Major social maladjustments lie ahead, and how best to respond to them is not a straightforward issue. If the negative impacts of automation are overemphasized, exaggerated fears over its deleterious effects may be aroused. But if their significance is underestimated, it is likely that precautions to minimize the individual and social costs will be neglected, and as a result people might quite rationally oppose replacing technologies.6 If history is any guide to workers’ reactions to the next wave of automation, it is telling that the Industrial Revolution was a time when many citizens fell between the cracks of transition and consequently, technological change was vehemently resisted (see chapter 5). On several occasions, the British government clashed with angry machine-smashing craftsmen, whom progress was forced upon. Yet resistance was not as vehement everywhere. The Old Poor Law helped ease the transition to the modern world. The economic historians Avner Greif and Murat Iyigun have shown that there was less popular resistance to technological change, and less social disorder, in parts of England where welfare institutions provided more generous support to the poor.7 There were also some contemporaries, even if they were few, who realized the importance of compensating the losers to progress to avoid social and political upheaval. In his 1797 book on poverty, Sir Frederick Eden rightly pointed out that machines “promote the general wealth,” but he added that they “throw many industrious individuals out of work; and thus create distresses that are sometimes exceedingly calamitous.” He declared that poor relief must be used so that the machines’ “inconvenience to individuals will be softened and mitigated, indeed, as far as it is practical.” He contended that failing to adequately do so would lead to stagnation, as people would then oppose machines the way they had in the preindustrial era.8
The rise and fall of the poor laws, discussed in chapter 11, reflected a shift in political power from the landed classes to the new urban elite, whose members saw little gain in helping people stay in the countryside. Instead, they needed workers for their factories. But the poor laws’ fall was also a consequence of the widespread belief that technology could not improve the human lot. Industrialization was advocated for in the national interest and unleashed to make sure that Britain did not lose ground to its rival nations in trade. And even though Malthusian forces in Britain had long disappeared, Malthusian logic was still thriving. Thomas Malthus’s contemporaries and the generation of political economists after him believed that population growth would always undo economic growth in per capita terms. One implication of this belief was that any attempt to redistribute income to make the benefits of industrialization more widely shared was always doomed (see chapter 2). Both Malthus and David Ricardo vehemently opposed poor relief, which they believed would just encourage the poor to have more children rather than helping them.9 We now know better.
In the twentieth century, governments assumed a wider responsibility for alleviating some of the adjustment costs imposed on the workforce. The labor movement, including its political branch, de facto accepted technology as the engine of growth but insisted on establishing a welfare system to provide credible assurance to all members of society that their personal income would not fall below a certain lower bound, making personal losses more narrowly constrained. And the newly generated wealth from industrialization allowed for more social spending, making it easier for society to compensate the less well off. As noted, one important reason why the socialist revolution that Karl Marx prophesied did not happen is that technology began to work in workers’ interests, and consequently laborers quite rightly came to regard it as the engine of their good fortune. The adoption of the steam engine and, later, electrification created new and better-paying jobs for workers, who eventually acquired the skills required to run the machines. But another reason is that governments diffused the threat of revolution from below by expanding the franchise, creating a welfare state, and building an educational system that eased adjustment to the accelerating pace of change. Thus, quite naturally, the coming AI revolution has prompted calls for a capitalist reinvention of similar magnitude.
What Can Be Done?
Historically, the worst times for labor have been those characterized by both worker-replacing technological change and slow productivity growth. If AI technologies turn out to be as brilliant as some of us think, we should be more optimistic about the long run. As Daron Acemoglu and Pascual Restrepo have pointed out, brilliant technologies are much preferable for labor than mediocre ones because as they make us richer, they create more demand for other goods and services produced by humans.10 Indeed, wages grew faster between 1995 and 2000, when computers prompted a brief productivity boom, than in the preceding and succeeding years. But while high productivity growth is always preferable to slow growth, growth in wages may fall behind that in productivity if technology is of the replacing sort, and some workers might see their incomes vanish in the process—even as new jobs are created elsewhere in the economy. That is what has happened in recent years, and it is also what happened during the classic years of industrialization.11
National unemployment in America today stands at 4 percent. Work is seemingly not about to come to an end, despite the rise of the robots. Instead, automation has manifested itself in falling wages for large swaths of the population, leading some to drop out of the workforce. The rising percentage of workers that are now outside the workforce, who are not accounted for in the unemployment rate, is particularly troubling. In Men without Work, Nicholas Eberstadt estimates that if the current trend continues, 24 percent of men ages 25–45 will be out of work by 2050. Joblessness is especially prevalent among men without a college degree, who lack the skills to compete in the ever-higher-tech economy.12 They are the ones who have seen their earning potential diminish due to automation, and because they lack the necessary skills, they have been excluded from the new and emerging well-paying jobs (see chapter 9).
