The helpful comments of David Autor and Robert J. Gordon on an earlier version are acknowledged.

Carl Benedikt Frey rose to sudden fame in the economics profession in 2013 when he circulated a working paper with Michael Osborne, “The Future of Employment: How Susceptible Are Jobs to Computerisation?,” which was widely cited as arguing that 47 percent of U.S. jobs were at a “high risk” of automation (the paper was eventually published in 2017 in Technological Forecasting and Social Change), and made him an instant celebrity in doom-and-gloom circles. In a recent retrospective on the ensuing debate, The Economist referred to him as “an accidental doom-monger” and pointed out that he was in fact much more optimistic than he had been made out to be. The famous 47 percent of workers were simply in the risk class with the relative highest ranking, but all that meant was that they were the most vulnerable, not that they were all inevitably going to be made redundant. Frey has spent much of his career thinking about the effects that technological progress has had on labor, employment, and the distribution of income. This long, detailed, and erudite book is the result. It is based on wide reading in the economic history literature as well as in labor economics and the economics of technological change.

Frey’s interest in economic history is driven by asking what technological progress has done to the demand for labor in the past and what the resulting changes to income distribution were. He argues that changes in distribution unfavorable to labor creates political resistance by workers to technological change, and thus threatens further progress. He relies on an analysis of the past to understand current-day concerns regarding the effects of automation, robotization, artificial intelligence, and other perceived technological threats to workers. At the center of the argument is a distinction between “laborenabling” and “labor-replacing” technological progress. The former is job-creating and raises the real income of labor and its share in national income, as innovation is on the whole complementary with labor. In the “labor-replacing” case, innovation and labor are gross substitutes and therefore technology threatens the jobs and reduces the income share of workers.

The methodology he deploys draws heavily on economic history in search of parallels with contemporary issues. Frey draws an analogy between the current-day growth of labor-replacing digital technology with the age of the Industrial Revolution. As he sees it, the two events are “exceedingly similar” (p. 244, 329), and when assessing the impact of modern technology and automation he states categorically that “the message of this book is that we have been here before,” that is, during the British Industrial Revolution (p. 348). Such analogies are understandably popular; for all but professional scholars. Clichés about history repeating itself have an irresistible attraction. In all fairness, Frey is sophisticated enough to realize that history does not quite repeat itself, but it may sometimes rhyme (another quote attributed to but never uttered by Mark Twain).

Yet, some rhymes are better than others. Drawing lessons from the past to make economic policy is, to say the least, controversial. There are times when such lessons are useful, such as the lessons from the Great Depression that inspired the Obama stimulus package of 2008. But on the whole, one should be cautious in drawing historical analogies, especially when it comes to technological change, where the past may be an especially poor guide to the present, let alone the future. Was the technological progress during British Industrial Revolution much like the modern revolution in robots, automation, artificial intelligence, and other putative threats to labor?

To start with, it is always useful to recall John Stuart Mill’s warning that the great danger in the study of history is not so much mistaking falsehood for truth, as to mistake a part for the whole. The technological advances during the Industrial Revolution advanced on a broad front, and while economic history books have understandably emphasized the spectacular advances in the cotton industry and the steam engine, scholars have long stressed that other improvements—by no means all labor-saving— affected both productivity and living standards. Even in textiles, moreover, the view that technological change was all labor-replacing does not do justice to the complex reality, in which machinery produced not only cheaper goods but simply better ones. The higher quality of the yarn spun by the throstle, to cite just one example, made cotton warps possible, and thus paved the road to a much superior all-cotton fabric. Better and cheaper iron and steel, gas lighting, cheaper watches, cement, paper, improved ships, machine tools, navigational instruments—all of them were part and parcel of the Industrial Revolution. They replaced some workers, but created employment for others. The net effect on the demand for labor, much less economic welfare, is hard to assess.

To be sure, many of the advances did replace labor and made some occupations redundant, most famously the domestic spinsters and handloom weavers. However, whether on balance technological change was labor-saving is hard to say. Twenty five years ago, G.N. von Tunzelmann concluded that it was not (G.N. von Tunzelmann, “Technology in the Early Nineteenth Century.” In Roderick C. Floud and D.N McCloskey, eds., Economic History of Britain since 1700. 2nd ed., Vol. 1. Cambridge: Cambridge University Press, 289-91). To repeat, this is not to say that specific industries and occupations were not adversely affected—they inevitably were if technological change was to occur at this rate.

