Ever since General Electric installed the first industrial robot in 1961, Americans have been worrying that automation could destroy the country’s labor force. During the Great Recession and its aftermath, these voices grew even louder. “We’re not going to have a jobless recovery,” business writer Jeff Jarvis predicted in 2011. “We’re going to have a jobless future.” “Smart machines won’t kill us all, but they’ll definitely take our jobs and sooner than you think,” Mother Jones warned in 2013.
But which jobs, exactly, are going to disappear? To hear many pundits tell it, the advance of technology is specifically threatening the middle ranks of the workforce. Automation, warned The Economist last October, will lead to “the further erosion of the middle class.” “Robots won’t destroy jobs, but they may destroy the middle class,” a Vox story was titled. The Associated Press produced a series of articles headlined, “What’s destroying the middle class? Not taxes. Not China. Think technology.”
Of the two political parties, Democrats have most loudly echoed the fears about automation and the middle class. Three major Democratic-aligned policy groups — the Third Way, the Center for American Progress, and the Hamilton Project — sponsored research in recent years arguing that the middle class was being harmed by automation.
But in the past year or two, some of the economists who study this issue have begun to veer in a different direction: They have produced research suggesting that automation is going to be relatively good to the middle class — creating new opportunities not only for professionals and managers but also for what MIT economist David Autor describes as “relatively well-remunerated, middle-skill” workers. These workers have attended some college or have bachelor’s degrees, and they make at or above the median wage. They are the heart of what Americans consider the middle class — and it now appears they aren’t disappearing at all.
This doesn’t mean automation is cost-free, however. According to this new line of thinking, the main victims of automation will not be middle-income workers but rather those with lower levels of income and education — that is, Americans who make less than $35,000 a year and have only a high school degree.
On the surface, the question of which demographic group is being most harmed by automation may sound like just a highly technical argument among economists. But, in fact, it sheds light on the utter inadequacy of U.S. politics and policy. While Democrats worry that the middle class is being threatened by robots, and Republicans woo the middle class with tax cuts, neither party is doing nearly enough to address the technology-driven disaster facing lower-income, less-educated Americans.
BROADLY SPEAKING, automation arrived in the U.S. economy in three overlapping waves. The first wave began in the 1960s with the introduction of the mainframe computer, the transistor, and the silicon chip. Firms started installing computer-coded robots and machine tools that took over much of the work of the factory floor — from the production of a physical good to its packaging for sale. The objective eventually became the creation of a “lights-out” factory, in which production would take place 24 hours a day without human supervision. While no American firm has yet achieved this for its entire operation, many now boast unlit, unheated, football-field-size areas where coded machines have taken over the tasks that humans used to perform.
The effects of this first wave on the economy were profound. From 1979 to 2013, U.S. manufacturing employment fell 37 percent, from 19.3 million to 12.1 million. Some of that was due to outsourcing — one study estimates that Chinese-import competition accounted for a quarter of the loss of manufacturing employment between 1990 and 2007 — but a great deal of the rest was due to the automation of basic, repetitive factory labor, the kind you would find on an automobile assembly line. Indeed, largely because of automation, American manufacturers today produce far more goods with far less labor than they did just a few decades ago. (Productivity in manufacturing increased 75 percentage points between 1987 and 2007.)
The second wave of the automation revolution came in the 1980s. This time, the key change was the introduction of the personal computer, which gradually began to replace lower-level office staff: secretaries, tax-preparers, typesetters, and file clerks.
You can see evidence of both waves at companies like Conveyers & Automation — a Towson, Maryland, firm that uses robots to create lights-out packaging facilities for Coca-Cola, Pepsi-Cola, Corning Glass, and other big manufacturers. When I visited recently, what was most striking was not only the way the large yellow robots picked up and put down cans and bottles, but also the fact that at the firm’s headquarters, I couldn’t spot a single clerical or blue-collar worker.
The third wave arrived in the early 1990s with the rise of the Internet. It transformed the distribution as well as the production of goods and services, creating what economist W. Brian Arthur calls a “secondary economy.” “Business processes that once took place among human beings are now being executed electronically,” Arthur writes. “They are taking place in an unseen domain that is strictly digital.” The secondary economy can make financial decisions, do inventory, diagnose illnesses, wage war, regulate electricity use, and sell everything from books to automobiles to machine tools.
Amazon began as the archetypical third-wave firm. It sold books and later other goods through the Internet, threatening the existence of bookstores and shopping malls. But it has also become a pioneer in fulfilling the promise of automation’s first wave. The company bought Kiva Systems, which produces orange robots the size of ottomans that roam unlit, unheated sections of Amazon’s warehouses all day and night, transporting shelves of goods to stations where they are packed and sent off. Amazon won’t discuss how much labor these devices have saved, but a manager of Amazon’s subsidiary Zappos has estimated that they cut labor in half. Moreover, the robots are supposed to speed up packaging by 400 percent.
