The Rise of the Machines How Computers Have Changed Work

UBS Center The Rise of the Public Paper #4 Machines How Computers Have Changed Work

David Dorn

No. 4, December 2015 UBS Center Public Paper The Rise of the Machines: How Computers Have Changed Work

Contents

3 About the Author Abstract

4 Introduction

5 Is Technological Development Just Beginning or About to End?

8 Drawing on Past Experience: A Short History of Technological Change in Textile Production

12 The Computer Revolution: Renewed Fears about the Automation of Work

14 Computers in Economic Theory: Skill-Biased vs Task-Biased Technological Change

17 Labor Market Polarization

21 Technology versus Globalization

24 How Can Workers Succeed in a Computerized Labor Market?

26 Will the Lessons of the Past Remain Relevant in the Future?

28 Conclusions

29 References

31 Edition About Us

2 About the Author Abstract

The so-called “Rise of the Machines” has Prof. David Dorn fundamentally transformed the organization of work during the last four decades. While enthusiasts are captivated by the new technologies, many worry that these machines will eventually lead to mass unemployment, as robots and computers substitute for human labor.

This Public Paper shows that these concerns are likely to be exaggerated. Despite rapid technological progress and automa- – Professor at the Department of Economics, tion, unemployment has not dramatically University of Zurich expanded over time. Instead, employment – Affiliated Professor at the UBS Center shifted from the most highly automated sectors to other sectors that experienced Contact [email protected] less technological progress, as well as emerging sectors that were created by new technology. David Dorn is Professor of International Trade and Labor Markets at the University of Zurich and Affil- While computers have little impact on iated Professor at the UBS Center. He was previously overall employment, however, they con- a tenured Associate Professor at the Center for Mon- tribute to rising inequality. Machines etary and Financial Studies (CEMFI) in Madrid, a have overtaken humans in their capability Visiting Professor at Harvard University, and a Visit- to execute well-defined routine tasks pre- ing Scholar at Boston University, MIT, and the Uni- cisely, and many of the production and versity of Chicago. clerical jobs that specialize in these tasks have been irreversibly lost. As a result, the His research connects the fields of labor economics, employment structure of labor markets in international trade, economic geography, and mac- developed countries has become increas- roeconomics. In particular, he studies how globali- ingly polarized as employment concen- zation and technology affect labor markets. trates in a set of highly paid and a set of lowly paid occupations, both of which are Professor Dorn is a Research Fellow of the Centre for difficult to automate. Economic Policy Research (CEPR) in London, the Institute for the Study of Labor (IZA) in Bonn, and As computerization changes the composi- the Center for Economic Studies/ifo Institute (CESifo) tion of human labor rather than decreas- in Munich. He is on the Editorial Board of The Re- ing its overall amount, policymakers view of Economic Studies, and an Associate Editor of should not be primarily concerned about the Journal of the European Economic Association. mass unemployment. Instead, the more immanent policy challenges caused by His research on the labor market impacts of techno- computerization result from changing logy was funded in part by the Swiss National Science skill demands in the labor market and ris- Foundation (SNSF) and the Spanish Ministry of ing economic inequality among workers. Science and Innovation.

3 UBS Center Public Paper The Rise of the Machines: How Computers Have Changed Work

Introduction

The first fifteen years of the 21st century This essay discusses the impact of tech- have been a difficult period for workers nology on the labor markets in devel- in developed economies. In many oped countries. It argues that an wealthy European, North American, and imminent large decline in the demand for East Asian countries, the share of popu- human labor is far from certain, and that lation holding a job has declined, and speculation about the long-term evolu- wage growth for the average worker has tion of employment is inherently diffi- slowed or even turned negative. The cult. However, there are already “Great Recession” of the years 2007– hundreds of years of experience regard- 2009 is to blame for an important part ing the technological change of the past, of this decline in workers’ fortunes. and researchers have been able to study However, the employment rate in the the impact of computer technology on United States had already been falling for the labor market for several decades. I several years prior to this crisis, and argue that the evidence from the distant labor markets in many countries and recent past reveals recurring pat- remained depressed for a remarkably terns, which may usefully guide expecta- long time after the recession had offi- tions about technology’s impact on labor cially ended.1 New evidence also suggests markets in the near future. that the fraction of national income obtained by workers has been declining for at least three decades in developed countries.2

It is therefore natural to hypothesize that labor markets are not just in a temporary slump, but instead face a fundamental force that increasingly deteriorates workers’ outlook for finding a job. Computer technology is an obvious candidate for that role. Whether one enters an office building or a factory, the widespread use of personal computers, communication devices, computer-guided machines, and robots is a striking feature of the workplace of the 21st century. Computer technology often replaces work tasks that humans previously executed, and one thus wonders whether continued technological development will eventually lead to the obsolescence of most human labor.

4 Is Technological Development Just Beginning or About to End?

Gordon E. Moore, co-founder of Fairchild computing operations has been falling Semiconductor and Intel Corporation, spectacularly since the 1950s (Figure 1).3 predicted in 1975 that technological progress in the semiconductor industry would Computers are not only becoming cheaper allow a doubling in the number of elec- and more powerful, but also more versa- tronic components per microchip every tile. New applications in areas such as two years, thus leading to rapid progress machine learning, artificial intelligence, in computer processing power. This pre- and mobile robotics hold great promise diction, which became known as for expanding the use of computer tech- “Moore’s Law,” has held up remarkably nology to an ever-wider set of tasks. One well since then. The dramatic progress in of the most publicized innovations in semiconductor technology led to great recent years is the self-driving car, which performance improvements in many elec- may profoundly reshape ground transpor- tronic devices, and to rapidly declining tation in the 21st century. prices of computing equipment. Data compiled by the economist William Nord- Erik Brynjolfsson and Andrew McAfee, haus indicates that the cost per million two scholars at the MIT Sloan Business

Fig. 1 Evolution of cost per computing operation (1850–2010)

In $

1,000,000

10,000

100

0.01

0.0001

0.000001

0.00000001 1850 1870 1890 1910 1930 1950 1970 1990 2010

Manual Mechanical Vacuum Transistor Microprocessor

Note: The costs represent $ per million standardized computing operations per second (MSOPS).

