Could AI Level the Playing Field of Earnings Inequality?

How we got here

For decades, technology has reshaped job markets throughout the world, often strengthening the relative position of the highly educated and skilled and those performing non-routine tasks. For example, in the US in the 1980s and 1990s, new technologies automated routine tasks such as data entry or working along an assembly line that were often done by less-educated workers. This caused the low skilled wages to grow more sluggishly than high skilled wages. At the same time, the new technologies complemented tasks of high-skill workers such as engineers who maintained robots or the managers of highly automated production firms. Thus, the wages of this group increased strongly during the same period. Consequently, earnings inequality in the U.S. grew rapidly in the 1980s and 1990s.

Also notable is the fact that the skill premium in the US leveled off in the 2000s. What could be the reasons for that? For sure there is more than one definitive explanation but in our most recent research paper we show that Artificial Intelligence (AI) might be playing an important role.

The AI Twist: Levelling the Playing Field

AI’s impact on jobs and wages could be quite different from earlier technological advancements. Unlike industrial robots or other types of machines that do routine tasks, AI is designed to take on more complex, non-routine tasks. This includes diagnosing diseases, developing new drugs, translating texts seamlessly between different languages, generating creative content, and writing computer code—all tasks traditionally reserved for highly-educated, highly-paid workers. Reducing the relative demand for high-skill workers, AI may exert downward pressure on skill differentials.

We study a theoretical model with traditional machines and assembly lines that are operated by humans, automation tools such as industrial robots and 3D printers for autonomous production, and AI capital that can do high-skill non-routine tasks. We pay particular attention to the subtle differences in the substitutability between the different types of capital and different skill levels of workers. We found that, as AI expands, it indeed reduces the skill premium, thereby appreciably narrowing the wage gap between high-skill and low-skill workers.

It is important to recognize that AI is just one part of the contemporary technology landscape. Industrial robots, for example, continue to increasingly perform routine tasks in manufacturing and other sectors. This means that while AI might reduce the skill premium directly, ongoing technological progress in other areas could still depress wages relatively for low-skill workers and, thus, obscure the direct effect of AI on the skill premium.

In addition to this cautionary note, our research does not focus on wealth inequality, which may increase with the use of AI because the cutting-edge AI models are developed by rather large technology companies with the financial muscle to design and train the AI models, a highly expensive endeavor. Thus, many of the benefits of AI may be channeled toward AI developers and the owners of the AI developing firms. This, in turn, could exacerbate wealth inequality, an issue that deserves more attention in future research.

Where to go from here?

While technology has historically contributed to rising wage inequality, the effect of AI could be different. Its ability to perform non-routine, high-skill tasks may shrink the wage gap between low- and high-skill workers. Yet, the full impact of AI on earnings inequality can be shaped by the way that industries and policymakers adapt to the new technology landscape.

Enacting wise regulations and rules for the use of AI and ensuring that as many people as possible are able to work with AI through education and (re-)training are important prerequisites for fostering a more inclusive and equitable job market, where everyone has the potential to benefit from the advances in technology.

References

Bloom, D E; K Prettner; J Saadaoui; and M Veruete (2024), “Artificial intelligence and the skill premium”, NBER Working Paper 32430.

Co-Authors

This blog article is co-authered by David E Bloom (Harvard T.H. Chan School of Public Health), Jamel Saadaoui (Université Paris VIII Vincennes-Saint-Denis) and Mario Veruete (Quantum DataLab).

CORA 2025

Klaus Prettner will be the host of the next CORA - Conference on Robots and Automation on 23/24 September 2025 in Vienna, Austria.

image: Pixabay