FIGURE 19: The Trajectories of Gross Domestic Product (GDP) per Hour and Average Weekly Leisure in the U.S, 1890–2000
Sources: V. A. Ramey and N. Francis, 2009, “A Century of Work and Leisure,” American Economic Journal: Macroeconomics 1 (2): 189–224. See figure 9 for data on GDP per hour.
To be sure, Keynes did not overestimate the potential scope of mechanization. He was broadly right in thinking that “we may be able to perform all the operations of agriculture, mining, and manufacture with a quarter of the human effort to which we have been accustomed.”97 The economist Robert Heilbroner observed thirty-six years later, in the midst of the 1960s automation debates:
One can maintain that the labor-displacing effects have run ahead of the job-creating effects within the two historically most important areas of work—the farm and the factory.… In mining, we see the same absolute shrinkage as in agriculture, in spite of a huge rise in output (as in agriculture). Eight hundred thousand men entered the earth or worked at the mine surface in 1900; only 600,000 in 1965.… Thus it seems beyond dispute that the labor-displacing effect of investment can outpace the job-creating effect, and in fact has done so in many key sectors of the economy.98
Heilbroner was, of course, well aware that while workers had been replaced from agriculture and mining, they had not become detached from the labor market altogether. On the contrary, the percentage of the population engaged in paid employment had risen as more and more women entered the workforce. As home production became increasingly mechanized, women could have decided to take full advantage of the new entertainment possibilities provided by phonographs, radios, and televisions and enjoyed their newly found leisure time at home. But instead they decided to move into the labor market to take on paid work. More broadly, an average American worker in 2015 who merely wishes to maintain the average income level of 1915 could do so by working just seventeen weeks per year, aided by modern technology.99 But most citizens do not find this trade-off desirable. Instead, their demand for new goods and services has risen along with productivity. As labor-saving technology has given us the means to do more with less, most of us have preferred to take on other productive tasks instead of opting for more leisure.
The concern going forward, Heilbroner argued, was twofold. This time, it was not just jobs in agriculture and industry that were being affected. Automation, he worried, was rendering employment in the service sector redundant, too. And the demand for services produced by labor, he predicted, would eventually be fully met:
But here is the critical point: Technology today seems to be invading the services as well as other kinds of work. One secretary, with a machine, can now type and style a writer’s work.… There is no reason why technology should not penetrate deep into the job skills of the white-collar class. Where, then, will the new labor emigrants go? … Suppose that we can employ most of the population as psychiatrists, artists, or whatever. I am afraid there is still an upper limit on employment due, very simply, to the prospect of a ceiling on the total demand that can be generated for marketable goods and services.100
Whether there is a saturation point or not is controversial, but if our “basic needs” have all been met, higher incomes should no longer translate into higher subjective well-being. To that end, the economists Betsey Stevenson and Justin Wolfers examined whether there’s a critical income level beyond which the well-being–income relationship diminishes. Analyzing multiple data sets, and using various definitions of basic needs and different measures of well-being, they found no evidence of a satiation point so far. When they compared average levels of subjective well-being and gross domestic product (GDP) per capita across countries, they found that the well-being–income relationship observed among poor countries holds equally in the rich world. This relationship also holds across income levels within countries. In America, for example, there is no evidence of a significant break in the well-being–income relationship, even at annual incomes of half a million dollars.101 Thus, if there’s a satiation point, it is yet to be reached.
In 1966, Herbert Simon wrote a response to Heilbroner’s article, arguing that “insofar as they are economic problems at all, the world’s problems in this generation and the next are problems of scarcity, not of intolerable abundance.”102 It is tempting to conclude that Simon was right and nothing much has changed since. The dystopian belief that automation must lead to unemployment, and the utopian view that it will bring about a life of leisure, have both seemingly been wrong so far. Looking forward, one observer aptly summarizes the matter as follows:
That shorter hours are inevitable with automation and the kind of technological progress we envisage reflects, in many instances, a fear that unemployment will spread if work is not shared. Just as commonly, however, it reflects a utopianism that the new technology heralds a new day of all play and no work. Whether workers prefer shorter hours to additional income depends upon their judgment as to the relative worth of leisure and income. Progressive gains in productivity and levels of living make this choice easier to make in favor of leisure, but the outcome is hardly predictable. It becomes more and more uncertain as hours input and arduousness of work fall below a point where the physical strain and other detriments of work impinge on health, family life, and full participation in social life, while at the same time the material standards of consumption rise. I do not know what workweek industrial and other workers will choose in the future. It is interesting to note that with substantially full employment in recent years, little reduction in average hours worked has occurred in non-agricultural employment in the United States.… At present, workers seem inclined generally to place a higher value on additional income than on more leisure, but this may not always be the case.103
The extract above is from a Bureau of Labor Statistics report that was first published in 1956. I could have used those same words today. After a century of staggering advances in mechanization and soaring productivity, it is quite remarkable how little time Americans take in leisure.