In all these respects, automation has cut a wide swath through the economy. Like electricity, it’s a general-purpose technology; its effects are pervasive and not confined to a single industry. But while it has eliminated the jobs of clerical staff at firms like Conveyors & Automation, as well as the jobs of workers who used to scurry around Amazon warehouses grabbing packages off shelves, automation has not eliminated mid-skill, median-wage, middle-class positions.
That, at least, is the conclusion now drawn by MIT’s Autor and by two researchers at Oxford, Carl Benedikt Frey and Michael A. Osborne. In a presentation last summer to the Federal Reserve Bank of Kansas City, Autor, drawing on categories he had developed earlier with Harvard’s Frank Levy and Richard Murnane, set out the ways that automation has eliminated, added, modified, or left untouched various jobs. His categories (which I’ll use with minor variations) provide a useful way to see how automation has changed rather than destroyed the middle class.
First, there are the parts of the economy where automation has supplemented the human role — but not made it obsolete. These occupations fall into roughly three categories: complex abstract tasks (surgeons, dentists, lawyers, engineers, scientists, editors, architects, stock brokers, loan officers, therapists, school teachers, sales representatives, and a bevy of different kinds of technicians, especially in health care); jobs involving nonroutine personal interactions (specialized store clerks and technical-support personnel, home health aides, personal trainers, police, paramedics, and firefighters); plus those skilled crafts that cannot easily be reduced to routine instructions and now often require computer training (computer, utility, or telecommunications repair personnel, truck drivers, pilots, electricians, mechanics, and machinists).
Of course, some of these occupations may eventually fall victim to automation. Google, for instance, has developed a driverless car that could eventually be used to transport goods. But for now, none of these jobs are likely to be eliminated by technology.
So what jobs are in danger? Autor defines them as “routine tasks “¦ that follow an exhaustive set of rules and hence are readily amenable to computerization.” These include low-level clerical and secretarial work, rule-driven interpersonal encounters at banks, stores, and anywhere tickets are sold, and much blue-collar work in factories and warehouses.
In other words, the occupations that are safe from automation are all over the map in terms of income and education level — ranging from home health aides to telecom repair personnel to surgeons. But the jobs most threatened by automation are disproportionately unskilled and low-wage, requiring only a high school degree. In fact, nine of the 10 jobs that Autor and another MIT economist, David Dorn, cite as the most susceptible to computerization are low-skill and low-wage.
As recently as 2013, Autor had maintained that automation was polarizing the job market — creating a situation in which there was little middle zone between low-skilled service work and professional or managerial work. But last summer, in his presentation to the Federal Reserve Bank, he took a different tack. He now argues that “employment polarization will not continue indefinitely. While many middle-skill tasks are susceptible to automation, many middle-skill jobs” include tasks that are not. Many of these jobs, he explains, “will combine routine technical tasks with the set of nonroutine tasks in which workers hold comparative advantage — interpersonal interaction, flexibility, adaptability and problem-solving.”
Frey and Osborne reached a similar conclusion. In a 2013 paper in which they ranked occupations on the basis of whether they were computerizable, they concluded that the jobs “least susceptible” to computerization are high-skill professional and managerial jobs, but they also predicted that a host of middle-income occupations would survive. “Rather than reducing the demand for middle-income occupations,” they wrote, “our model predicts that computerization will mainly substitute for low-skill and low-wage jobs in the near future.”
Autor still maintains that, in the past, automation was destroying the middle class. But economist Stephen Rose argues that the middle class was never being destroyed, only gradually transformed. The difference lies partly in how they count the loss of many blue-collar production jobs. Autor defines these jobs as middle-income and middle-skilled — and sees their loss as evidence of the destruction of the middle class. Rose, on the other hand, sees these jobs as middle-income (because of fleeting corporate acquiescence to union wages) but lower-skilled. That changes the whole calculation of what has happened over the past few decades.
There is little disagreement among economists about how to help the victims of automation.
Rose has created an extensive database of occupations based not on Labor Department classifications, but on a job’s specific function, the skill required to perform it, and its place in the chain of output. According to his figures, from 1967 to 2007 — roughly the period during which the automation revolution took hold — the share of managerial and professional jobs in the labor force grew from 22 to 35 percent; the share of mid-skill jobs declined only slightly from 39 to 36 percent; and the share of low-skill jobs fell from 39 to 29 percent.
As Rose describes it, the trend has been gradually moving, with some stops and starts — for instance, during the last recession — toward an upgraded rather than a polarized workforce. It’s hard for me to say conclusively who is right, but Rose’s picture fits more with what I’ve seen as I’ve traveled around the United States over the years covering campaigns. The very wealthiest have become even wealthier; but below them, many Americans have gradually been moving upward. The middle class has not disappeared but edged ahead in education, income, and responsibility on the job. Americans’ standard of living has steadily risen — and that shows up in how they spend their money: According to Rose, Americans spent 46 percent of their income in 1947 on food, drink, and clothing. By 2007, they were only spending 18 percent. Much more of their money is now devoted to what John Maynard Keynes described as “relative” rather than “absolute” needs.
