The effects of traditional information and communication technologies (ICT) on the labour market have been extensively studied. However, studies on the labour market impact of artificial intelligence (AI) and robotics are still in their infant stage. Some studies estimate the future risk of the loss of human jobs due to computerisation from the viewpoint of technological substitutability (e.g. Frey and Osborne 2017, David 2017, Arntz et al. 2017). The theoretical models of automation have been advancing rapidly (e.g. Acemoglu and Restrepo 2018, Aghion et al. 2019), but empirical studies are lagging far behind owing to the lack of detailed data on the adoption and diffusion of recent automation technologies.
Among the new automation technologies, industrial robots are an exception. A large number of studies have explored the impacts of robots on employment using industry-level robot shipment data compiled by the International Federation of Robotics (e.g. Dauth et al. 2017 Graetz and Michaels 2018, Borjas and Freeman 2019, Acemoglu and Restrepo 2019, Destefano et al. 2019, Kromann et al. 2020). Although the results on the impact of robots on aggregate employment are not uniform across studies, these studies generally demonstrate a negative impact on low-skilled and middle-skilled workers. However, studies on the substitutability or complementarity of automation technologies other than industrial robots, such as machine learning and big data analytics, with human workers, have been scarce.
An exception is Morikawa (2017), in which I use data from a survey of Japanese firms and present evidence that the adoption of recent automation technologies is positively associated with the skill level of the firms’ employees. However, I do not explicitly distinguish between AI, big data analytics, and robotics within automation technologies. Accordingly, the paper does not explore the heterogeneity among automation technologies. Brynjolfsson and McElheran (2016) is another example of such a study. They document the diffusion of data-driven decision-making (DDD) in the US and the factors influencing its adoption, by using data from a statistical survey conducted by the government. They show that plants with a higher percentage of workers with a college education are more likely to adopt DDD, and hence establish a positive correlation between education and DDD adoption. However, the scope of their study is limited to DDD.
Heterogeneity among automation technologies
Against this background, we conducted an originally designed survey in 2019 and present new findings on the use of automation technologies in Japanese firms and its relationship with the skill composition of their employees (Morikawa 2020). The survey collected separates information about the firms’ use of the different automation technologies of AI, big data analytics, and robotics, in addition to the skill composition of their workforce from a representative sample of Japanese firms.1
Table 1 presents the cross-tabulation result on the relationship between the use of the automation technologies and the percentages of employees with higher education. The percentages are the ratios of employees with university or higher education (column 1) and employees with postgraduate education (column 2). In the case of AI and big data, the percentages of highly educated employees are higher in firms using these technologies than in firms not using them (Rows A and B). The differences are statistically significant at the 1% level.
Table 1 Use of automation technologies and education of employees
Note: ***: p<0.01, **: p<0.05.
Interestingly, the association between the use of robots and the education of employees is very different from the case for the other two technologies (row C). The percentage of employees with university or higher education is lower in firms using robots than in firms not using robots. When looking at the ratio of employees with postgraduate education, firms using robots exhibit slightly higher percentages than firms not using robots, but the difference is quantitatively small.
We conducted estimations to control firm characteristics such as size and industry and confirm that firms with a large share of highly educated employees tend to use AI and big data. The result suggests that complementary investments in human capital are needed in order to realise the benefits from the development of AI and big data analytics. Conversely, in the case of robots, the estimated coefficients on the share of highly educated employees are small and statistically insignificant.
From the viewpoint of complementarity with cognitive skills measured as educational attainment, this result suggests that automation technologies, such as AI and big data analytics that are used for white-collar tasks are markedly different from robots. Although these technologies are often categorised under one heading of ‘automation technologies’, we should be careful about the difference when considering their impacts on the labour market.