* * *
One relatively recent trend is noteworthy, however. Historically, the working poor were the ones who had to put in longer hours to be able to support themselves and their families. As Hans-Joachim Voth has shown, average working hours in Britain increased from fifty per week in 1760 to sixty hours in 1800.104 This was the time of Engels’s pause, when material standards for the working class faltered. It was also around the time when Jane Austen’s novels about the elites depicted a society at leisure, whose lives centered on refined conversations and literature. But in recent years, those who would have flocked into the factories in the postwar period have become less likely to work, also relative to the new cognitive elite. The economists Mark Aguiar and Erik Hurst find that the least educated increasingly “enjoy” more leisure hours than symbolic analysts. Data from the American Time Use Surveys also show that college-educated Americans now work about two hours more per day than people who did not go to college.105 The most convincing explanation for this pattern is simply that it reflects diminishing opportunity in the labor market for the would-be working class. As we saw in chapter 9, as automation has progressed, opportunity for the unskilled has regressed. Faced with falling wages and fading job options, some people may have chosen welfare instead of work, while others have struggled to find a job.
In 1983, when computers finally began to enter the workplace in large numbers, Wassily Leontief remarked: “Think what would happen if all unemployed steel and auto workers were retrained to operate computers.… There aren’t enough computers to go around.… More and more workers will be replaced by machines. I do not see that the new industries can employ everybody who wants a job.”106 One reason why there are still so many jobs is that computers did create new tasks for labor (see chapter 9). But those tasks were primarily created for the highly skilled. This stands in contrast to the period of the Second Industrial Revolution, when technological change spun off new tasks for semiskilled workers, leaving the middle cla
ss with more and increasingly well-paying jobs (see chapter 8). The industries of the computer age failed to provide the same opportunities for the middle class as the smokestack industries that preceded them.
It is hard—not to say impossible—to predict exactly what new jobs and tasks AI technologies will bring about in the future. Yet we should put some faith in the observations of Frederic Bastiat. In his brilliant 1850 essay, “That Which Is Seen, and That Which Is Not Seen,” he wrote: “In the department of economy, an act, a habit, an institution, a law, gives birth not only to an effect, but to a series of effects. Of these effects, the first only is immediate; it manifests itself simultaneously with its cause—it is seen. The others unfold in succession—they are not seen: it is well for us, if they are foreseen.”107 With regard to machines, substitution is the observable first-order effect. What isn’t seen are the new jobs that will be created. Few jobs that exist in America today existed in 1750, at the onset of the Industrial Revolution. And many of today’s jobs did not exist in official occupational classifications even in the 1970s, including those of robot engineers, database administrators, and computer support specialists. Almost half of employment growth between 1980 and the Great Recession happened in new types of work.108
The unseen will always be unknown, but it seems unlikely that AI technologies will reverse the twentieth-century pattern of rising skill requirements. With some exceptions, the jobs that are least likely to be replaced in the next wave are indeed the jobs of the skilled. And if we look to new industries that did not exist in 2000, most of these industries are related to digital technologies, and the majority of the workers they employ have a college degree (many a degree in science, technology, engineering, or mathematics).109 Thus, the next wave of automation is likely to have effects similar to those of earlier computer technologies, but it is likely to affect more people. Those who would have taken on jobs in the factories in the postwar era have already seen their job options diminish since the computer revolution. And with retail, construction, transport, and logistics becoming more exposed to automation, too, the options of those people will likely deteriorate even further. Indeed, even if the next three decades mirror the past three, that is not all that comforting since automation recently has pushed up joblessness among groups in the labor market and put downward pressure on the wages of those with no more than a high school degree. In their best-selling book The Second Machine Age, Erik Brynjolfsson and Andrew McAfee make a similar observation: “Technological progress is going to leave behind some people, perhaps even a lot of people, as it races ahead.… There’s never been a better time to be a worker with special skills or the right education, because these people can use technology to create and capture value. However, there’s never been a worse time to be a worker with only ‘ordinary’ skills and abilities to offer, because computers, robots, and other digital technologies are acquiring these skills and abilities at an extraordinary rate.”110
FIGURE 20: Most Common Occupation by U.S. State in 2016
Source: S. Ruggles et al., 2018, IPUMS USA, version 8.0 (dataset), https://usa.ipums.org/usa/.
Today, the largest single occupation in most American states is that of the truck driver (figure 20). It is true, as the economist Austan Goolsbee has pointed out, that if all 3.5 million truck, bus, and taxi drivers lose their jobs to autonomous vehicles over a fifteen-year period, that would amount to just over nineteen thousand per month: in 2017, 5.1 million Americans were separated from their jobs on a monthly basis, while 5.3 million jobs were generated on average. In this scenario, autonomous vehicles would increase the separation rate by less than four-tenths of a percent.111 And this would be very unlikely to happen over a fifteen-year period. Technology adoption is never frictionless, and it will take much longer for taxis to become fully autonomous than long-haul trucks. The worrying part is that the alternative options for large groups in the labor market are continuously worsening. Even assuming that replaced truck drivers are reabsorbed with relative ease in the ceaseless churn in the labor market, we have to ask ourselves into which jobs and at what wages? And if their options look unattractive, will they choose to work at all?
