Labor market impacts of digitalization and automation

The past decades have been characterized by a tremendous rise in computing power, reducing the costs of automating so-called routine tasks which follow clear, explicit rules and can thus be put into computer code. This has led to a polarization of labor markets in advanced economies with declining shares of middle-paid, routine-intensive occupations and rising shares of both, high- and low-paid jobs.

While this computerization has not led to employment declines, the question whether this holds true for the effects of further technological advances in the near future remains open. Whereas previous automation methods were limited to problems that are sufficiently well understood to be put into algorithms of well-defined steps, now even less structured problems appear automatable using big data and machine learning.

Continued increases in computing power, the growing availability of big data, and significant advances in Machine Learning methods are shifting the boundaries of what can be automated by machines. Against this background, some studies predict that about half of the U.S. workforce is “at risk of automation”, which has spurred public fears of technology-induced mass unemployment.

Workers adjust to automation

A new IZA/ZEW paper by Melanie Arntz, Terry Gregory and Ulrich Zierahn contrasts such fears with the scientific debate. The study shows that many estimates of automation potentials are severely upward biased as they often are conducted at the occupational level, ignoring the great variation in what people actually do at work.

As many workers in seemingly automatable occupations already adjust their task schedules to non-automatable tasks, they often face much lower exposure to automation. The study finds that the share of workers in automatable jobs is likely to be less than 10% in the U.S. and many other countries. These numbers, however, only refer to technological potentials and must not be equated with actual job losses or employment effects as is often done in the public debate.

The study outlines three main reasons why the job destruction potential of automation is overrated:

1. The diffusion of new technologies into the economy is a rather slow process, leaving workers time to adjust. Diffusion is slow due to high costs, uncertainty, the need to undergo organizational change for implementing the technologies, and the need for acquiring workers with suitable skills.
2. Workers are flexible and adjust. In fact, much of the adjustment to automation is not made by making seemingly replaceable occupations redundant, but by workers doing other tasks in the same occupations. Being in an occupation that is “at risk” thus does not necessarily imply that the worker is about to lose his or her job, but that the worker has to adapt by switching to the right tasks and learning the right skills.
3. While automation indeed does displace jobs, it simultaneously creates new jobs. The overall effect on the employment has actually been positive, not negative. It thus remains an open question whether the next wave of digitalization and automation will lead to fewer or even more jobs.

The paper also describes scenarios for the potential impact of digitalization and automation via cutting-edge technologies on the German labor market, exploiting a recent survey on the adoption of new digital technologies.

The results suggest that the net effect remains small, and is actually positive in the next five years. However, there appear large structural shifts between occupations and industries, which are accompanied by rising inequality and employment polarization.

Coping with structural change

The main challenge for the future thus is not mass unemployment, but structural change. In addition, the simulations suggest that firms are currently in an investment phase where they first have to incur high investment costs and need to acquire the right skilled workers before being able to reap large productivity gains.

Therefore, the effects of these cutting-edge technologies may change in the medium to long run when the technologies mature. Nonetheless, this does not imply that they will reduce employment in the longer run. Once they mature, productivity effects will also raise the demand for labor. It remains to be seen whether the job-creating effects continue to dominate the job-destruction effects in the longer run.

These results entail three main policy implications:

1. Promoting the adoption of new technologies seems to be a reasonable policy goal, as these technologies apparently raise employment and production. The focus should be on medium and small firms who currently seem to fall behind.
2. The introduction of these technologies requires workers with the right skills. The lack of such workers seems to partly hinder the introduction of new technologies. The second recommendation thus is to address skill shortages by education, qualification, and further training.
3. The coming wave of technological change seems to be associated with a further rise in inequality, as high-skilled, high-wage occupations are on the rise, whereas low- and medium-paid jobs further fall behind. In order to prevent further rising inequality, targeted training and qualification measures may help workers to switch to the expanding occupations, thus helping them to participate in the technology-induced benefits while lowering the losses of those who cannot change their skills and jobs and thus remain in shrinking occupations and sectors.