Source: Nordhaus (2001)

5 UBS Center Public Paper The Rise of the Machines: How Computers Have Changed Work

School, argue that the rapid fall in com- electricity and the development of a pan- puter prices and the wider applicability oply of electrical devices; the invention of of computer technology are heralding a the internal combustion engine which “Second Machine Age” that will trans- revolutionized transportation; a great form the economies of developed coun- improvement in sanitary and living con- tries to an even greater degree than the ditions due to running water, indoor Industrial Revolution.4 As ever-cheaper plumbing, and central heating; the ability robots are able to execute an ever-greater to rearrange molecules which permitted range of tasks, firms will only hold on to great progress in the development of workers as long as a machine cannot exe- pharmaceuticals, plastics, and other cute their jobs more cheaply. The result is chemical products; and the introduction enormous downward pressure on work- of major communication and entertain- ers’ wages, and as these wages fall below ment technology, following the invention a reservation level—the minimum level of of the telephone, the phonograph, pho- wages that makes it worthwhile for peo- tography, radio, and motion pictures ple to hold a job—a rapidly growing pool within a span of just fifteen years. of unemployed individuals will form. Annual growth of U.S. GDP per capita A technology-driven disappearance of accelerated during the Great Innovation most employment opportunities will pro- period, and peaked at a 2.5% average foundly change society, and will require a gain per year between 1950 and the onset new organization of many public institu- of the oil crisis in 1973. In the four tions. Despite the chilling outlook of mass decades since, average annual GDP unemployment, the Second Machine Age growth has been one-third lower at 1.6% will not only produce losers, however. per year. While annual growth even The widespread use of cheap robots in exceeded 4% in the ten years during the production will lower production costs, 1950–1973 period, such a high growth and the resulting productivity increase rate has never again been attained during raises aggregate societal wealth. However, the last thirty years. This slowdown in these gains are likely to be concentrated economic growth is not unique to the among a small group of owners of com- United States, but it is even more pro- puter capital, while a much larger fraction nounced in Japan and in the major Euro- of the society will suffer from the loss of pean economies. gainful employment. It is an interesting intellectual exercise to think about appro- Gordon argues that the slowdown in eco- priate public policies that could success- nomic growth results from a slowdown fully deal with a workless and highly in innovation. The Great Innovation cen- unequal society. tury, which was characterized by major developments in multiple important tech- The jobless robot age predicted by Bryn- nologies, has been followed by a period jolfsson and McAfee is, however, far with a comparatively one-dimensional from certain. Quite to the opposite, Rob- development in computer technology ert Gordon, an economic historian from alone. Indeed, Gordon makes the provoc- Northwestern University, posits that ative prediction that innovation and eco- technological change is slowing rather nomic growth will continue to slow as it than accelerating.5 He argues that the becomes increasingly difficult for human- period of greatest technological advance ity to come up with fundamentally novel in history is not the computerization of discoveries (see Figure 2). the late 20th and early 21st century, but the hundred years from the early 1870s Of course, pathbreaking future innova- to the early 1970s. This century of tion is extremely difficult to predict. Who “Great Innovation” saw the discovery of would have foreseen today’s omnipres-

6 Fig. 2 Economic growth in countries with technological leadership (1300 – 2100)

In % per year

3.0

2.5

2.0

1.5

1.0

0.5

0.0 1300 1400 1500 1600 1700 1800 1900 2000 2100

Actual UK Actual US Interpolated Hypothetical Path

Note: Growth in real GDP per capita with actual and hypothetical paths.

Source: Gordon (2012)

ence of the internet-connected multipur- pose smartphone just thirty years ago? But as much as it is problematic to extrapolate the slowdown in economic growth into the future, one can also not confidently extrapolate past trends in the development of computer technology—or foresee an acceleration of that development—in order to conclude that an age of robots is immanent and inevitable.

While long-term predictions are notori- ously difficult, this essay argues that one can learn important lessons from past experience. Historical evidence not only allows assessing the impacts of past technological change on the labor market, but these impacts can be contrasted with past predictions about the transformative effects of technology. Indeed, concerns about the replacement of workers by machines date back many centuries, and there also is mounting evidence on the role of computers and robots in the labor market during the last three to four decades.

7 UBS Center Public Paper The Rise of the Machines: How Computers Have Changed Work

Drawing on Past Experience: A Short History of Technological Change in Textile Production

Humans have been producing textile would lose their livelihood if the use of clothing for thousands of years. In this the new, more productive technology production process, cotton, wool, or were permitted.6 other fibers are first converted to yarn, then yarn is converted to cloth, and Mechanization also reached the second finally cloth is converted to clothing. production step of the textile industry, While the basic sequence of production the weaving of yarn to cloth. As of the steps remained unchanged over time, 17th century, a predecessor of the mecha- there were dramatic improvements in the nized loom permitted the simultaneous execution of each step. During the Indus- production of up to 24 woven ribbons, a trial Revolution, the textile industry was dramatic improvement over the classical at the forefront of a broader trend loom that could only produce one ribbon towards mechanization of production. at a time. To prevent employment loss Yet already prior to that transformative among ribbon weavers, the German period, the textile sector provided a use- emperor prohibited the use of this mech- ful example for the study of technological anized ribbon loom in 1685, and regents change and its labor market implications. of many other European territories did the same. A first major innovation in the textile sector affected the process of spinning, The imposition of technology bans in an i.e., the conversion of fiber to yarn. His- attempt to prevent employment loss, torically, humans would attach fibers to however, proved counterproductive in the a spindle and rotate that spindle by hand long run. Right outside the borders of the in order to twist fibers to yarn. From German empire, craftsmen in the Swiss about the 13th century onwards, these city of Basel adopted the new ribbon- hand spindles were gradually replaced by weaving machine, and were thus able to spinning wheels, which took advantage produce at much lower cost than their of the rotational energy of a wheel, and German counterparts. Competitive pres- greatly increased the productivity of the sure from technology adopters eventually spinning process. led to growing opposition against the German empire’s machine ban, which The introduction of the spinning wheel was finally revoked in the mid-18th cen- met occasional resistance from the crafts tury.7 guilds that controlled production in many European territories. According to The process of textile production soon historical documents, the city council of changed even more dramatically with a Cologne decided in the year 1412 that a series of inventions in the late 18th cen- local merchant by the name of Walter tury, at the start of the Industrial Revolu- Kesinger would not be allowed to con- tion. The spinning wheel was replaced by struct a spinning wheel, after he had seen the “Spinning Jenny,” a multispindle such a machine during travels in Italy. spinning frame that could produce many The council argued that many spinners yarns at a time. Weaving was revolution-

8 ized by the introduction of the power loom, an automated loom that was pow- Key technological changes in the ered by steam or water energy rather than by human hand. spinning industry

The new production technology required Hand spinning expensive machines and access to a cen- Woman using a hand spindle. Detail from tral power source. Therefore, the decen- an Ancient Greek Attic white-ground oino- tralized home production of yarn and choe (wine jug), ca. 490 BC, from Locri, cloth by spinning wheels and handlooms Italy. was replaced by mass production in factories. In a mechanized factory, a single worker was able to produce an output that would have required dozens of workers prior to the Industrial Revolu- tion. © British Museum, London

The massively reduced need for human Spinning wheel labor in textile production led to popular Woman spinning with a wheel, from the unrest in England. Unemployed workers Elizabethan era, early 17th century. protested against the new system of factory production, and in some cases attacked factories and smashed machines. The British government reacted by mak- ing “machine breaking” a capital crime. The protesting workers, who became known as Luddites after their alleged © New York: American Heritage Publishing, 1967 leader Ned Ludd, were persecuted and their movement broken up. Spinning frame A common theme during these centuries The improved “Spinning Jenny” that was used in textile mills, England, 18th century. of technological progress in the textile A worker operating a “Spinning Jenny” with industry was the fear that new labor-sav- 60 spindels could replace about 25 hand ing technologies would lead to long-term spinners. mass unemployment. Indeed, the introduction of new machines certainly dis- rupted the labor market, and many workers lost their jobs when their work © 2004–2015 Florida Center for tasks became mechanized. While con- Instructional Technology cerns about technology-induced unemployment were well founded in the short Automated spinning machine run, the predicted long-term decline in Factory building with “Mule Jenny,” England, employment never materialized. 1834. A worker operating a “Mule Jenny” with 14 spindles could replace about 175 An intuitive, and yet profoundly mis- hand spinners. taken view of the labor market is that there is a fixed amount of work, which can either be done by humans or by machines. According to this view— known to economists as the “lump of © Science Museum/Science and labor fallacy”—an increasing use of Society Picture Library machines in the production process nec-

