By Rodrigo Adão (Chicago Booth), Martin Beraja (UC Berkeley Haas), and Nitya Pandalai-Nayar (UT Austin)
As artificial intelligence reshapes the workplace, policymakers and business leaders are confronted with a pivotal question: how quickly will workers adapt to AI-driven changes? To understand what shapes these transition speeds, in a recent paper Adão et al. (2024), we examine two major technological revolutions of the past century: the manufacturing innovations of the early 1900s and the emergence of Information & Communications Technology (ICT) from the 1980s onward. We find that the skill requirements of new technologies critically determine how quickly economies adapt. The adjustment was faster for manufacturing because it required skills more similar to existing occupations, whereas the ICT revolution required different skill sets, resulting in a slower transition. While older workers found it challenging to transition into ICT-intensive occupations, younger generations gradually filled these roles.
Comparing Technological Shifts
Historically, innovations such as steam engines, electricity, and computers have led to fundamental shifts in economic activity. Looking to the future, artificial intelligence and robotics promise similar transformative impacts. When innovations favor specific skills, inequality may rise rapidly (Katz and Murphy, 1992). The reorganization of labor markets might take decades as workers shift roles and newer generations acquire necessary skills (Chari and Hopenhayn, 1991). But do economies adjust at different paces depending on technology? The literature mainly explains common patterns across adjustment episodes rather than examining how specific technologies shape them (Helpman, 1998; Herrendorf et al., 2014).
We present new insights into how exposure to technological innovations affected employment and wages across different occupations in the U.S. during two eras: the manufacturing-enhancing technologies of the early 20th century and the ICT innovations later that century. The contrast between these historical episodes is striking. Panel A of Figure 1 shows that the relative employment growth in more exposed occupations was both faster and stronger overall following the manufacturing innovations of the early 1900s compared to the ICT innovations of the late twentieth century. When assembly lines and electrical machinery transformed manufacturing in the early 20th century, workers rapidly shifted into manufacturing-intensive occupations. Panel B indicates a quick surge in demand for ICT-intensive occupations between 1980 and 1990, with a corresponding labor supply response after 2000.
Figure 1
Panels C and D of Figure 1 depict the employment response for older and younger workers, respectively. The ICT expansion post-1980 was almost solely driven by younger workers, as older workers did not shift to ICT roles—the black dots in Panel C show negligible and statistically insignificant changes for seasoned workers. However, the black dots in Panel D illustrate a significant increase in young workers entering ICT roles over forty years. Overall, the data indicate that the labor supply adjustment to ICT innovations was limited and delayed due to the minimal reallocation of older generation.
The Role of Skill Specificity
What explains these contrasting patterns? Our research shows that a key factor is how different the required skills are between old and new jobs – what we call “skill specificity.” When new technologies demand skills that differ from those used in existing jobs, the transition tends to be slower and more reliant on new generations entering the workforce.
Figure 2
To investigate this, we develop systematic measures of skill similarity between occupations. Figure 2 depicts two histograms, one for each episode, showing the task distance measure across occupations. The task distance distribution for ICT exposure in the latter episode has more mass on higher distance values than the distribution for manufacturing exposure in the earlier episode. For instance, the analytical and programming skills needed in software development had limited overlap with most existing jobs in 1980. This high skill specificity made it difficult for experienced workers to transition into ICT roles. In contrast, manufacturing jobs in the early 1900s required skills more similar to existing occupations. A craftsman could more readily adapt his manual and technical abilities to operate new industrial machinery. This lower skill specificity enabled the faster reallocation of incumbent workers.
Dynamics of Adjustment
The degree of skill specificity shapes transition dynamics through two key channels: First, when skills are more specific, experienced workers find it harder to switch to new occupations, even when wages are higher. This creates a direct barrier to workforce reallocation. Second, this limited mobility of incumbent workers leads to higher wage premiums in new occupations. These wage increases provide strong incentives for young workers entering the labor force to invest in new skills. This explains why technological transitions with high skill specificity are driven primarily by new generations. We develop a theoretical model that formalizes these mechanisms and demonstrates how they can quantitatively account for the different adjustment patterns observed in the manufacturing and ICT transitions.
Implications for the AI Revolution
Our findings have important implications for how economies might adapt to artificial intelligence. Early evidence suggests that some AI applications, like large language models, may be relatively easy to use with existing skills. As discussed in Acemoglu et al. (2023), AI could augment rather than replace current capabilities, suggesting a potentially faster adaptation than during the ICT revolution. However, more advanced AI applications may require highly specific skills with limited transferability from existing occupations. If this proves true, the transition could be slow, primarily driven by new workers, similar to the ICT experience. This scenario could exacerbate generational inequality and slow down aggregate productivity gains.
Conclusion
The lessons from past technological revolutions highlight that while the nature of jobs might evolve, the fundamental challenge of equipping workers with relevant skills remains constant. As we face potentially one of history’s most significant technological shifts with the rise of AI, a proactive approach will be essential to ensure all segments of the workforce can benefit. Our findings provide valuable guidance in anticipating and addressing these challenges, enabling a smoother transition into an AI-driven future.
References
Acemoglu, D., Autor, D., & Johnson, S. (2023), “How AI can become pro-worker”, VoxEU.org, 4 October.
Adão, R., Beraja, M., & Pandalai-Nayar, N. (2024). Fast and slow technological transitions. Journal of Political Economy Macroeconomics, 2(2), 183-227.
Chari, V. V. and Hopenhayn, H. (1991). Vintage human capital, growth, and the diffusion of new technology. Journal of Political Economy, 99(6):1142–1165.
Helpman, E. (1998). General purpose technologies and economic growth. MIT press.
Herrendorf, B., Rogerson, R., and Valentinyi, Á. (2014). Growth and structural transformation. Handbook of economic growth, 2, 855-941.
Katz, L. F. and Murphy, K. M. (1992). Changes in relative wages, 1963–1987: supply and demand factors. The Quarterly Journal of Economics, 107(1):35–78.
For more details, please see the associated working paper.