If current trends continue in the coming years, the divide between the winners and losers to automation will become even wider. And there are good reasons to think that it will. Looking at the automatability of existing jobs, we have seen that most occupations that require a college degree remain hard to automate, while many unskilled jobs—like those of cashiers, food preparers, call center agents, and truck drivers—seem set to vanish, though how soon is highly uncertain. But there are also unskilled jobs that remain outside the realms of AI. Many in-person service jobs—like those of fitness trainers, hairstylists, concierges, and massage therapists—that center on complex social interactions remain safe from automation.13
There is no way of knowing exactly what jobs the future will bring. At the advent of the Industrial Revolution, nobody could have foretold that many Englishmen would become telegraphers, locomotive engineers, and railroad repairmen. Today, futurologists are just as ill equipped to predict the jobs that AI will create. Official employment statistics are always behind the curve when it comes to capturing new occupations, which are not included in the data until they have reached a critical mass in terms of the number of people in them. But other sources, like LinkedIn data, allow us at least to nowcast some emerging jobs. Among them are the jobs of machine learning engineers, big data architects, data scientists, digital marketing specialists, and Android developers.14 But we also find jobs like Zumba instructors and Beachbody coaches.15
In a world that is becoming increasingly technologically sophisticated, rising returns on skills are unlikely to disappear and likely to intensify. Like computers, AI seems set to spawn more skilled jobs for labor, in the process creating more demand for in-person service jobs that remain hard to automate. As noted above, much of recent job creation has centered on the so-called labor multiplier. Computers have created jobs for software engineers and programmers, which in turn have raised the demand for in-person service jobs in the places where they work and live (see chapter 10). Thus, where skilled jobs are abundant, the unskilled earn better wages, too. In San Jose, California, fitness trainers and aerobics instructors mad
e $57,230 on average in 2017. In Flint, Michigan, they averaged $35,550 annually. Of course, direct comparisons are complicated by a variety of factors. It is true that the cost of living in the Bay Area is higher than it is in Flint. But it is just as true that amenities are more plentiful, health outcomes and public services are better, and crime rates are lower.
Automation then represents a double whammy. Where machines have replaced middle-class workers, the demand for local services has also suffered. The growing divide between the skilled and the unskilled has been amplified by the staggering divergence between skilled and unskilled places. The Bay Area has prospered from the miracles of software engineering, while labor in the Rust Belt has suffered from the implementation of new technologies that were invented elsewhere. And in many places, where middle-class jobs have dried up, vanishing incomes have brought a range of social problems like rising crime, faltering marriages, and deteriorating health (see chapter 10). Many of these problems, as we all know, are negatively correlated with rates of intergenerational mobility. They could have long-lasting effects on communities as they also have adverse consequences for the next generation. In this light, the appeal of populism is not hard to understand. It gives voice to the anger of those who have been excluded from the engines of growth and trapped in places of despair.
The message of this book is that we have been here before. We should recall Maxine Berg’s noting of the “unprecedented demands for mobility, both geographical and occupational,” that accompanied the Industrial Revolution. We should remember that machines “meant, or at least threatened, unemployment, an unemployment which at best was transitional between and within sectors of the economy.” But above all, we should bear in mind that “the conceptual changes in political economy over the period are also very closely connected to class struggle [which was evident] in the very seriousness attached by political economists to the 1826 anti-machinery riots in Lancashire and to the 1830 agricultural riots.”16
Engels’s pause eventually came to an end, as enabling technologies came to the rescue and workers acquired new skills. But by that time, three generations of ordinary Englishmen had seen living standards decline. Governments today can mercifully assume wider responsibility for the social costs brought by technological change. Indeed, the growing percentage of men in their prime who are not working and the steady decline in the earnings capacity of those with no more than a high school degree suggest that we must think carefully about short-run dynamics as AI-enabled automation progresses. As productivity growth makes the pie larger, everyone could in principle be made better off. The challenge lies in the sphere of politics, not in that of technology. Given the enormous potential for AI to make us richer on the one hand, and the specter of disruptions to labor on the other hand, governments must carefully manage the short run, which was a lifetime for many during the classic years of industrialization.
As the former secretary of the treasury Lawrence Summers puts it, “Little is certain. But we will do better going forward than backward [which] means embracing rather than rejecting technological progress.… This will be a major debate that I suspect will define a large part of the politics of the industrial world over the next decade.”17 To avoid the technology trap, governments must pursue policies to kick-start productivity growth while helping workers adjust to the onrushing wave of automation. Addressing the social costs of automation will require major reforms in education, providing relocation vouchers to help people move, reducing barriers to switching jobs, getting rid of zoning restrictions that spur social and economic divisions, boosting the incomes of low-income households through tax credits, providing wage insurance for people who lose their jobs to machines, and investing more in early childhood education to mitigate the adverse consequences for the next generation. In the next sections, we shall look at what can be done in more detail.
