Artificial Intelligence (AI) is evolving rapidly – from chatbots like ChatGPT and image generators like DALL·E to complex agent systems – and it is increasingly impacting high-skilled, non-routine jobs such as those in medicine, law, and finance. This shift is causing widespread concern about job losses. A Deloitte survey from 2023 found that 43% of workers fear losing their jobs to AI in the next five years, rising to 69% among highly qualified professionals. But how justified is this fear?

Historical Perspective: Tech Anxiety Isn’t New

Fear of job loss due to technology goes back to the 19th century, when “Luddites” protested against the replacement of their jobs by mechanized looms. Economic historians have identified three recurring concerns: (1) that new technologies eliminate jobs and increase short-term inequality; (2) that they dehumanize or eliminate meaningful work; and (3) that they may not lead to long-term economic or productivity gains. Such anxieties tend to intensify during uncertain economic times, as in 1811 when Britain faced severe economic hardships from crop failures and trade disruptions.

Despite such fears, history has also shown that technological shifts tend to transform rather than eliminate employment in the long term. While old jobs vanish, new ones emerge – especially in areas requiring creativity, social intelligence, or adaptability. Yet, public debate tends to highlight risks over potential gains, as people are psychologically inclined to perceive uncertainty as threatening.

Rethinking Automation Risks

The global debate around AI and automation was reignited by a 2013 study by Carl Benedikt Frey and Michael Osborne, who argued that AI would differ from previous technologies. They estimated that 47% of U.S. jobs would be at risk of automation over the next decade or two (Frey & Osborne, 2013). Though widely cited, the study has been criticized for assuming that entire occupations can be fully automated by AI, without considering that jobs often involve a variety of tasks, some of which are less susceptible to automation.

In our critique of Frey and Osborne, we examine job automation from a different angle by re-assessing the potential of AI and related technologies using a task-based, rather than occupation-based, approach (Arntz et al., 2017). We show that many tasks – especially those involving interaction, complexity, or creativity – are far less automatable than Frey and Osborne assumed. As a result, we estimate that only about one in ten jobs are automatable – a figure substantially lower than earlier projections. The findings suggest that automation potentials tend to be overestimated and should not be conflated with actual job losses.

Why AI May Not Lead to Mass Unemployment

A large misperception in the debate on “automation risks” is that automation potentials must not translate into employment losses. First, our research also suggests that new technologies tend to be adopted slowly, as firms face high costs, organizational challenges, and regulatory hurdles (Arntz et al., 2024). Second, workers have historically shown adaptability by reskilling and shifting into new roles. Third, innovation itself generates employment, with research showing that 60% of jobs created in the US between 1980 and 2015 emerged in entirely new occupations (Autor et al., 2024).

Whether AI ultimately increases or reduces employment depends on the net effect of the following forces: the displacement of human labor by machines, productivity-driven job creation through increased demand, and the reassignment of workers to new, more productive roles. Focusing on automation technologies more generally, we found that despite some displacement, automation has contributed to net employment growth across Europe (Gregory et al., 2022). Importantly, we show that many new jobs emerged outside the automating sectors. Rising incomes from productivity gains have further fueled consumption and created employment in service industries like retail, hospitality, and health.

The Missing Productivity Boost

Similar estimates on the employment effect of actual AI adoption are still lacking due to insufficient data on the micro level. Whether a positive net effect will occur depends critically on whether AI delivers the expected productivity gains. Despite AI’s rapid development, many industrialized nations have not yet experienced the productivity gains it promised – this is evident from the absence of noticeable increases in aggregate productivity statistics, a phenomenon often referred to as the Productivity Paradox. Optimists argue that we are in a transition phase – the “Between Times” – where the real productivity benefits will emerge only after companies fully integrate AI into redesigned workflows, rather than simply use it to optimize existing ones (Agrawal et al., 2022).

Currently, most firms apply AI to improve efficiency, for example, through automated document processing or fraud detection. Real transformation happens when AI reshapes entire business models. One example is in German retail, where inventory and sales forecasting decisions are being centralized and guided by AI-driven analysis of weather, location, and historical data. This represents a system-wide change with the potential for large-scale scalability and growth – alongside a redistribution of decision-making power.

AI as a Tool for Human Enhancement

There are also reasons for caution that technology does not inherently benefit everyone. Nobel laureates Daron Acemoglu and Simon Johnson argue in their analysis of 1,000 years of innovation that outcomes depend on who controls the technology and how it is used (Acemoglu & Johnson, 2023). If AI is deployed mainly for cost-cutting, such as through customer service chatbots or aggressive automation, job losses and service degradation are likely. However, if AI augments human work, like diagnostic tools in healthcare or adaptive learning platforms in education, it can enhance productivity and quality without eliminating human roles.

This speaks for a human-centered approach to AI where machines complement, not replace, people. This strategy prioritizes meaningful jobs, high-quality services, and broad social benefits over short-term profit and labor reduction.

The Role of Institutions in Shaping Outcomes

Noteworthy is the call by the Nobel laureates for strong institutions to ensure that technological progress leads to shared prosperity. They show that inclusive innovation occurs when political and social institutions, such as worker representation, collective bargaining, and continuing education, are actively involved in shaping change.

Germany can be seen as a positive example, with strong co-determination rights, training systems, and active unions that help workers navigate transitions. In contrast, the U.S. – where worker representation has weakened considerably – shows more negative employment effects from automation. There is a need for further research to determine which institutional arrangements are most effective today in guiding technological change for the public good.

Conclusion: The Future of Work Is in Our Hands

AI is poised to transform the labor market, but whether this transformation leads to opportunity or precarity depends on the choices we make. Experience suggests that inclusive, well-regulated innovation, backed by strong institutions, can lead to job growth and shared prosperity. The critical question is not whether to adopt AI, but how it is used and whose interests it serves. The power to shape that future does not lie in the algorithms – it lies in human hands.

Bibliography

Acemoglu, D., & Johnson, S. (2023). Power and progress: Our thousand-year struggle over technology and prosperity. Basic Books.

Agrawal, A., Gans, J., & Goldfarb, A. (2022). Power and prediction: The disruptive economics of artificial intelligence. Harvard Business Review Press.

Arntz, M., Genz, S., Gregory, T., Lehmer, F., & Zierahn-Weilage, U. (2024). Frontier technology adopters and the aggregate decline of routine jobs (IZA Discussion Paper No. 16740). https://docs.iza.org/dp16740.pdf

Arntz, M., Gregory, T., & Zierahn, U. (2017). Revisiting the risk of automation. Economics Letters, 159, 157–160. https://doi.org/10.1016/j.econlet.2017.07.001

Autor, D., Chin, C., Salomons, A., & Seegmiller, B. (2024). New frontiers: The origins and content of new work, 1940–2018. The Quarterly Journal of Economics, 139(3), 1399-1465. https://doi.org/10.1093/qje/qjae004

Frey, C. B., & Osborne, M. A. (2013). The future of employment: How susceptible are jobs to computerization?. In Oxford Martin Programme on Technology and Employment. https://oms-www.files.svdcdn.com/production/downloads/academic/future-of-employment.pdf

Gregory, T., Salomons, A., & Zierahn, U. (2022). Racing with or against the machine? Evidence from Europe. Journal of the European Economic Association, 20(2), 869–906. https://doi.org/10.1093/jeea/jvab012