9 UBS Center Public Paper The Rise of the Machines: How Computers Have Changed Work

essarily reduces the work that is available ignoring the emergence of new employ- to humans. This contention is a fallacy ment opportunities, many past observers because it fails to recognize two channels of technological change have fallen vic- through which even labor-replacing new tim to the “Luddite fallacy” of wrongly technologies can create additional predicting a rise of long-term unemploy- employment. ment (see box on the next page).

First, new technology is often associated with the emergence of new industries and occupations. The development of spinning machines and power looms, for instance, led to the creation of many jobs in a new machine-producing industry. Second, there is a less obvious but prob- ably more important job creation effect that operates via changes in prices and consumer spending. Mass production in the textile industry led to a dramatic drop in the price of clothing. This price decline allowed consumers either to buy a greater quantity of clothing with the same amount of money, or to buy the same amount of clothing at lower cost while expanding purchases of other goods and services. The resulting increase in demand for clothing and other outputs led to greater production and rising employment in many sectors of the economy, particularly in those that were not directly exposed to labor-saving technology.

Over time, technological change did not eliminate employment, but it strongly changed its composition. From the 19th to the late 20th century, a large part of employment first moved from agriculture to manufacturing, and later from manufacturing to the service sector. It would have been unthinkable two centuries ago that agriculture, which employed the large majority of workers at that time, would account merely for a few percentage points of overall employment today, despite producing a much greater output. Yet unemployment has not shown a pronounced upward trend over time, as workers have found new employment opportunities in industries like health- care, finance, and entertainment, where the rising employment shares would have been equally difficult to foresee. By

10 The Luddite fallacy in numbers

Fig. 3 Unemployment rate Historical data from the United Kingdom, Unemployed in % of economically active population aged 16 and over where the Luddites staged their protests in the early 19th century, illustrate the extent of 20 the so-called “Luddite fallacy.” Despite fundamental technological change, unemploy- 15 ment levels show no secular trend. They are, 10 however, characterized by strong cyclical movements, rising rapidly after World War I, 5 in the Great Depression, during the “stagfla- 0 tion” period of the 1970s, and in the wake of the recent financial crisis. This absence of a 1915 2015 1975 1875 1925 1935 1955 1905 1895 1995 1855 1965 1985 1945 2005 1865 1885 long-term trend is not due to changing defini- tions of unemployment. As Figure 4 makes clear, the percentage of workers in the total population has remained remarkably stable, Fig. 4 Employment levels hovering around 50 percent for a very long People in work in % of total population time.

100 80 60 40 20 0 1915 2015 1975 1875 1925 1935 1955 1905 1895 1995 1855 1965 1985 1945 2005 1865 1885

Source: Bank of England "The UK recession in context–what do three centuries of data tell us?" Data Annex, Version 2.2, July 2015.

11 UBS Center Public Paper The Rise of the Machines: How Computers Have Changed Work

The Computer Revolution: Renewed Fears about the Automation of Work

The last and continuing wave of techno- The computer revolution quickly ignited logical change that reached the labor mar- fears about mass unemployment. As early ket is the adoption of computer tech- as in 1978, the German news magazine nology in the workplace. Its predecessors Der Spiegel titled: “The computer revolu- date back at least to the Industrial Revo- tion: progress creates unemployment.” It lution, when the French inventor Joseph argued that computers differed from ear- Marie Jacquard developed a loom oper- lier technological innovations, since they ated by replaceable punched cards, which not only eliminated many jobs in auto- controlled a particular sequence of the mating sectors, but also failed to create a loom’s operations. The same principle of meaningful number of new jobs in com- operating machines with programs stored puter production or elsewhere in the on punched cards was later used in main- economy. The article cited a British frame computers from the 1960s on- union leader who predicted that by the wards. year 2000, most jobs would have been replaced by computers. This pessimistic The start of the computer revolution is, prediction turned out to be quite mis- however, often dated to the late 1970s or taken. The unemployment rate of the early 1980s. Production in factories was United Kingdom (see Figure 3), which already changing rapidly at that time, as computer-guided production machines became more widely and cheaply available. These devices included computer numerical control (CNC) machines that were operated by a computer program, and industrial robots that could move a robot arm around multiple axes. Comput- erization also started to affect office work, most notably due to the introduction of personal computers in the early 1980s. The IBM Personal Computer was released in 1981; the downmarket Com- modore 64, which became the best-selling computer of all time, followed in 1982, and Apple introduced its iconic Macin- tosh in 1984. Continued technological development has since led to ever more powerful and versatile computers, production machines, and robots. And great advances in communication technology, including the World Wide Web and wire- Cover of the 16/1978 edition of the German less communication devices, have further news magazine Der Spiegel. increased the reach of these machines. © DER SPIEGEL

12 stood at 6% in 1978, again hovered around 6% in the year 2000, when sup- posedly most work should already have been lost. And it remained at 6% in 2015, a full 37 years after Der Spiegel wrongly predicted a “social catastrophe” of unprecedented mass unemployment.

The fact that computers failed to eliminate most human labor in the past four decades despite predictions to the opposite invites a healthy skepticism about renewed claims that robots are just about to take over from humans.

More likely than not, today’s technology enthusiasts will be seen as the next vic- tims of the Luddite fallacy within a few decades, thus joining the many previous pundits that predicted the end of human work in the past.

Of course, the fact that computers did not create mass unemployment does not mean that they had no impact on labor markets. Quite to the contrary, many economists consider computerization, along with the globalization of goods flows and worker flows, as the main driver of change in labor markets during the last three or four decades. A closer look at computers’ effects on workers during that period is warranted both in order to understand the recent past, and to provide some guidance for thinking about possible future impacts of computer technology.

