A robot is coming for your job? Well, probably not

In recent years, there has been a revival of public concerns that automation and digitization might result in a jobless future. This debate has been particularly fueled by Carl Benedikt Frey and Michael A. Osborne, who argue that 47% of jobs in the US are ‘at risk of computerization.'

These alarming figures spurred a series of studies that find similarly high shares of workers at risk of automation for other European countries. By focusing on the level of occupations, these studies typically neglect the fact that occupations involve bundles of tasks, many of which are hard to automate. As a result, these studies potentially overestimate the share of workers at risk of automation.

For instance, bookkeeping, accounting, and auditing clerks in Frey and Osborne’s study face a risk of 98% of becoming automated in the near future, although most workers in this profession perform tasks such as face-to-face interactions that seem rather unlikely to become automated soon without extraordinary effort and costs.

In a recent study for the OECD, we analyzed the heterogeneity of tasks at the level of jobs to re-evaluate the share of jobs that could be performed by machines. For this we use internationally comparable task information for 21 OECD countries at the worker level provided by the PIAAC data base.[1] Our findings suggest that the automatibility of jobs is significantly lower when taking into account the actual tasks performed by workers at their individual jobs.

Overall, we find that, on average, only 9 % of all jobs across OECD countries are automatable. The share of automatable jobs varies between 6 % in Korea and 12 % in Austria (see chart). These differences appear to correlate with general differences in workplace organization, differences in previous investments into automation technologies as well as differences in the education of workers across countries. In particular, lower educated workers face considerably larger automation potentials than low qualified workers across all countries in our study. Accordingly, countries with lower shares of tertiary educated workers typically have larger automation potentials.

Whether these results are informative for emerging economies is questionable, as the technological endowment of these countries is typically less advanced. Instead, evidence on past technological change in developed economies might be more informative for potential developments in emerging economies, assuming these countries first have to catch up to the current technological level in developed economies.

The threat from technological advances thus seems to be much less pronounced than suggested by the widely cited set of automation studies applying the method of Frey and Osborne. Still, even our lower figure of 9% seems worrisome at first place, as it suggests that every tenth worker might experience a machine or robot taking over his job. However, the estimated share of ‘jobs at risk’ must not be equated with actual or expected employment losses from technological advances for three reasons:

- First, the utilization of new technologies is a slow and gradual process, due to economic, legal and societal hurdles. For instance, despite the potential for automation, new technologies need to pay off before firms start implementing them in their business strategy. This also involves considering additional costs that are associated with new challenges of digital technologies, such as data and cybersecurity. Moreover, there might be legal and ethical reasons, as in the case of the driverless cars, where humans might not be willing to transfer the decision making to a machine. Similarly, machines might not be accepted as an adequate substitute for humans in health care services or as waiters, despite technical possibilities.

-Second, even if new technologies are introduced, workers typically adjust to changing technological endowments by focusing on those tasks that machines cannot perform. The reason is that new technologies may substitute for certain tasks on the job, but they typically also complement others. As a result, workers may perform different tasks rather than becoming unemployed because of technological advances. In fact, existing evidence suggests that workers increasingly shift worktime from routine work (e.g. reading, writing) to non-routine tasks (e.g. creative, cognitive work) that are harder to automate. For the future of work this implies that monotone and repetitive labor may be replaced by more versatile and enjoyable work.

-Third, technological change also generates additional jobs through the demand for new products and services and through higher competitiveness of firms that implement new technologies. A recent study suggests that computerization in Europe has on net generated positive labor demand effects since the millennium. While technology did substitute for labor, these labor-saving effects were exceeded by additional labor demand created through rising consumption and production. According to the study, labor has been racing with rather than against the machine in the recent decade.The economic outlook for the future of work might not be as pessimistic as many suggest. In particular, automation and digitization are unlikely to destroy large numbers of jobs. It seems more likely that workplaces will change and require different skills from workers compared to previous decades. Nevertheless, low qualified workers may face harder challenges to adjust in the digital transformation as the automatibility of their jobs is typically significantly higher compared to those of highly qualified workers.

This feature is strikingly consistent across all countries in the sample. From a policy perspective, the likely challenge for the future lies in counteracting rising inequality and ensuring that low qualified workers cope with the changing skill requirements. In emerging economies, however, technological change might have different effects, as these countries face a different technological endowment.

[1] We exclude the Russian Federation from our sample. This is because when we restrict the Russian PIAAC sample to those observations where all relevant variables are non-missing, then the distribution of these variables is not representative.