The concern about an artificial intelligence, or AI, workforce shortage in the United States is rapidly becoming a top national security priority. Calls for additional legislative action are mounting as the national security community sees talent as a key enabler in outcompeting China. An increasing number of proposals, including those in the 2021 National Defense Authorization Act and others based on the recommendations of the National Security Commission on Artificial Intelligence, have the goal of growing and cultivating the domestic AI workforce based on the premise of shortages.
However, there is little data on actual U.S. AI labor market dynamics to inform whether there is an AI workforce shortage, and if so, what type and to what extent. Moreover, there is no standard definition of “AI workforce.” This makes it difficult, if not impossible, to determine which workers are in short supply and how to best address it.
Workforce shortages generally come in two distinct forms, which is important for targeting policy. The first is a skills shortage in the traditional economic sense: an insufficient supply of talent with specific, in-demand skills often due to high barriers to entry. For example, a critical AI occupation often discussed synonymously with “AI workforce,” computer research scientists, requires years of advanced education and training. Domestic supply is limited, and as a result, the United States also relies on foreign talent.
The second type is a local talent shortage, in an organization or geographic area. There are many possible reasons for local shortages — workers’ geographic preferences, state occupational licensing requirements and, at the organizational level, limited career advancement opportunities, poor hiring practices, uniquely specific skills and/or experience requirements, below market average wages and compensation, among others.
New CSET research provides insight into the reality of U.S. AI talent shortages. We assess the state of the U.S. AI workforce, analyzing traditional economic indicators like employment and wages. We use an occupation-based definition of the AI workforce, enabling analysis using occupational data collected by the federal government over time. We also consider the entire team of talent involved with the design, development and implementation of AI.
To assess the extent of shortages, we focus on five key occupations: Computer and information research scientists; software developers; mathematicians, statisticians and data scientists; user experience designers and project management specialists.
Although it is difficult to make an evidence-based determination, we find three variations in workforce dynamics across the five occupations. First, extremely strong employment and wage growth for computer and information research scientists over 2015 to 2019, coupled with high barriers to entry, likely indicates there is more demand than supply. Second, in other high-demand occupations such as software developers and data scientists, our analysis suggests existing education pipelines have responded as designed to meet rising demand. Over 2015 to 2018, for example, computer science and engineering were the fastest growing undergraduate majors, adding more than 200,000 new graduates. Third, in the case of project management specialists and user experience designers, we do not find evidence of a gap in demand relative to supply.
The divergent trends within the AI workforce have implications for future policy. Importantly, we need to prioritize growing, cultivating, and attracting the highest tier of AI talent. Previous CSET research recommends increasing graduate education scholarships and reforming and streamlining immigration pathways for top AI talent entering the U.S. This also involves a better understanding of STEM workforce pipeline leakages domestically, which start early and continue throughout the entire education lifecycle and well into one’s career.
For technical occupations where talent pipelines are working, we need to ensure their sustainability; for example, by incentivizing AI companies to closely partner with state workforce boards and universities to help schools prepare students with the necessary skills. For nontechnical AI occupations, the diversity of potential backgrounds suggests prioritizing AI literacy education for everyone.
Finally, in the quest to expand the talent pipeline, the prevailing wisdom of the need for a four-year college degree should be carefully considered. The current proliferation of certifications, coding academies and other online courses aimed at upskilling U.S. workers provide an alternate pathway for some. However, it is unclear if most employers are accepting these alternative credentials. This creates a risk of leaving many potentially qualified workers on the sidelines, if non-college graduates are less able to compete.
Effective education and workforce policy requires understanding the relevant labor market dynamics, including insights on the existence of workforce shortages. Our research suggests that policies aimed at equipping America’s talent for tomorrow’s jobs cannot and should not be a one-size-fits-all approach.
Diana Gehlhaus is a research fellow at the Center for Security and Emerging Technology (CSET), focused on AI education and workforce issues as it relates to the United States, China, and U.S. Department of Defense. Follow her on Twitter @dianagcarew.