Frey argues that the first two generations of the Industrial Revolution experienced the harsh consequences of labor-replacing industrialization and workers on the whole suffered. Between 1760 and 1850 there was indeed little increase in real wages. Other indicators also are consistent with living standards that fluctuated, but in the long term were more or less stationary before the middle of the nineteenth century. On balance, however, it is not clear what the exact significance was of what Robert Allen has called “Engels’ Pause,”—an actual decline in labor share concurrent with the Industrial Revolution. Was it really driven by technology? Evidence for rising income inequality on the economy-level (the rising part of the famous Kuznets curve) during the Industrial Revolution is quite dubious as Charles Feinstein showed 30 years ago in his classic “The Rise and Fall of the Williamson Curve” (this JOURNAL, Vol. 48, No. 3 (Sept. 1988), 699-729). Real wages and workers’ living standards rose less than labor productivity, so that the share of pure labor must have declined. But factory workers were still only a small segment of the labor force: it is less clear what happened to small-scale selfemployed producers of manufactures and services, from artisans to shop- and tavern- keepers, to wagon-drivers, physicians, scriveners, and hundreds of other occupations listed in the 1851 Census.

Moreover, technology was not the only driver of income distribution during the British Industrial Revolution. The failure of living standards to pick up before 1850 was due to a variety of well-documented factors. One pitfall of comparing the present age with that of the Industrial Revolution is that the demographics today are very different: the age of the Industrial Revolution in Britain was one of rapidly growing population, driven by rising birth rate and hence characterized by increasing dependency rates, which would tend to depress living standards. It was also an age of costly and disruptive wars between 1776 and 1815, followed by a nasty depression. A string of poor harvests, coupled to the Corn Laws which protected landlords at the expense of industrial workers, and the abolition of the old poor law in 1834 compounded the misery. Wage income may also have been affected by composition effects, as low-paid women and children joined the formal labor force. In short, a schematic two-factor model may not capture what happened to income distribution. Our own age, of course, is one of stagnant populations and below-replacement fertility in all industrialized nations. Because of ageing, it shares with the age of the Industrial Revolution the feature of rising dependency rates, but the mechanism is quite different. Moreover, the social safety nets in industrialized nations today are much more advanced than they were in the increasingly laissez-faire economy of nineteenth-century Britain.

As Frey reminds us, there was resentment toward machinery during the Industrial Revolution. Not all machine-breaking riots, however, were directed against mechanization as such—in Lancashire the attacks on machines were less against new techniques per se than against convenient targets in an unorthodox bargaining game between employees and employers. Perhaps the surprising feature of the Luddite riots was that there were so few of them, given the rapid spread of industrialization after 1815. Political grievances mixed with economic grievances (resulting for instance in the Peterloo massacre), and much unhappiness may have resulted from urbanization, congestion, pollution, and harsh work conditions. Much of the work that was replaced was in domestic rural industry, in part carried out by women especially in spinning and lace work. With few exceptions it was hard for such workers to organize.

After 1850, Frey argues, resentment against machinery subsided because technological progress became more worker-enabling. New occupations were created by the new industries and products created by the second Industrial Revolution and absorbed the workers thrown out of work by mechanization. There is undoubtedly much truth to this, and indeed living standards began increasing steadily after 1850. Again, however, other factors may have been at work as well. Not everything should be forced into a model of biased technological change. Overseas, migration after 1850 absorbed large number of European workers—an option that today’s workers in industrialized economies do not have (and do not need). Living standards after 1860 were also affected by the rapidly growing global trade in agricultural products and the growing reliance of Europeans on products from the “ghost acreage” of North America, Australia, and Argentina.