But amidst this improvement in Americans’ standard of living, there are also pockets of high unemployment and poverty throughout the country — in cities like Detroit or East St. Louis, in much of West Virginia or eastern Kentucky, and in rural parts of the Deep South. The people in these places have not benefited from the promise of automation, and if the economists are right about the rapid disappearance of low-skill jobs, things are only going to get much worse. The question, then, is: What should be done to help these people? And it’s a question that neither Democrats nor Republicans — both of whom would rather talk about the middle class than the poor — seem especially eager to face.
IN TRUTH, THERE is little disagreement among economists about how to help the victims of automation. “Our findings “¦ imply that as technology races ahead, low-skill workers will reallocate to tasks that are non-susceptible to computerization, that is, tasks requiring creative and social intelligence,” Frey and Osborne write. “For workers to win the race, however, they will have to acquire creative and social skills.” And the first place they must do that is in school.
As the automation revolution has developed, it has steadily raised workers’ educational requirements — and workers have responded by going to school more than they had. According to the census, in 1967, only about half of Americans over 25 had graduated from high school; by 2009, that number was 87 percent. In 1967, less than 10 percent over 25 had graduated from college; by 2009, it was 30 percent.
But that still leaves a lot of workers unprepared for today’s job market. Currently, 34 percent of the labor force over the age of 25 has not made it beyond high school. Many of these workers are falling through the cracks: The labor participation rate of high school graduates is 17 percentage points lower than the participation rate of four-year-college graduates.
The problem is, in part, the wide disparity in the quality of public schools. And the challenge is starkest in the states where high school and college graduation rates are the lowest — states such as Mississippi, Alabama, Louisiana, Kentucky, Arkansas, West Virginia, South Carolina, Tennessee, and Nevada. In these locales, many children don’t have a chance to move up the ladder of the new workforce that automation is creating.
What’s needed is obvious: an enormous, sustained, and probably expensive effort to increase the quality of the country’s worst schools. That would include money to attract outstanding teachers and to make it possible for lower-income children to go to college. (It would also involve revitalizing the communities around the schools.) Yet so far, Republicans and Democrats have mustered only half-measures — and even those half-measures have proved astoundingly difficult to implement.
In George W. Bush’s first year in office, for instance, Democrats and Republicans passed the No Child Left Behind Act, which imposed minimum standards on all states. Each state would administer standardized tests, and students would have to show progress for a school to receive federal funding. But the content of the tests was left up to the states, and some of the states that most needed to improve their school systems balked at doing so.
Then, in 2009, with the support of the Obama administration, 44 states and Washington, D.C., adopted common standards for math and reading; known as the Common Core, the benchmarks were designed to prepare students for college and the new economy. “These parsimonious standards give greater weight to today’s foundational skills than the weight given by most state standards,” Levy and Murnane wrote recently.
But the proposal has become a major source of controversy on the right, generating enormous opposition from Republican governors and from tea-party groups. Nineteen states — all with GOP governors at the time — have either rejected outright or downgraded the Common Core standards. These include many of the states — such as Mississippi, Louisiana, and South Carolina — for which the proposal was designed. While former Florida Gov. Jeb Bush remains supportive, two other Republican presidential aspirants who had previously backed Common Core — Louisiana Gov. Bobby Jindal and New Jersey Gov. Chris Christie — now either oppose or express “grave concerns” about the program.
President Obama and many Democratic policymakers continue to back the Common Core, but some Democratic politicians, moved in part by teachers’ wariness of standardized tests, have begun to distance themselves. Last fall, New York Gov. Andrew Cuomo claimed he had “nothing to do with Common Core” and promised to “disregard Common Core scores for at least five years.” In Washington state, the Democratic Party passed a resolution rejecting Common Core. Hillary Clinton has yet to take a clear position on the issue.
The Common Core standards certainly don’t represent a final answer to the question of what to do about the 34 percent of Americans who are going to suffer most from automation. These standards are merely a small starting point, but the fact that they are being rejected where they are most needed — and that some Democrats are joining the many Republicans who have backed away from them — says much about our political system’s inability to address the social challenge of automation.
One element of the problem may be Americans’ traditional aversion to government, but part of it may also be structural. In a parliamentary system with proportional representation, the 34 percent might have a party that would summon them to the polls and represent their distinct interests. By contrast, in America’s two-party, winner-take-all system, these voters are inevitably ignored in favor of larger, wealthier groups that are more likely to vote.
And so, Democrats bemoan the destruction of a middle class that actually remains quite vibrant, and Republicans cater to middle-class complaints about taxes and spending, while trying to lure poor whites with rhetoric about abortion and gun control. In the meantime, the revolution in automation will continue — and the 34 percent of Americans who are not prepared to join it will be lost in the shuffle.