13 UBS Center Public Paper The Rise of the Machines: How Computers Have Changed Work

Computers in Economic Theory Skill-Biased vs Task-Biased Technological Change

Computers became an important topic on a long sequence of technological innova- the research agenda of macroeconomists tions that have favored more highly edu- and labor economists in the 1990s. cated workers. But how exactly could Researchers had observed that the wage one explain that computers raise the rela- differential between university-educated tive productivity of university-educated workers and those with lower educational workers? David Autor and Frank Levy attainment had risen rapidly during the from the Massachusetts Institute of Tech- previous decade, both in the United States nology and Richard Murnane of Harvard and in several other developed countries. University conducted field studies that A leading hypothesis to explain the analyzed the introduction of computer growth of wage inequality was skill- technology in firms, and they observed biased technological change (SBTC). It computers’ impact on employment levels, posits that new technology and machines wages, and job content of different types augment the productivity of all workers, of workers. Based on their findings, they but that productivity gains are larger for formulated a more refined theory for the more highly educated workers. The grow- impact of computers on the labor mar- ing productivity advantage of educated ket, which has become known as the workers increases firms’ demand for these task-biased technological change (TBTC) high-skill employees. hypothesis.9

The Harvard economists Claudia Goldin In the TBTC model, computers do not and Lawrence Katz have argued that such have a differential impact on workers SBTC has taken place throughout the based on their education levels, but based 20th century, thus continuously raising on the task content of their occupations. the demand for skilled labor.8 That grow- This model draws on the key observation ing demand coincided with a rapidly that computers have distinct strengths and growing supply of skilled labor, as aver- weaknesses when it comes to executing age education levels increased dramati- different work tasks. A personal computer, cally in all regions of the world. The CNC machine, or robot is directed by 1980s, however, marked a period during software that was prespecified by a pro- which the growth in the supply of skilled grammer. Computers are thus good at workers slowed in the United States, executing tasks that follow a well-defined whereas demand for these workers con- procedural routine. These “routine tasks” tinued to grow or even accelerated. The are often found in repetitive production excess growth of skill demand relative to work. The production of car bodies in the skill supply generated an increase in the automotive industry, for instance, requires relative wage of workers with higher edu- that the exact same work steps be cation, and thus greater wage inequality repeated over and over, while the margin in the labor market. for error is small in order to ensure high quality and replaceability of parts. Robots Through the lens of the SBTC hypothesis, are much better than humans at executing computers are seen as the continuation of such high-precision repetitive work.

14 Fig. 5 Which jobs are most exposed to automation?

Percentage of routine-intensive occupations

0 0 20 40 60 80 100

Job’s percentile in ranking of occupational mean wages

Note: Occupations are ranked by their average hourly wage in 1980. A job at the 20th percentile earns more than 20% of all jobs in the economy, but less than 80% of all jobs. Routine occupations are defined as the ⅓ of occupations with highest routine task intensity according to job task data.

Source: Autor and Dorn (2013)

Routine tasks also appear in a second with people and objects. This includes area of the labor market. Many clerical limitations in verbal communication with occupations deal with data work, includ- humans, as well as difficulties with the ing data processing, data storage, data recognition and physical handling of retrieval, and data transmission. When- objects—all tasks that require versatility ever these data tasks are executed accord- and adaptation to the environment. ing to clearly specified rules, then they belong to the set of routine tasks that can The TBTC hypothesis summarizes tasks readily be done by computers. For that involve problem solving, creativity, instance, computers now execute many or managerial leadership under the term tasks that were previously done by “abstract tasks.” These tasks all draw on accountants, file clerks, or secretaries. human ability to react to new developments and problems, and to come up with While computers are often better than new ideas and solutions. Occupations humans at doing routine tasks, they face that strongly rely on abstract tasks important limitations in the execution of include managers, engineers, medical doc- other tasks. Machines that follow a pre- tors, and researchers. A common feature specified program cannot readily produce of these jobs is that they require a high new ideas and inventions, and struggle to level of cognitive skill, and typically cor- react to unforeseen influences on their responding to a university education. work. Moreover, computers and machines While computers tend to be poor substi- often lack good interfaces for dealing tutes for humans in these occupations,

15 UBS Center Public Paper The Rise of the Machines: How Computers Have Changed Work

they can instead be valuable comple- of the next room. But in practice, every ments. Many abstract task-intensive occu- guest will leave her room in a slightly dif- pations become more productive when ferent state. Apart from differences in computers allow a cheaper and more cleanliness, guests can leave towels, pil- rapid processing, storage, and transmis- lows, toiletries, pens and many other sion of data. For instance, an engineer objects that belong to the hotel in differ- who designs bridges benefits when com- ent spots within the room. It would be puters permit a rapid calculation of a very challenging for a robot to find and planned object’s static properties, while a recognize all of the hotel’s objects, assess manager of a large company benefits their state of cleanliness, and take the from access to real-time data that indicate appropriate measures of cleaning or the state of operations at the firm’s multi- replacing them. Compared to humans, ple plants. Far from being replaced by robots are often very limited in their machines, these workers specializing in physical adaptability, and cannot grip or abstract tasks hence stand to benefit from clean many different types of objects. An a more widespread use of computers. even greater challenge arises when hotel guests leave behind novel objects that The third group of tasks in the TBTC they brought into the room, like a pizza model is referred to as “manual tasks,” box or a jewelry box. It is easy for a and is characterized by a combination of human to recognize these objects, and to fine motoric movement, visual recogni- decide that the pizza box should be dis- tion, and verbal communication. Manual carded while the jewelry box should be tasks are important in personal service kept. The same task, however, presents a occupations such as waiters, childcare major obstacle for a machine. workers, or hairdressers, but also in transportation, repair, and construction jobs. Workers in these occupations typically require relatively little formal schooling, since manual tasks build on basic abilities such as verbal communication, seeing and recognizing persons and objects, as well as holding and moving objects with the human hand. Computers have little direct impact on these manual task-intensive jobs, which are not easily automated, but which also do not benefit substantially from an interaction with computers in the workplace.

Cleaners of hotel rooms are an excellent example not only for illustrating a manual task-intensive job, but also for explaining the difference between repetitive and routine work. The sequential cleaning of hotel rooms is certainly a repetitive chore. However, this repetitive- ness does not translate to routineness in the sense used here. For hotel cleaning to be a routine job, it would be necessary that the cleaning of one room would encompass exactly the same work steps and physical movements as the cleaning

16 Labor Market Polarization

In summary, the TBTC hypothesis pre- managers and professionals, but higher dicts a decline of employment in routine wages than manual task jobs such as per- task-intensive occupations as cheaper and sonal service workers. As a consequence, more powerful computers become avail- declining employment in routine task- able, while occupations specializing in intensive occupations translates to a pat- abstract or manual tasks cannot be read- tern of employment polarization, as ily replaced by machines. These predic- workers become increasingly concen- tions are supported by evidence from trated in the highest and lowest paid many countries in North America, occupations of the labor market. In fol- Europe, and East Asia. For instance, my low-up research, Goos, Manning, and research with David Autor observes that Anna Salomons showed that this polar- the routine task-intensive occupation ization is remarkably pervasive across groups of production workers, machine countries. In all 16 European countries operators and clerical workers accounted they studied as well as in the U.S., the for a roughly constant 37 to 38 percent employment share of occupations with of U.S. labor input from 1950 to 1980, before declining rapidly to just 28 per- 10 cent of U.S. labor in 2005. As routine Fig. 6 Employment polarization in Europe and in the United States occupations contracted, managerial and professional occupations that intensively Change in employment shares by occupation 1993 – 2006 in 16 European countries use abstract tasks grew rapidly. Employ- ment in low-skilled service occupations, USA such as waiters, cleaners, and childcare EU Average Italy workers, has been expanding since the Austria 1980s; this work is rich in manual tasks. France Occupations in farming, mining, con- Luxembourg struction, repair, and transportation, Denmark which also mostly perform manual tasks, Belgium were declining rapidly until 1990, but Spain have since stabilized their employment Germany share. Sweden UK While the TBTC hypothesis provides Greece clear predictions for the distinct effects of Netherlands computerization on occupations that use Norway Finland different job tasks, it also has indirect Ireland implications for the inequality between Portugal workers of different education or income levels. The economists Maarten Goos –0.15 –0.1 –0.05 0 0.05 0.1 0.15 0.2 and Alan Manning pointed out that rou- Percent change tine occupations in production and clerical work tend to be clustered towards the Lowest-paying 3rd Middle-paying 3rd Highest-paying 3rd middle of the occupational wage distribu- Note: Occupations grouped by wage tercile: Low, middle, high. tion.11 These jobs typically have lower wages than abstract task jobs such as Source: Acemoglu and Autor (2011) and Goos, Manning and Salomons (2009)

