By sifting through 35,000 job categories in U.S. census data, economists have developed a new method for estimating the impact of technology on employment growth and loss.
Since the Luddites were destroying machine looms, it has been evident that new technologies can wipe out jobs. However, technical innovations also create new jobs: Consider a computer programmer or someone installing solar panels on a roof.
Overall, does technology replace more jobs than it creates? What is the net balance between these two things? Until now, that has not been measured. However, a new research project led by MIT economist David Autor has developed an answer, at least for U.S. history since 1940.
The study uses new methods to examine how many jobs have been lost to machine automation and how many have been generated through “augmentation,” in which technology creates new tasks. The study finds that, particularly since 1980, technology has replaced more U.S. jobs than it has generated.
“There does appear to be a faster rate of automation, and a slower rate of augmentation, in the last four decades, from 1980 to the present, than in the four decades prior,” says Autor, co-author of a newly published paper detailing the results.
However, that finding is only one of the study’s advances. The researchers have also developed an entirely new method for studying the issue, based on an analysis of tens of thousands of U.S. census job categories about a comprehensive look at the text of U.S. patents over the last century. That has allowed them, for the first time, to quantify the effects of technology on both job loss and job creation.
Previously, scholars had largely been able to quantify job losses produced by new technologies only, not job gains.
“I feel like a paleontologist who was looking for dinosaur bones that we thought must have existed but had not been able to find until now,” Autor says. “This research breaks ground on things we suspected were true, but we did not have direct proof of them before this study.”
The paper “New Frontiers: The Origins and Content of New York, 1940-2018” appears in the Quarterly Journal of Economics. The co-authors are Autor, the Ford Professor of Economics; Caroline Chin, a PhD student in economics at MIT; Anna Salomons, a professor in the School of Economics at Utrecht University; and Bryan Seegmiller SM ’20, PhD ’22, an assistant professor at the Kellogg School of Northwestern University.
Automation versus Augmentation
The study finds that overall, about 60 percent of jobs in the U.S. represent new types of work created since 1940. That computer programmer may have been working on a farm a century ago.
To determine this, Autor and his colleagues combed through about 35,000 job categories in the U.S. Census Bureau reports, tracking how they emerge over time. They also used natural language processing tools to analyze the text of every U.S. patent filed since 1920. The research examined how words were “embedded” in the census and patent documents to unearth related text passages. That allowed them to determine links between new technologies and their effects on employment.
“You can think of automation as a machine that takes a job’s inputs and does it for the worker,” Autor explains. “We think of augmentation as a technology that increases the variety of things that people can do, the quality of things people can do, or their productivity.”
From about 1940 through 1980, jobs like elevator operator and typesetter tended to get automated. But at the same time, more workers filled roles such as shipping and receiving clerks, buyers and department heads, and civil and aeronautical engineers, where technology created a need for more employees.
From 1980 through 2018, automation has thinned the ranks of cabinetmakers and machinists, among others, while industrial engineers, operations, and systems researchers and analysts have enjoyed growth.
Ultimately, the research suggests that the adverse effects of automation on employment were more than twice as significant in the 1980-2018 period as in the 1940-1980 period. There was a more modest and positive change in the effect of augmentation on employment in 1980-2018, as compared to 1940-1980.
“There’s no law these things have to be one-for-one balanced, although there’s been no period where we haven’t also created new work,” Autor observes.
What will AI do?
However, the research must also cover more nuances in this process since automation and augmentation often occur within the same industries. It is not just that technology decimates the ranks of farmers while creating air traffic controllers. For example, there may be fewer machinists but more systems analysts within the same large manufacturing firm.
Relatedly, over the last 40 years, technological trends have exacerbated a wage gap in the U.S., with highly educated professionals being more likely to work in new fields, split between high-paying and lower-income jobs.
“The new work is bifurcated,” Autor says. “As old work has been erased in the middle, new work has grown on either side.”
The research also shows that technology is not the only driving new work. Demographic shifts also lie behind growth in numerous sectors of the service industries. Intriguingly, the latest research suggests that large-scale consumer demand drives technological innovation. Inventions are not just supplied by bright people thinking outside the box but in response to precise societal needs.
The 80 years of data also suggest that future pathways for innovation and their employment implications are hard to forecast. Consider the possible uses of AI in workplaces.
“AI is different,” Autor says. It may substitute some high-skill expertise but may complement decision-making tasks. We’re in an era where we have this new tool and don’t know what it’s good for. New technologies have strengths and weaknesses, and it takes a while to figure them out. GPS was invented for military purposes, and it took decades to be in smartphones.”
He adds: “We’re hoping our research approach gives us the ability to say more about that going forward.”
Autor recognizes that the research team’s methods can be further refined. For now, he believes the research opens up new ground for study.
“The missing link was documenting and quantifying how much technology augments people’s jobs,” Autor says. “All the prior measures showed automation and its effects on displacing workers. We were amazed we could identify, classify, and quantify augmentation. So that itself, to me, is pretty foundational.”
The Carnegie Corporation, Google, Instituut Gak, the MIT Work of the Future Task Force, Schmidt Futures, the Smith Richardson Foundation, and the Washington Center for Equitable Growth provided part of the research support.