). 

Figure 2 Traditional capital to real value-added

Source: EU KLEMS.
Notes: In each country, capital is first averaged across industries using each industry’s within-country employment share, then averaged across countries without weights. 

Figure 3 ICT and software to real value-added

Source: EU KLEMS.
Notes: In each country, capital is first averaged across industries using each industry’s within-country employment share, then averaged across countries without weights. 

Our specifications always include country-year and country-industry fixed effects, implying that the coefficients of interest are identified by within-country changes in the capital inputs between industries. Industries with faster growth in software capital experienced employment losses for low-skill workers and young workers, relative to other industries. In contrast, faster growth in traditional (non-ICT) capital is associated with employment gains for the low-skilled and women, and faster growth in ICT capital is associated with employment gains for all worker types.

To understand these patterns, we differentiated industries by their exposure to automation, as measured by the routine-share index (RSH) created by Autor et al. (2003). How different types of capital are associated with the demand for different worker groups varies by an industry's RSH: non-ICT capital is associated with higher employment growth in routine-intensive industries, suggesting that relatively low-tech machines may complement routine tasks. By contrast, both ICT and software are associated with employment losses in more routine-intensive industries, and with employment gains in less routine-intensive industries.

Industrial robots and the demand for labour 

These interesting results are just conditional correlations. So we used a novel empirical strategy to identify the causal effects of one of the most prominent forms of new automating technologies: industrial robots. 

A country importing robots from countries that experienced relatively high growth rates in their worldwide exports of robots is more exposed to this type of technology. Hence, we constructed a measure of exposure to the worldwide surge in robots for each of our ten countries using UN COMTRADE data on bilateral trade in industrial robots, available starting in 1996. We then interacted this country-level measure with the RSH index that captures the industry-level scope for automation. Figure 4 shows the trend for each country.

Figure 4 Normalised robot exposure variable

Source: Authors’ calculations based on UN COMTRADE.

Including this new variable shows that industrial robots decrease low-skill employment, while they increase the income shares of high and medium-skill workers, old workers, and men. We also investigated how these effects differ between manufacturing and service sectors, both of which intensively use robotic technology, but have experienced very different employment trends. 

Most of the effects are starkly different between the two sectors: In manufacturing, robots lower low-skill, young, and female employment, while in services, they increase medium-skill and male employment. In both sectors, robots increase the income shares of high-skill, old, and male workers.

Our results are consistent with the view that robots replace workers who perform routine tasks, especially in sectors where automation is more widespread, such as manufacturing. By contrast, they increase employment and incomes in sectors where automation has started more recently, such as in services, a sector in which new occupations are appearing. Given the industrial and occupational composition of these sectors, that robots are likely to complement engineers, product designers and managers – that is, occupations that are dominated by high-skill, more senior, and male workers. Software has a similar effect to robots, whereas ICT capital is associated with employment gains mostly for medium and low-skill workers.

Are robots biased against women?

We also looked for gender bias by studying the effects of robots on the employment levels and income shares of men and women, stratified by skill levels. As noted above, the high-skill employment share of women rose more than men. The substitutability between robots and female workers is mostly driven by lower-skill workers. By contrast, the complementarity of robots with male workers is driven by the high-skilled.

These results may at first seem puzzling. Even as women’s overall labour market outcomes were improving over time, technology seems to have had a negative impact on their employment and relative incomes. The response of women over the medium-run reconciles these facts: robots did not replace women indiscriminately, but rather, only those at lower skill levels. At the same time, women responded by acquiring higher levels of skill, and at a faster rate than men. Similarly, the positive effect of robots on men’s outcomes is also because men were traditionally more educated.

Conclusions

Our findings highlight the importance of distinguishing between technologies that replace humans, such as robots and software, and those that are used by humans, such as ICT. As demonstrated by our investigation of gender bias, our results deliver an encouraging message: it is possible for workers to flourish from advances in technology by acquiring new skills that are complementary to machines, rather than remaining in jobs that are destroyed by them.

Authors’ note: This column is based on a paper prepared for the special issue of Economic Policy on “Automation, Artificial Intelligence and the Economy”. The views expressed here are those of the authors and do not necessarily reflect the views of the National Bank of Belgium or the Eurosystem.

References

Acemoglu, D and P Restrepo (2019), "Robots and Jobs: Evidence from the US Labor Markets", Journal of Political Economy, forthcoming.

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

Autor, D H and D Dorn (2013), "The Growth of Low-Skill Service Jobs and the Polarization of the US Labor Market", American Economic Review 103(5): 1553–1597.

Autor, D H and A Salomons (2017), "Robocalypse Now-Does Productivity Growth Threaten Employment?" Working paper, MIT.

Blanas, S, G Gancia, and S Y T Lee (2019), "Who is Afraid of Machines?" CEPR discussion paper 13802, forthcoming in Economic Policy.

Graetz, G and G Michaels (2018), "Robots at Work", Review of Economics and Statistics 100(5): 753-768.

a

A

Topics:  Labour markets Productivity and Innovation

Tags:  technology, automation, technological change, computers, ICT, productivity, innovation, labour, employment

Research Economist-Microeconometrician, National Bank of Belgium

Professor of Economics, Queen Mary University of London and CEPR Research Fellow

Reader, Queen Mary University of London

Events

CEPR Policy Research