17 UBS Center Public Paper The Rise of the Machines: How Computers Have Changed Work

intermediate wage levels has declined, or falling wages, depending on country and in all but one of these countries, both and time period. employment in low-wage and in high- wage occupations has grown relative to Indeed, the U.S. experience of a com- the middle-wage jobs (Figure 6).12 While bined employment and wage growth in the exact extent and shape of this low-skilled service occupations is sur- employment polarization varies by coun- prising in the context of the SBTC and try, it is striking that the same basic pat- TBTC theories. Both theories predict tern is so pervasive despite substantial that computers enhance the productivity international differences in industry of skilled workers in occupations with structures, labor market regulations, and abstract tasks, and the resulting growing local economic growth. The average changes across European countries are not just qualitatively, but even quantita- Fig. 7 Employment polarization tively very similar to the trends observed in the United States. Change in employment share 1980–2005, in %

The polarization of the occupational employment structure invites the ques- 40 tion whether similar patterns can be 30 observed for wages. In the United States, 10 this is indeed the case. For Figures 7 and 0 8, the several hundred occupations that are observed in the U.S. Census have –10 been ordered according to their average –20 wage in 1980. During the next 25 years, 0 20 40 60 80 100 both occupational employment and occu- Job’s wage percentile pational wage growth was larger in the highest-paid occupations (on the right side of both graphs) and in the lowest- Fig. 8 Wage polarization paid occupations (on the left) than in occupations with intermediate wages Wage growth 1980–2005, in % (towards the center). Closer inspection of the data suggests that the very pro- 30 nounced wage growth in high-wage occu- 25 pations was driven by managerial and 20 professional occupations, whereas wage 15 growth in low-wage occupations 10 stemmed largely from low-skilled service jobs. 5 0 20 40 60 80 100

International evidence on wage polariza- Job’s wage percentile tion is sparser and less homogeneous than in case of employment polarization. Note: The graph shows occupational employment and wage growth in the United States. Occupations are ordered on the x-axis according to their per- Highly paid occupations experienced centile in the occupational wage distribution of 1980. Figure 7 measures the both employment and wage gains in relative change in an occupation’s share in total employment (for instance, many countries, suggesting that the an occupation that expands from 1% to 1.2% of total employment has grown growing supply of highly skilled workers by 20%). Figure 8 measures the change of the mean log real hourly wage by for these jobs has not kept up with grow- occupation group. Wage changes are expressed in log points, which approxi- mately correspond to percentage changes. The data series in the graph are ing demand. The expansion of employ- smoothed. ment in low-paid occupations, however, has coincided with either rising, stagnant Source: Autor and Dorn (2013)

18 demand of firms for skilled workers can The observation that the demand for readily raise their employment and low-skilled services is increasing is impor- wages. However, both theories suggest tant because these jobs can provide that computers have little effect on the employment opportunities for workers productivity of low-skilled workers. In who have little education or job-specific the task-based view, the employment training. It strongly contradicts the share of low-skill service occupations notion that all low-skilled work is rapidly and other jobs with manual tasks will becoming obsolete as a consequence of expand relative to employment in the computerization. routine occupations that become automated. However, as redundant clerical and production workers seek new employment in manual task jobs such as cleaners and waiters, which are readily accessible for workers with little formal education, one would expect wages in these occupations to fall. The observation of a combined increase of employment and wages in low-skilled service occupations thus suggests that a second force must be at work in addition to labor reallocation from routine to manual task-intensive jobs.

David Autor and I argue that this second force is a growing demand for services that are produced with low-skilled labor. Computerization reduces production costs primarily for manufactured goods, but not for low-skilled services such as cleaning or childcare, where computers hardly affect the production process.13 When consumers perceive goods and services to be poor substitutes in consump- tion, then they react to the falling price of goods not by buying a larger quantity, but by using some of the money that is saved on cheaper goods to purchase more services. Increased consumer spending on, for instance, restaurant meals then increases firms’ labor demand for workers who produce that service. The rising demand for low-skilled services and the growing supply of workers to these jobs due to the automation of routine work combine to generate growing employment in low-skilled service occupations. This employment growth can be accom- panied by wage growth if the demand for low-skilled services grows sufficiently rapidly relative to the increasing supply of workers.

19 UBS Center Public Paper The Rise of the Machines: How Computers Have Changed Work

Labor market polarization in Switzerland

Fig. 9 Employment polarization in Switzerland A recent paper of the UBS Center Working Paper Series studies whether the Swiss labor Average change in employment share per decade since 1980, in % market has polarized analogously to the labor +4 markets in the United States and European Union. Its authors, Andreas Beerli and Ronald Indergand, indeed find evidence for polariza- +2 tion in Switzerland.

0 Figure 9 shows a pronounced decline of the share of routine task-intensive occupation –2 groups in total employment during the period of 1980 to 2010. Both production and clerical jobs are in decline. Instead, a rising share of –4 .

c Swiss employment concentrates in abstract

craft task-intensive occupations at the top of the clerks wage distribution, such as managers, profes- operators managers

technicians sionals, and technicians. In addition, there is professionals

service & sales a small expansion of employment in low-paid elementary oc service and sales occupations. While Figure Note: These numbers represent the change in mean log hourly real wages 9 is based on data from Swiss-born workers per decade of these occupation groups between 1990 and 2010. For only, Beerli and Indergand also observe em- example, the share of managers grew on average by roughly 1.6 percentage points in every decade, so that their share of total employment rose from ployment polarization among immigrants. about 6% in 1980 to around 11% in 2010. The data includes Swiss-born The authors point out that the polarization workers only. of labor demand is a strong driver of the skill-composition of newly arriving immigrants explaining, for instance, why their level of Fig. 10 Wage polarization in Switzerland education has increased considerably in the last 30 years. Average change in hourly wages per decade, in %

4 These unequal fortunes of different occupation groups are mirrored in their rates of wage growth, as shown in Figure 10. From 1991 to 3 2011, real wages grew most in the lowest-paid occupations (service and sales) and in the 2 highest-paid ones (managers, professionals, and technicians). By contrast, workers in jobs 1 characterized by routine tasks are barely better off than in the early 1990s.