Technological progress in the past decades has had undeniable effects in “hollowing out” the demand for labor by replacing routine tasks, mechanical as well as cognitive, thereby shrinking and impoverishing large segments of the middle class. The disappearance of these medium-skill tasks contributed to a decline in the share of labor in national income. But the rise in inequality since 1980 has been far more pronounced in the United States and Britain than on the European continent or the industrialized nations of East Asia, suggesting that other factors may have been at work, since all were subject to the computer revolution that Frey blames for the worsening conditions of workers (p. 246). Moreover, in most industrialized economies innovation takes place in a much more encompassing welfare state, in which the worst effects of economic insecurity are cushioned. The deficiencies and distortions of the American welfare state and fiscal system, more than computers and automation, can be blamed for the misery that technological change inflicts. Digital technology may well have reduced the income of middle-skill and blue collar workers in favor of “symbolic analysts” (p. 245). Other factors, however, again confound the story. How much of the growth in post-tax inequality in the United States is due to the decline in unions, the falling real value of minimum wages, loophole-ridden tax laws, the emergence of winner-take-all markets, declining effective redistribution, and other factors?

The sharp distinction that Frey draws between technological progress, that is laborenabling and labor-replacing, abstracts from the General Equilibrium effects of technological progress. A labor-saving invention will lead to lower product prices, and depending on the respective elasticities of demand and substitution, an increase in output may lead to increased employment in the industry in question. Even if it does not, the income saved in the technologically advancing industry will be spent in other industries. Recent research by David Autor and Anna Salomons shows that the employment effects of labor saving innovations in one industry are offset by employment effects elsewhere, though this is not the case for the share of labor in value added (David Autor and Anna Salomons, “Is Automation Labor-Displacing? Productivity Growth, Employment, and the Labor Share.” Brookings Papers on Economic Activity 2018 (1): 1-87). But this is only half the story: every worker is also a consumer and living standards are determined by consumption; lower prices and improvements in the quality of products, not to mention altogether new goods and services, may go some way to compensate the losses sustained in labor markets.

To be sure, substantial segments of the U.S. economy have declined, and as Frey points out are “increasingly segregated from the rest of the economy.” He concedes that blaming all this on “labor-replacement” “remains an open question” (p. 256). But clearly he believes—and with some justification—that the replacement of workers by computers and machines is driving the decline in depressed areas and that middle-aged workers in rural areas and small towns are at the highest risk. During periods of rapid technological change, geography and age are destiny.

It is natural to connect the current state of the economy to automation. Modern research by labor economists has shown that “robotization” may have something to do with the declining share of labor, rising inequality, and hence political resentment of “elites” by blue-collar workers. Frey’s great concern is that political resistance against the short-term effects of labor-replacing innovation will derail progress altogether by bringing into power technophobic governments that will take us back to the dark age of pre-Industrial Revolution Europe, in which government throttled innovation (he believes) to avert labor market driven rebellions. This did not happen during the Industrial Revolution because the British government staunchly chose the side of the employers, but he fears that it could happen now. At least to date the evidence is, however, that the right-wing autocratic governments that have sprouted up in the past decade are more nationalist and tribalist than technophobic. People whose economic position is perceived to have deteriorated have turned their wrath in the direction of globalization and immigrants. Whether this makes sense or not, so far there seems to be little evidence of a “Luddite” political backlash against automation that Frey is concerned about. History may be a poor guide to the future.

Frey has a tendency to make past history just a little bit too neat and fit his scheme. In a chapter entitled “Why Mechanization Failed” he argues that many of the inventions of the Industrial Revolution could have been made many centuries earlier, but that governments, fearing worker backlash to the labor-replacing inventions, blocked them (p. 138). Only after the English Civil War and the Glorious Revolution, he claims, did the British government become sufficiently friendly to innovation to make the Industrial Revolution possible. But the timing of the Industrial Revolution required more than a government friendly to property and capital. It required skills and technical sophistication that produced the kind of workmanship and materials that were simply unavailable in 1600. What made steam power possible was not just the insights of learned people such as Torricelli and Papin, it was also the metallurgical skill of John Wilkinson, the ironmaster, the engineering talents of William Murdoch, Watt’s able assistant, and thousands like them. It is clear why Frey wants to make this point: he worries that the bad old days may be brought back by a reactionary regime driven by worker resentment. But the notion that pre-Industrial Revolution economies were held back primarily by rulers fearful of popular uprisings is unconvincing and does not fit the historical record.