A truck driver in the Midwest is not likely to become a software engineer in Silicon Valley. He might take up work as a janitor. Or he might find work in grounds maintenance, keeping parks, houses, and businesses attractive. (Both of these jobs, our estimates suggest, will not be exposed to the next wave of automation.) If he became a janitor he would trade a $41,340 job (2016 annual median income) as a truck driver for a $24,190 job. If he manages to become a ground maintenance worker, he would make $26,830 per year. Or he might get a job as a social care worker, earning $46,890 per year. But that would require him to get a college degree.
Leontief once joked that if horses had been given the right to vote, their disappearance from farms would have been less likely. Though the American middle class has hardly suffered the same fate as that of the farm horse, we would not expect Americans to simply accept falling wages. People might willingly accept automation if it reduces their incomes temporarily. But if their earnings seem unlikely to recover for years or even decades, they are more likely to resist it. Indeed, if individuals are unhappy with the verdict of the market, they can either try to block the technology or demand more redistribution through nonmarket mechanisms and political activism. We saw in chapter 3 that the Luddites and other groups vehemently opposed the spread of machines that threatened their livelihoods. In addition to rioting, they petitioned Parliament, appealing to the government to restrict the introduction of worker-replacing technologies—but their case was hopeless because they lacked political influence. Today, working people do not just have higher expectations for what governments must provide. They also have political rights.
13
THE ROAD TO RICHES
From time to time, people have thought that technological progress was about to come to an end. As the industries that constituted the key drivers of the Industrial Revolution—textiles, rail transport, and steam engineering—started to slow at the end of the nineteenth century, some observers asserted that the capitalist system had broken down.1 In a similar spirit, during the Great Depression, Marxist critics declared that capitalism was incapable of achieving sustained growth, and non-Marxist writers, like the economist Alvin Hansen, predicted that the U.S. economy was in for a period of secular stagnation in part due to faltering innovation: “When a revolutionary new industry like the railroad or the automobile … reaches maturity and ceases to grow, as all industries finally must, the whole economy must experience a profound stagnation.… And when giant new industries have spent their force, it may take a long time before something else of equal magnitude emerges.”2
Robert Gordon has recently put forward an equally bleak outlook on the future of growth in The Rise and Fall of American Growth.3 He argues that contemporary breakthroughs in artificial intelligence (AI), mobile robotics, drones, and other by-products of the computer revolution cannot match the great inventions of the early twentieth century. There is no way of knowing whether future productivity growth will reach rates seen in the golden years, but in light of the technologies on the horizon (see chapter 12), it seems to me that productivity will pick up again, provided that innovation is allowed to progress uninterrupted. The trouble is that many of those technologies are of the replacing sort, so that they will exert further pressure on the wages of the unskilled (see figure 18).
Before the dawn of automation, more than half of working American adults were employed in blue-collar and clerical jobs, which supported middle-class living for citizens with no more than a high school degree. Over the past thirty years, the number of these jobs has steadily declined, causing many of those who did not go to college to seek employment in low-income service jobs (see chapter 9). AI now threatens also to replace humans in many of those jobs, which used to be a safe haven for the unskilled—worsening their employment prospects even further. In this light, the
concern is not that productivity growth will fail to pick up. The more serious challenge, it seems to me, exists not in technology itself but in the area of political economy. As the brilliant David Landes observed, “Even assuming that the ingenuity of scientists and engineers will always generate new ideas to relay the old … there is no assurance that those men charged with utilizing these ideas will do so intelligently [and] there is no assurance that noneconomic exogenous factors—above all, man’s incompetence in dealing with his fellow-man—will not reduce the whole magnificent structure to dust.”4
If left unaddressed, the increasing divide between winners and losers from automation could have serious social costs that go far beyond those borne by the individuals whose jobs are directly affected (see chapter 10). Already, the growing economic divide has translated into a greater political divide that is challenging the very fabric of liberal democracy (see chapter 11). In the twentieth century, steadily growing incomes were taken as a given, and people still expect their material standards to improve. But in the age of automation, it has become harder for governments to deliver on that promise, as growth in middle-class wages has fallen behind that in productivity. The populist backlash, in large part, reflects the failure of governments to make the gains from growth more widely shared. Indeed, the wages of noncollege workers have been in decline for more than thirty years—a long-standing process that was unmasked by the Great Recession (see chapter 11). As Francis Fukuyama puts it, “The future of democracy in developed countries will depend on their ability to deal with the problem of a disappearing middle class.”5
The Technology Trap Page 38