0 clerks craft & operators & managers technicians, occupations professionals & elementary & service, sales

Note: These numbers represent the percentage changes per decade in the mean log hourly real wages of these occupation groups over the entire period from 1991 to 2011.

Source: Beerli and Indergand (2015).

20 Technology versus Globalization

The task-based model of technological This spatial reorganization of production change provides a useful framework for can take the form of offshoring, where thinking about the use of computers and multinational firms shift part of their robots in the workplace. It has also been operations to another country, or of an empirical success, as its predictions trade competition, where firms in devel- about patterns of occupational employ- oped countries reduce employment as ment growth turned out to be consistent they succumb to competitive pressure with evidence from many developed from imported goods. Global shifts in countries. However, the fact that comput- production are particularly apparent in erization could be a plausible explanation manufacturing, and closely tied to the for employment polarization does not spectacular economic development of imply that it has to be the only explana- China. Building on a series of market- tion for this trend. Most importantly, oriented reforms, China evolved from many economists have pointed out that being a minor player in international in the same period when labor markets trade in the early 1990s to becoming the were exposed to computerization, they world’s leading exporter of goods in also faced a rapidly progressing global- recent years. ization. The contemporaneous occurrence of Globalization is a process of interna- computerization and globalization makes tional integration that encompasses rising it difficult to estimate their separate trade in goods and services, growth in effects on employment polarization and international capital movements, migra- other aggregate outcomes in developed tion of workers, and dissemination of economies. My work with David Autor knowledge. Integration has deepened and Gordon Hanson proposes to study along all these dimensions in recent the impact of macroeconomic forces on decades, and progress in computer and local labor markets in the United States communication technology may well in order to overcome this dimensionality have acted as a catalyst for that develop- problem.14 The concept of local labor ment. The use of foreign suppliers, for markets builds on the empirical observa- instance, has become more attractive for tion that workers usually seek jobs at firms thanks to technology that allows workplaces located within commutable for easy and cheap communication over distance from their homes. As a conse- long distances and seamless tracking of quence, local labor supply and local shipments. labor demand combine to form separate market equilibria in different localities, Globalization provides an alternative and spatial variation in real wages and narrative that could explain the decline employment levels can be quite persistent of middle-wage occupations in many over time. developed economies. Workers in production and clerical occupations may not Firms in different cities and rural areas of only be replaced by computers and the United States should all have access robots, but also by workers in other to the same technologies. Nonetheless, countries where wage levels are lower. the impact of computerization will have

21 UBS Center Public Paper The Rise of the Machines: How Computers Have Changed Work

spatial variation. This variation stems indeed shows that historically routine from the fact that local labor markets task-intensive local labor markets, which vary in their industry, occupation, and are dispersed across all regions of the task mix. The historical source of local U.S., adopted more computers since the specialization can be the geographic 1980s while witnessing greater declines proximity to important raw materials, in routine work. As a consequence, local access to transportation infrastructure, or labor markets with a high initial employ- even the serendipitous emergence of an ment share of routine occupations experi- important firm around which other enced stronger employment polarization related businesses cluster.15 Importantly, than locations with comparatively little patterns of local specialization are reliance on routine work (Figure 11). remarkably stable even over long periods Moreover, there was also somewhat of time. For instance, local labor markets greater wage polarization in these local that made particularly large use of rou- labor markets with greater exposure to tine labor in 1950 still have a dispropor- computers.16 tionately routine-intensive occupation mix half a century later. The historical The comparison between local labor reliance on routine labor later creates a markets not only allows establishing a large potential for the replacement of more direct link between computeriza- workers by computers and robots, and tion and labor market polarization, but thus a particularly pronounced exposure also permits the separation of the effects of these locations to computerization. An of computerization from those of global- empirical analysis of U.S. Census data ization and other economic forces. In

Fig. 11 Employment polarization in U.S. local labor markets with different exposure to computers

Change in employment share 1980–2005, in %

–10

–20 0 20 40 60 80 100

Job’s wage percentile

Low exposure to computers High exposure to computers

Note: The graph shows employment polarization in the United States separately for local labor markets with above average and below average routine employment in 1980 (high and low exposure to computerization). The vertical axis represents the change in an occupation’s share of total employment times 100.

Source: Autor and Dorn (2013)

22 research with David Autor and Gordon Hanson, I show that the U.S. local labor markets which are most exposed to computerization only partially coincide with the regions that are most affected by the dramatic rise of import competition from China. Econometric analysis can therefore identify the local labor market impacts of these technology and trade forces separately.17

An empirical investigation of U.S. Census data shows that the extremely rapid increase of Chinese import competition since the 1990s has led to a substantial decline of employment in all occupation categories of the manufacturing sector. Due to slow reallocation of manufacturing workers to other jobs, this import shock has also reduced overall employment. By contrast, routine task-intensive local labor markets have not experienced a significant overall job loss due to computerization, but their occupational employment structure has polarized both within the manufacturing and non- manufacturing sectors. Consistent with the timeline of technology development, the loss of routine jobs started with the automation of production work in manufacturing during the 1980s, and later became larger in the service sector as computers started to replace clerical work.18

Empirical studies focusing on European data reach the same conclusion as the analyses for the United States: The adoption of computers and robots has not caused a notable decline in overall employment. However, computerization contributes to labor market polarization, as middle-wage routine occupations decline while high-wage abstract and low-wage manual occupations expand.19

23 UBS Center Public Paper The Rise of the Machines: How Computers Have Changed Work

How Can Workers Succeed in a Computerized Labor Market?

Computers are transforming the occupa- outdo computers in terms of information tional composition of the labor market. storage or calculation. Victories of com- Young adults who enter the labor market puters over extremely accomplished today face a very different set of job humans in quiz shows and chess competi- opportunities than their parents a genera- tions have impressively shown the tre- tion ago. Many production and clerical mendous advantage that machines now occupations with intermediate wages hire have in such tasks. fewer workers than they used to, and employment polarization is particularly pronounced among the young, who are becoming disproportionately concen- Humans retain an advantage trated in high-wage and low-wage jobs.20 The continued availability of jobs in low- over machines when it comes skilled service occupations also offers employment opportunities to workers to problem solving, creativity, who have little formal education. Yet the workers who stand to gain most are those and interaction with other in the highly paid managerial and professional occupations where productivity has humans. risen thanks to computer technology.