Three further caveats against jumping too quickly into alarmist conclusions are in order. The first is that technological progress in the twenty-first century is far wider than the digital technologies that are the main focus of Frey’s book. Advances in nanotechnology, genetic engineering, material sciences, energy physics—to name but a few— may affect economies just as much as the much-feared advance of robots and AI. We do not know precisely how these advances will play out, but there is no doubt that they will create new jobs, the kind of jobs resulting from what Acemoglu and Restrepo call a “reinstatement effect.” After all, one of the characteristics of technological progress is that it creates future jobs that nobody could foresee. As Acemoglu and Restrepo argue, these effects “increase the demand for labor and tend to raise the labor share. When they go hand-in-hand with automation, the growth process is balanced, and it need not imply a dismal scenario for labor” (Daron Acemoglu and Pascual Restrepo, “Artificial Intelligence, Automation and Work.” NBER Working Paper No. 24196, January 2018, p. 32). The displaced handloom weavers of the 1810s and 1820s could not imagine their children growing up to become telegraph operators or railroad engineers any more than anyone in 1914 could have imagined their grandchildren as video game designers, cyber security consultants, or geriatric exercise physiologists. By focusing on the partial equilibrium effects of technological change being either labor-replacing or labor-enabling, Frey provides an incomplete picture of the effects of technological change on labor demand.

A second caveat points a deeper difference between our own age and the age of the Industrial Revolution and the hazards of reading too much in historical analogues. Given the inescapable demographic trends in the industrialized worlds coupled to the growing reluctance to accept immigrant workers, the industrialized economies are more likely to be threatened by labor shortages than by structural unemployment. Economies with especially bad ageing problems tend to invest more in robots, and for good reason. Moreover, labor supply may be affected from an unexpected corner: as leisure technology improves, the marginal rate of substitution between leisure and income may rise. Prime-age workers may work less (or consume more leisure at work) as video games become more attractive, sports events become ever-more accessible, and social media more sophisticated. As labor forces shrink, and dependency ratios rise, robots are not a threat to workers but an indispensable supplement. If faced with a choice between more immigrants and more robots, the current political climate seems to indicate that most governments will choose the latter. Ageing may also affect the demand for labor—geriatric care is especially labor intensive, and while new technology may help, many tasks require a human touch.

Finally, Frey’s account does not seem to realize that the threat of labor-replacing technology may be exaggerated. Artificial intelligence may be great at playing “GO” but in production it seems much more of a complement to labor than a replacement. In two places (p. 236, 310—the book has a tendency toward repetitiveness), Frey cites computer scientist Hans Moravec as pointing out that digital technology is good at tasks that people are poor at, and vice versa. That suggests strong complementarity. Moreover, some AI experts have been warning that its promise is overhyped. Oren Etzioni, the CEO of the Allen Institute for Artificial Intelligence, is one of the more sensible voices in the deafening clamor about AI. He points out that there is a profound difference between autonomy and intelligence. People have both. “AI systems, even if they have very high intelligence for a specific task, tend to be low in autonomy. And that lack of autonomy limits the technology’s ability to take on many broad problems” (MIT Technology Review 26 March 2018). Not only that horror scenarios of hundreds of thousands of truck drivers being all of a sudden replaced by self-driving semi-trailers are fanciful, an increasingly digital world and the much-hyped “internet of things” are increasingly vulnerable to cybercrime and sabotage, which will slow down their adoption. The one thing about human workers is that they cannot be hacked.

Despite such misgivings, this is an important book that economic historians should read. Frey is erudite and thoughtful, and the questions he raises are important and pertinent. The analogy between two historical episodes, even if a bit overdrawn, is tempting and a lot can be learned from it. While most practitioners of economic history may be somewhat hesitant to map historical experiences to current-day questions (much less policy recommendations, that Frey inevitably takes up in his closing chapter entitled “the road to riches),” it is undeniably fun to think along such lines even if one feels a bit guilty afterward. Moreover, for teachers of undergraduate economic history such exercises are useful in classes, provided that students understand both the insights that historical analogies provide and their pitfalls.