Education
If people race alongside the machine, they are less likely to rage against it. And historically, education has been the way workers have adjusted to accelerating technological change. The seminal 2008 book by the economists Claudia Goldin and Lawrence Katz, The Race between Education and Technology, shows that the solid performance of the U.S. economy and the expansion of education over the first three-quarters of the twentieth century were not coincidental. The former was, at least in part, the consequence of the latter. The fact that the twentieth century was dominated by America and was the human capital century, the authors write, was not a historical accident: “Economic growth in the more modern period requires educated workers, managers, entrepreneurs, and citizens. Modern technologies must be invented, innovated, put in place, and maintained. They must have capable workers at the helm. Rapid technological advance, measured in various ways, has characterized the twentieth century. Because the American people were the most educated in the world, they were in the best position to invent, be entrepreneurial, and produce goods and services using advanced technologies.”18
We saw in chapter 8 that the race between technology and education did a good job of explaining much of what was going on in the American labor market up until 1980, when technological change increasingly became of the replacing type. Yet replacing technological change made education more rather than less important. As discussed in chapter 9, people have adjusted very differently to automation depending on their educational background. As middle-income jobs for the semiskilled began to dry up, people falling into low-paying service jobs or out of the workforce were overwhelmingly citizens without a college degree. Those with a college education, in contrast, were more likely to move up in the ranks.
Unskilled work is not coming to an end, but as noted, low-skilled jobs are more exposed to future automation, while occupations that require a college degree remain relatively safe. And though it remains to be seen what the jobs of the future will be, and exactly what skills they will require, we do know what some of the barriers to acquiring new skills are. Arguably the greatest policy challenge is that in study after study, children from disadvantaged backgrounds are shown to have consistently lower educational attainment. As is well known, deficits in basic skills like math and reading, which surface in the early years of life, mean that children typically fail to catch up with their peers in later grades. One reason that such deficits arise in the first place is that children from low-income families often lack the intellectual stimulation from in-home reading and daily conversation that are so common in families where one or both parents have completed college. We also know that parents in the top fifth of the income distribution spend seven times more on enriching extracurricular activities and educational materials for their children—like books, computers, and music lessons—than those in the bottom fifth.19 In this light, it stands to reason that as automation causes the incomes of many parents to vanish, it also diminishes the future prospects of their children. The economist Jeffrey Sachs and colleagues have indeed argued that AI threatens not just to reduce the jobs, wages, and savings of the current generation, but also to impoverish future generations as a consequence.20
To level the playing field, governments are advised to invest more heavily in early childhood education. Gaps in knowledge and ability between children from disadvantaged backgrounds and their relatively advantaged peers open early on and tend to persist throughout life. A proactive approach through investments in high-quality early childhood programs will therefore be more effective and economically preferable than efforts to bridge the gap later on. And preschool education for children from poor families pays for itself. Those are the findings of James Heckman, winner of the Nobel Prize in Economics, and colleagues—whose research shows dramatic long-term effects of early interventions, with a rate of return on investment of 7–10 percent per annum through much improved educational outcomes, better health, productivity, and reduced crime.21 Other work by Arthur Reynolds and colleagues, published in Science, reaches a similar conclusion. In their study, they traced the fates of more than fourteen hundred participa
nts in the Chicago-based Child-Parent Center Education Program over twenty-five years. Their findings show that relative to the control group, program participants did strikingly better in terms of educational attainment, income, substance abuse, and crime, with the strongest enduring effects for males and children of high school dropouts.22 As things stand, the overall societal costs of the opportunity gap, though hard to estimate, are striking by all accounts. Economists have put the aggregate annual costs of child poverty to the U.S. economy at $500 billion per year, which is equivalent to almost 4 percent of the gross domestic product. These costs stem from low productivity growth, higher crime rates, and cascading health expenditures.23
It is true that none of these studies account for the fact that the opportunity gap almost certainly will have some bearing on the future rate of innovation. In a pathbreaking study, the economist Alexander Bell and colleagues set out to analyze why some Americans are more likely than others to become inventors. Drawing upon data on 1.2 million inventors from patent records, the authors found that children from low-income families are much less likely to become inventors even if they exhibit the same ability as high-income children, as measured by test scores.24 This innovation gap grows in later grades—which, the authors argue, “is because low-income children steadily fall behind their high-income peers over time, perhaps because of differences in their schools and childhood environments.”25
The Technology Trap Page 39