Who's Susceptible To The Automation Boom?

By Lucas Kitzmueller & Bhavik Nagda | May, 2021

The last few decades have witnessed rapid advances in automation technology amidst growing anxiety about the labor market. As emerging technologies mature and gain commercial traction, concerns of jobs displacement will only heighten. New technologies create winners and losers. They improve employment opportunities for some workers but lower the demand for others. Scroll down to learn more about the impacts of automation on certain economies, geographies, and demographics.

A Brief History of Automation Risk

In the last 30 years, it was primarily workers in middle-skilled occupations in production and administration who found themselves on the wrong side of the demand curve. These jobs have in common that they require challenging, yet at the end of the day, repetitive routine tasks that can be automated.
For example, industrial robots reduced the need for factory workers and computers have reduced the demand for secretaries.
While employment in middle-skill occupations was declining, high-skill occupations such as managers and professionals were growing . These occupations rely on cognitive non-routine tasks, which to date have been proven hard to automate.
Similarly, low-skill occupations with a large share of manual routine tasks have been growing. This includes, for example, workers in personal care services.

The Data


However, as the commercial adoption of AI is still nascent in many industries, it is still unclear what workers will be affected by AI. While there exist many studies on the labor market impact of robot adoption, there are no comparable studies for AI technologies. Fortunately, the Stanford economist Michael Webb has developed a clever way to measure future exposure to AI technologies. His idea is the following: The text of AI-based patents contains information about what technologies do, and the text of job descriptions compiled by the Department of Labor contains information about the tasks people do in their jobs. Combining these two datasets, we can quantify what jobs AI may soon be capable of performing at the workplace. For example, one of the tasks of doctors is “Interpret tests to diagnose patient’s condition”. From these task descriptions, Webb extracts verb-object pairs. In this case, the pairs would be (interpret, test) and (diagnose, condition). To then measure this task’s exposure to AI, Webb calculates the frequency of these verb-object pairs in the titles of all AI-based patents. With this method, he is able to rank 964 occupations in the database according to their exposure to AI technology. To contrast the impact of AI with past technological shocks, Webb also calculates the exposure to robots and software technologies.

Michael Webb's Process

Does Education Affect Automation Risk?

For this analysis, we grouped occupations by their typical entry level education as determined by the Bureau of Labor Statistics. The graph shows distribution of Webb’s exposure estimates within each educational group. Occupations are weighted by size (i.e., the number of workers employed in the occupation in 2019).
The data for robot exposure confirm the earlier finding: less-educated workers, typically employed in jobs with a high share of manual routine tasks, are more exposed to disruption from robotics than higher-educated workers.
The exposure estimates for AI, however, reveal an opposite pattern. Some lower-skill occupations, such as power plant operators and dispatchers, are still highly exposed. But many highly educated workers will be impacted too, such as clinical laboratory technicians, chemical engineers, and optometrists. In fact, workers with a bachelor’s degree appear to be exposed the most.

The Geographical Distribution of Automation Risk

Risk Percentile for Robot Exposure

Worker displacement represents a challenge for policymakers, especially if it is concentrated in few labor markets. The map shows the exposure of states to different technologies, based on the occupational composition of their workforce as calculated from the Current Population Survey.
As expected, states involved in manufacturing, such as Wisconsin or Kentucky, are heavily impacted by robotics.
However, as AI is increasingly linked to production, the states involved in manufacturing appear to be also disproportionately exposed to AI. At the same time, states with a large high-tech sector and managerial workforce, e.g., Washington DC, Boston, and Washington state, also have high exposure to AI technologies.
These findings suggest that this time will be different: while past automation shocks primarily affected lower and middle-skilled workers, AI is set to displace higher-skilled workers. And while increases in robots primarily impact regions intensive in manufacturing, the impact of AI will be felt more widely across the US.
Sources: Data from Michael Webb's Paper and the St. Louis Fed

Thank you to the 6.859 course staff for a wonderful semester!