It is thus an easy policy recommendation that more schooling is desirable in order An education that prepares young people to help cohorts of young workers succeed for the task demands of the 21st century in the labor market. After all, the rise of should thus seek to strengthen skills in the wage differential between workers these areas, for instance by fostering with and without university education problem-solving abilities and communica- suggests that the growth in the relative tion skills through case study projects, supply of highly educated individuals has group work, and other modern forms of fallen short of the growth in relative teaching that complement a more tradi- demand for their labor. tional mode of instruction based on lec- tures and memorization. The policy response to technological change should, however, not just be more A greater focus on individualized cus- education, but also different education. tomer interaction and on innovation and Computers have changed, and will con- problem solving can also provide a per- tinue to change the demand for job tasks spective for some of the shrinking middle- in the labor market. Therefore, education wage occupations. Clerical and produc- should build skills in those tasks where tion jobs that integrate these non- human capabilities remain superior to routine tasks cannot be as readily auto- machines, and not in dimensions where mated as jobs that only execute routine machines have the edge. An education tasks, at least not without a substantial that emphasizes rote memorization and loss in quality. A machine operator who mental arithmetic is no longer able to has a thorough understanding of the produce skilled workers who can hope to machine’s operation, and of the produc-

24 tion process it is embedded in, is harder to replace than an operator who is just familiar with a few buttons on the machine’s operating panel. The former will be able to quickly resolve problems and may even propose improvements that raise the efficiency of the production process. Similarly, a salesperson who expertly advises customers and carefully responds to individual customer requests will not as easily lose her job to a machine as a colleague who merely swipes credit cards at the cash register. These jobs with virtu- ous bundles of job tasks do not require a university education, but benefit from a high-quality vocational training system that combines hands-on experience on the job with schooling that is tailored to the need of a specific occupation.

25 UBS Center Public Paper The Rise of the Machines: How Computers Have Changed Work

Will the Lessons of the Past Remain Relevant in the Future?

The prediction that labor-saving technol- as great breakthroughs, including super- ogy inevitably causes a long-term rise in sonic airplanes, maglev trains, or solar unemployment has been proven wrong vehicles, have never become cheap and many times in history, and again during powerful enough to be adopted widely in the first few decades of computer adop- the economy. tion. Yet the question persists whether the theory of task-biased technological A further limit to automation stems from change, which currently guides many the fact that human workers often execute economists’ understanding of computers’ a more diverse bundle of tasks than those impact on the labor market, will remain that a computer or robot can replace. A useful in the future as technology evolves truck driver for instance not only steers a further. truck through traffic, but also loads, con- trols, and unloads the cargo; deals with A prime example for the expanding possi- the accompanying paper work; and takes bilities of technology is the driverless car. care of maintenance and small repairs. Just over a decade ago, academic research The driverless technology alone will thus listed truck drivers as one example of a not be able to substitute for all the work a manual task-intensive occupation that truck driver does. In some situations, it is cannot readily be substituted for by tech- feasible to split up the task bundle of an nology.21 In the meantime, however, an occupation, and have machines execute immanent automation of their jobs seems some parts while other tasks remain in the inevitable to many observers. And once it hands of humans. One example is bank is possible to automate drivers, then the tellers. The work that used to be done by replacement of other manual occupations a teller is now split into dispensation of by robots may not be far away. cash, a routine task that automated teller machines handle, and a large set of other However, enthusiastic predictions about customer services, which human employ- rapid development and adoption of tech- ees still execute. A counterexample to this nology frequently underestimate the chal- unbundling of tasks is airline pilots. The lenges on the path from an experimental first crossing of the Atlantic Ocean by a prototype to a large-scale market intro- plane flown by an autopilot took place in duction of a new product. The driverless 1947. But in the almost seven decades car, for instance, will only find wide- since, the job of the airline pilot has not spread use once improved technology disappeared because pilots are still needed allows it to negotiate difficult road condi- onboard a plane in order to react to tions, once its production costs have been unforeseen conditions such as a failure of lowered substantially, and once the legal the aircraft’s engines or instruments. questions surrounding its use have been resolved. None of these arguments Overall, past experience suggests that negates the possibility that driverless cars predictions of immanent dramatic tech- will dominate the streetscape a few years nology development should be assessed from now. Yet each of these obstacles has with a healthy dose of skepticism. Yet it the potential to greatly diminish the new is still reasonable to expect that the technology’s success. Other means of boundaries between automatable and transportation that were once announced non-automatable tasks will continue to

26 shift as computers and robots become Fig. 12 The next frontier? Objects correctly and incorrectly identified more powerful and versatile. So far, com- as chairs by machine learning puters’ primary strength has been the execution of tasks that can be characterized by precisely defined routines, and it stands to reason that this fundamental insight will remain valid even as a gradually expanding set of tasks becomes accessible for computer routines.

If instead a paradigm shift should occur, then it may be due to advances in machine learning. A key obstacle to the automation of non-routine tasks is the difficulty for human programmers to specify all the work steps of these tasks Source: Grabner, Gall and Van Gool (2011) exactly in a computer program. Machine learning can help overcome this obstacle by allowing computers to guess the opti- How “intelligent” can computers become mal answer to a problem using inductive through machine learning? Some research- statistical techniques rather than formal ers predict that growing computing power procedural rules. A concrete example is and better training databases will eventu- the task of visually identifying a chair.22 ally allow machines to reach or even A conventional computer program may exceed many human capabilities. Others, specify the typical constituting features of however, expect that machines will remain a chair, such as a raised surface, four legs, error-prone. Figure 12 shows the result of and a backrest. An accordingly pro- a machine learning experiment where a grammed computer will recognize many computer identified two objects as chairs regular chairs, but not atypical models that both have a space to sit on and a that have only three legs or lack a back- backrest. However, one of these objects is rest. However, if the programmer were to actually an overturned table. While it relax the criteria for the identification of a seems certain that machine learning tech- chair, then the computer would start mis- nology will improve, it remains an open classifying other objects like cabinets or question whether computers will ever be tables as chairs. While it is easy for able to fully understand the intended pur- humans to distinguish chairs from other pose of an object, just as a human would pieces of furniture intuitively, it is surpris- immediately realize that an overturned ingly difficult to convey this human intu- table is not meant for seating. ition to a computer. Machine learning seeks to circumvent this problem. Instead of providing the computer with a rule- based script to identify chairs, the machine is fed with a training library of labeled pictures or 3-D scans of chairs and other objects. This data allows the computer to statistically infer which attri- butes increase the likelihood that a given object is a chair. If the statistical model is sufficiently good, then the computer should become able to identify other chairs accurately that were not part of the initial training library.

27 UBS Center Public Paper The Rise of the Machines: How Computers Have Changed Work

Conclusions

Computer technology has transformed the As computerization changes the composi- organization of work during the last four tion of human labor rather than decreas- decades. As robots and computers can ing its overall amount, policymakers substitute for human labor, many observ- should not primarily be concerned about ers worry that technological progress will mass unemployment. Instead, the more inevitably lead to mass unemployment. immanent policy challenges caused by Historical evidence suggests that these computerization result from changing skill concerns are likely exaggerated. Despite demands in the labor market and rising rapid technological progress and automa- economic inequality among workers. tion since at least the Industrial Revolu- tion, unemployment has not dramatically expanded over time. Instead, employment shifted from the sectors that automated the most to other sectors that experienced less technological progress, as well as emerging sectors that were created by new technology.

Empirical research finds that computers have little impact on overall employment, but contribute to rising labor market inequality. Machines have overtaken humans in their capability to execute well- defined routine tasks precisely, and many of the production and clerical jobs that specialize in such tasks have been irreversibly lost. The employment structure of labor markets in developed countries becomes increasingly polarized as employment concentrates in a set of highly paid and a set of lowly paid occupations that both are difficult to automate. The former set includes managerial and professional occupations that require such skills as leadership, creativity, or problem solving, while the latter include service occupations that combine the tasks of visual recognition, verbal communication, and fine motoric movement.

