In February, McKinsey Global Institute predicted that 45 million Americans—one-quarter of the workforce—would lose their jobs to automation by 2030. That was up from its 2017 estimate that 39 million would be automated out of work, due to the economic dislocation of COVID-19. Historically, firms tend to replace some of the workers they fire during recessions with machines.
Fear of robot-driven mass unemployment has become increasingly mainstream. Andrew Yang, who is currently leading the polls for the Democratic nomination to be the next mayor of New York City, made it a pillar of his unorthodox 2020 presidential campaign. The coming lack of jobs, Yang said, justified giving all Americans a $1,000 monthly government income.
But look closely at the studies predicting automation-driven job loss, and you’ll find less reason for alarm (though there are still reasons to consider a universal basic income). The robots are mostly not coming—at least, not soon.
To start, there’s a huge difference between “robots” and “automation.” Once, many elevators were operated by people. An office building in D.C. still had elevator operators in the 2010s. They weren’t replaced by humanoid robots that listen to rider requests and manipulate a lever with mechanical fingers. They were replaced by a row of buttons riders press themselves. A lot of automation works this way.
The distinction matters because automation happens all the time. Over the past 150 years, we’ve gone from a nation of farmers to a nation of factory workers to a nation of white collar and service employees, with much of that momentous change driven by automation. But while regional economies have been disrupted and recessions have created periodic unemployment crises, there has never been a chronic, structural shortage of jobs nationwide. New inventions create new markets and jobs to go with them.
The robot job apocalypse scenario is based on the assumption that the next wave of automation technology will be fundamentally different. Artificial intelligence in particular is thought to be advancing so quickly that replacement jobs won’t keep pace. People wonder whether our fragile, imperfect species will be necessary much longer.
But that’s not what the forecasters are saying. The robot job loss prediction boom was kicked into high gear in 2013, when a pair of Oxford University researchers estimated that 47 percent of American jobs are “at risk” of computerization. The report was widely cited, including in official White House reports.
To arrive at that estimate, a team of machine learning experts examined 70 occupations, each of which had been analyzed by the U.S. Department of Labor and broken down into dozens of discrete tasks and competencies. The experts looked at each task and made an informed guess as to whether it could be automated, assuming state-of-the-art technology, the enormous data sets that fuel modern A.I., and future engineering breakthroughs that have not yet occurred. They used those estimates to write an algorithm that automatically analyzed hundreds of other jobs.
“At risk” of automation does not, in that analysis, mean “likely to be automated.” It means, “could theoretically be automated if someone had unlimited time, money, and access to the latest A.I.” That’s an enormous difference. Perhaps the engineers at Boston Dynamics, which makes those viral videos of disturbingly humanoid robots, could spend millions of dollars building a robot version of the guy who stands on the street corner twirling the big pointy sign that says “Going Out of Business Sale!!!” But they won’t, because nobody would buy that robot, because they can just hire the guy for $10 an hour.
The recent McKinsey report takes this into account, estimating the cost of developing new automation technology, the price of the labor it would be replacing, and the time it would take for widespread adoption. That’s why its estimate is 27 percent of jobs, not 47 percent. But here, too, definitions matter.
McKinsey predicts that of the 49.1 million who will have their jobs displaced by automation, 32 million will stay in the same occupation, and another 2.2 million will stay in the same occupational category. The number of people who will lose their jobs in the “have to find a new line of work” sense is only 14.9 million. Not 27 percent, but 9 percent.
That’s because automation is more likely to change jobs than destroy them. Machines will perform an increasing share of boring, rote tasks, and people will move to more human work. When hundreds of thousands of ATMs were deployed in the 1980s and 1990s, the number of bank tellers went up, not down, because reduced labor costs allowed banks to open more branches. Now machines count the money, and people sell you auto loans. Automation works especially well when workers are partners in designing their new relationships to machines.
Nine percent of jobs is still a lot. But the optimal number isn’t zero. The White House automation report notes that about 6 percent of jobs in the American economy are eliminated every three months through the normal process of some businesses shrinking or shutting down as others start up and expand.
Automation-driven job loss definitely exists. In 2020, economists Daron Acemoglu and Pascual Restrepo found that each new industrial robot deployed in the United States between 1990 and 2007 replaced 3.3 workers, even after accounting for the positive economic effects of more productive firms. It was a small impact—one worker in 1,000—but very real.
The question of who gets replaced is also fraught. Nineteenth-century automation often replaced higher-paid skilled craftsmen. Twenty-first-century automation hits lower-paid, less-skilled workers the hardest. Recent recessions have been brutal for working-class families, who often never regain lost economic ground. The American unemployment insurance system is creaky, inadequate, and in dire need of reform.
The wilder robot dystopia scenarios often proceed from failures of metaphor. Many powerful new A.I. systems use methods called “neural networks,” which people assume means, “like a human brain.” They are not like human brains. A.I. is pattern recognition. Alexa knows that certain spoken sounds correspond to the sequence of letters “peanut butter,” which is remarkable. But Alexa has no idea what “peanut butter” means or why it tastes good with jelly.
The soberest predictions of automation job loss still rely on a lattice of interlocking predictions that may not come true. Five years ago, it seemed like we were on the cusp of robot taxis and freight trucks becoming widespread. Today, we’re stuck on the cusp. The last mile between “almost good enough” and “good enough” can be very long.
Even the simple, routine tasks that are the heart of most job loss scenarios can be fiendishly difficult to automate. Amazon uses hundreds of thousands of cutting-edge robots in its warehouses. But they’re not androids that pick items off of shelves. The robots are the shelves, which move to humans, who still do the picking.
Those simple, deft movements of eye and hand, recognizing and grasping myriad shapes in three dimensions, are the products of millions of years of evolution. Scientists and engineers are working hard to catch up. But they’re not going to fully solve those problems all at once, or in the next nine years, or for a long time after that.
Future Tense is a partnership of Slate, New America, and Arizona State University that examines emerging technologies, public policy, and society.