Robots in manufacturing and service industries
However, even among robots, service robots or general-purpose robots expected to be adopted in the service industries may differ in character from industrial robots prevalent in manufacturing facilities. In considering this possibility, we divide the sample into manufacturing firms and firms in the service industries (Table 2). The relationship between the use of robots and education level of employees is markedly different by industry. The relationship in service industries is similar to that found for the use of AI and big data (Row B). However, industrial robots used in manufacturing industries are not positively associated with the percentages of highly educated employees (Row A). The estimations controlling for firm characteristics are generally consistent with the findings observed from the simple tabulation table. This result suggests that industrial robots used mainly in manufacturing industries are different from emerging service robots from the viewpoint of complementarity with employee skill.
Table 2 Use of robots and education of employees
Note: ***: p<0.01, *: p<0.1.
To summarise, our findings from the survey have similar implications with those of Agrawal et al. (2019), which states, “we caution on drawing broad inferences from the research on factory automation in forecasting the net near-term consequences of artificial intelligence for labour markets.” In other words, we should be careful in projecting the results of studies on the impacts of industrial robots on employment onto other new automation technologies including AI.
Acemoglu, D and P Restrepo (2018), “Modeling Automation”, AEA Papers and Proceedings 108: 43–47.
Acemoglu, D and P Restrepo (2019), “Automation and New Tasks: How Technology Displaces and Reinstates Labor”, Journal of Economic Perspectives 33(2): 3–30.
Aghion, P, B F Jones and C I Jones (2019), “Artificial Intelligence and Economic Growth”, in A Agrawal, J Gans, and A Goldfarb (eds.), The Economics of Artificial Intelligence: An Agenda, Chicago: University of Chicago Press, pp. 237–289.
Agrawal, A, J S Gans and A Goldfarb (2019), “Artificial Intelligence: The Ambiguous Labor Market Impact of Automating Prediction”, Journal of Economic Perspectives 33(2): 31–50.
Arntz, M, T Gregory and U Zierahn (2017), “Revisiting the Risk of Automation”, Economics Letters 159 (October): 157–160.
Borjas, G J and R B Freeman (2019), “From Immigrants to Robots: The Changing Locus of Substitutes for Workers”, NBER Working Paper No. 25438.
Brynjolfsson, E and K McElheran (2016), “The Rapid Adoption of Data-Driven Decision-Making”, American Economic Review 106(5): 133–139.
Dauth, W, S Findeisen, J Südekum and N Woessner (2017), “German Robots: The Impact of Industrial Robots on Workers”, CEPR Discussion Paper No. 12306.
David, B (2017), “Computer Technology and Probable Job Destructions in Japan: An Evaluation”, Journal of the Japanese and International Economies 43 (March): 77–87.
Destefano, T, S Haneda and H U Kwon (2019), “Determinants of Structural Adjustment and Employment Use in Japan: Firm Characteristics, Offshoring and Industrial Robotics”, RIETI Discussion Paper, 19-E-067.
Frey, C B and M A Osborne (2017), “The Future of Employment: How Susceptible Are Jobs to Computerisation?”, Technological Forecasting & Social Change 114 (January): 254–280.
Graetz, G and G Michaels (2018), “Robots at Work”, Review of Economics and Statistics 100(5): 753–768.
Kromann, L, N Malchow-Møller, J R Skaksen and A Sørensen (2020), “Automation and Productivity: A Cross-Country, Cross-Industry Comparison”, Industrial and Corporate Change 29(2): 265–287.
Leduc, S and Z Liu (2020), “Can Pandemic-Induced Job Uncertainty Stimulate Automation?”, FEB San Francisco Working Paper, No.2020-19.
Morikawa, M (2017), “Firms’ Expectations about the Impact of AI and Robotics: Evidence from a Survey”, Economic Inquiry 55(2): 1054–1063.
Morikawa, M (2020), “Heterogeneous Relationships between Automation Technologies and Skilled Labor: Evidence from a Firm Survey”, RIETI Discussion Paper, 20-E-004.
1 The survey was conducted by the Research Institute of Economy, Trade and Industry. The number of firms responded to the survey is 2,535.