28 References

1. Acemoglu, Daron, David Autor, David Dorn, Gordon Hanson and Brendan Price. 2016. Import Competition and the Great U.S. Employment Sag of the 2000s. Journal of Labor Economics, forthcoming.

2. Karabarbounis, Loukas and Brent Neiman. 2014. The Global Decline of the Labor Share. Quarterly Journal of Economics, 129(1): 61–103.

3. Nordhaus, William D. 2001. The Progress of Computing. Cowles Foundation Working Paper no. 1324.

4. Brynjolfsson, Erik and Andrew McAfee. 2011. Race Against the Machine. Lex- ington MA: Digital Frontier Press; and Brynjolfsson, Erik and Andrew McAfee. 2014. The Second Machine Age. New York NY: W.W. Norton & Company.

5. Gordon, Robert J. 2012. Is U.S. Economic Growth Over? Faltering Innovation Confronts the Six Headwinds. NBER Working Paper no. 18315; and Gordon, Robert J. 2014. The Demise of U.S. Economic Growth: Restatement, Rebuttal, and Reflections. NBER Working Paper no. 19895.

6. Isenmann, Eberhard. 2014. Die Deutsche Stadt im Mittelalter 1150–1550. Vienna: Böhlau Verlag.

7. Bayerl, Günter. 2013. Technik in Mittelalter und Früher Neuzeit. Stuttgart: Theiss Verlag.

8. Goldin, Claudia and Lawrence F. Katz. 2008. The Race Between Education and Technology. Cambridge MA: Belknap Press.

9. Autor, David, Frank Levy and Richard Murnane. 2003. The Skill Content of Recent Technological Change: An Empirical Exploration. Quarterly Journal of Economics, 118(4): 1279–1333.

10. Autor, David and David Dorn. 2013. The Growth of Low-Skill Service Jobs and the Polarization of the US Labor Market. American Economic Review, 103(5): 1553–1597.

11. Goos, Maarten and Alan Manning. 2007. Lousy and Lovely Jobs: The Rising Polarization of Work in Britain. Review of Economics and Statistics, 89(1), 113– 133.

12. Goos, Maarten, Alan Manning and Anna Salomons. 2009. Job Polarization in Europe. American Economic Review Papers and Proceedings, 99(2): 58–63; and Acemoglu, Daron and David Autor. 2010. Skills, Tasks and Technologies: Impli- cations for Employment and Earnings. In: Orley Ashenfelter and David Card (eds.), Handbook of Labor Economics Volume 4, Amsterdam: Elsevier.

29 UBS Center Public Paper The Rise of the Machines: How Computers Have Changed Work

13. Autor, David and David Dorn. 2013. The Growth of Low-Skill Service Jobs and the Polarization of the US Labor Market. American Economic Review, 103(5): 1553–1597.

14. Autor, David and David Dorn. 2013. The Growth of Low-Skill Service Jobs and the Polarization of the US Labor Market. American Economic Review, 103(5): 1553–1597; and Autor, David, David Dorn and Gordon Hanson. 2013. The Geography of Trade and Technology Shocks in the United States. American Eco- nomic Review Papers and Proceedings, 103(3): 220–225.

15. Moretti, Enrico. 2012. The New Geography of Jobs. New York: Houghton Miff- lin Harcourt Publishing.

16. Autor, David and David Dorn. 2013. The Growth of Low-Skill Service Jobs and the Polarization of the US Labor Market. American Economic Review, 103(5): 1553–1597.

17. Autor, David, David Dorn and Gordon Hanson. 2013. The Geography of Trade and Technology Shocks in the United States. American Economic Review Papers and Proceedings, 103(3): 220–225.

18. Autor, David, David Dorn and Gordon Hanson. 2015. Untangling Trade and Technology: Evidence from Local Labor Markets. Economic Journal, 125(584): 621–646.

19. Goos, Maarten, Alan Manning and Anna Salomons. 2014. Explaining Job Polar- ization: Routine-Biased Technological Change and Offshoring. American Eco- nomic Review, 104(8): 2509–2526; and Michaels, Guy, Ashwini Natraj and John Van Reenen. 2014. Has ICT Polarized Skill Demand? Evidence from Eleven Countries over Twenty-Five Years. Review of Economics and Statistics, 96(1): 60–77; and Michaels, Guy and Georg Graetz. 2015. Robots at Work. LSE-CEP Discussion Paper no. 1335.

20. Autor, David and David Dorn. 2009. This Job is Getting Old: Measuring Changes in Job Opportunities Using Occupational Age Structure. American Eco- nomic Review Papers and Proceedings, 99(2): 45–51.

21. Autor, David, Frank Levy and Richard Murnane. 2003. The Skill Content of Recent Technological Change: An Empirical Exploration. Quarterly Journal of Economics, 118(4): 1279–1333.

22. Grabner, Helmut, Juergen Gall and Luc Van Gool. 2011. What Makes a Chair a Chair? In IIEE Conference on Computer Vision and Pattern Recognition 2011: 1529–1536; and Autor, David. 2015. Why Are There Still So Many Jobs? The History and Future of Workplace Automation. Journal of Economic Perspectives, 29(3): 3–30.

30 Edition About Us

The UBS International Center of Economics in Society, UBS Center in UBS Center Public Papers short, is an Associated Institute at the Department of Economics of the University of Zurich.

It was established in 2012, enabled by a founding donation by UBS, which the bank made on the occasion of its 150th anniversary. In view of the very generous donation, the university Issue 1 Issue 2 named the UBS Center after its bene- November 2013 April 2014 factor. China’s Great Conver- Fear, Folly, and gence and Beyond Financial Crises The new center of excellence serves two main aims, the first of which is to finance world-class research in economics on all levels, to be conducted at the university’s Department of Eco- nomics. It thereby supports the department’s ambition to further strengthen its position as one of the top economics departments in Europe and to make Zurich one of the best places for Issue 3 Issue 4 research in economics in the world. June 2015 December 2015 The center’s other aim is to serve as a The Economics of The Rise of the platform for dialogue between aca- Effective Leadership Machines demia, business, politics, and the broader public, fostering continuous Economics. For Society knowledge transfer in the process. The UBS Center Public Papers make research on topics Delivering on these aims will in turn of key societal relevance available to a broader audience also contribute to the quality of Zurich, in a simplified, compact and highly readable format. and Switzerland more generally, They are written by leading academics associated with the as leading locations for education UBS International Center of Economics in Society. and business. www.ubscenter.uzh.ch/en/publications For more information, please consult Imprint www.ubscenter.uzh.ch Publisher: UBS International Center of Economics in Society, Zurich Author: Prof. David Dorn Concept / Layout: Roman von Arx, Monika Salzgeber Printed by Abächerli Media AG, Switzerland www.ubscenter.uzh.ch

31 www.ubscenter.uzh.ch [email protected] Tel. +41446345722 CH-8001 Zurich Schönberggasse 1 Department ofEconomics University ofZurich UBS InternationalCenterofEconomicsin Society