Automation and New Tasks

February 22, 2019

Previous | Next

Daron Acemoglu and Pascual Restrepo, “Automation and New Tasks: The Implications of the Task Content of Production for Labor Demand.” Prepared for Journal of Economic Perspectives, November 6, 2018.

This paper really helped me think more systematically about the impact of automation on the demand for labor, something I’ve been thinking a lot about lately. My last post on the subject, which also featured a paper by Acemoglu and Restrepo, is here.

Technology and labor demand

One of the key issues is how to reconcile the recent concern that robotics will destroy too many good jobs with the traditional optimism of economists about the effects of technology. Throughout most of U.S. history, new technologies like farm machinery or industrial assembly lines have not stood in the way of economy-wide gains in labor demand, as reflected in real wages per capita. And yet, too many workers today are experiencing unemployment or wage loss because of automation. How do we sort this out in a way that enables us to anticipate where we may be headed and formulate sensible policy responses?

Acemoglu and Restrepo conceptualize the labor demand of an industry as a product of “value added” and “labor share”:

Labor demand = Value added x Labor share

Value added is an industry’s addition to economic output and income. Labor share is the portion of that added value that is received by labor. Labor demand is the resulting wages per capita.

The authors say that economists have primarily conceptualized technological progress as “factor-augmenting.” That is, new technologies increase the productivity of labor or machinery or both. More productive factors of production mean more value added. Assuming that labor’s share doesn’t decline, labor demand as reflected in real wages ought to increase. “Factor augmenting technologies affect labor demand mostly via the productivity effect and have a small impact on the labor share of an industry.” That was the experience, for example, of manufacturing industries in the mid-20th century.

However, an analysis that focuses on value added while holding labor share constant is not complete. Technological change can also effect how much labor is needed in a particular process of production, and how much of the income from that process goes to workers.

The task content of production

The authors’ key concept is the task content of production, which involves the allocation of tasks between labor and capital (the latter including machines and software). Focusing on the potential of new technologies to add value “often misses the major implications of technological changes that directly alter the allocation of tasks to factors.”

What makes the implications of automation hard to assess is that it has contradictory effects on labor demand–a productivity effect and a displacement effect. By raising productivity and increasing value added, it increases the potential income to be distributed to the human workers who remain. But by replacing labor with machinery, it reduces labor’s share of income. Whether wages actually rise or fall depends on the relative strength of the two effects.

To complicate things further, displaced labor does not necessarily go missing from an industry, let alone the economy as a whole. The introduction of new tasks for humans to do has a reinstatement effect that raises the labor share of production and therefore labor demand. (Personally, I’m not too fond of the term “reinstatement”, since it sounds as if workers are getting their old jobs back, which is not at all what is intended. I would prefer the term “redeployment”.) But whatever term is used, the point is important. As machines take over familiar tasks, humans can move into new economic activities where they have some advantage over machines, especially because of their general intelligence, flexibility and creativity. This has been true in the past, and how much it remains true in the robotic age is an important question for the future of work.

Labor demand in a multi-sector economy

When the authors move from discussing individual industries to discussing the entire economy, they introduce an additional effect on labor demand, the composition effect. This occurs when labor moves from one economic sector to another, and the sectors differ in labor demand. If workers move out of a sector where wages and labor’s share of added value are falling, and into a sector where they are higher, that contributes positively to labor demand.

The classic example of this process is the mechanization of agriculture, “which started in the first half of the 19th century with the cotton gin and continued with horse-powered reapers, harvesters and plows later in the century and with tractors and combine harvesters in the 20th century.” That process displaced massive amounts of farm labor and reduced labor’s share of agricultural activity and income. Much of that displaced labor went into manufacturing. Manufacturing was mechanizing too, but it maintained labor demand by increasing output and creating new manufacturing tasks. “The composition and reinstatement effects explain why, despite the mechanization of a sector making up a third of the economy [agriculture in 1850], labor demand increased and the share of labor in national income remained stable during this period.”

Labor demand 1947-1987

For two recent periods, Acemoglu and Restrepo analyze changes in labor demand, measuring the relative contributions of the effects they have discussed. They chart developments in six industries: agriculture, mining, manufacturing, construction, transportation and services.

The postwar era of 1947-1987 was a period of strong labor demand and rising real wages, which grew at an average rate of 2.4% per year.

Nothing comparable to the displacement of labor by the mechanization of agriculture occurred during this time. The only industries to suffer a loss of labor share were the relatively small industries of mining and transportation.

There was some displacement of labor due to automation in the large manufacturing industry, but it was offset by the creation of new manufacturing jobs, such as managerial and clerical jobs in corporate bureaucracies, as well as jobs in the expanding service industries. “[T]here was plenty of automation, especially in manufacturing, but this was accompanied with the introduction of new tasks (or other changes increasing the task content of production in favor of labor) in both manufacturing and the rest of the economy that offset the adverse labor demand consequences of automation.”

With displacements due to automation balanced by reinstatements due to the creation of new tasks, labor’s share of output and income remained steady. The increase in actual real wages that occurred is accounted for almost entirely by the other effect that supports labor demand, higher productivity. New technologies added value, and labor got its share of that added value in the form of higher real wages.

If technological change always worked that way, we wouldn’t be so worried about the future. But there was trouble ahead. In their data for those years, we can already see labor’s share within manufacturing peaking around 1980 and starting downward.

Labor demand 1987-2017

In this more recent period, wage growth was slower, averaging only 1.3% per year.

Labor demand suffered in both ways suggested by the same formula:

Labor demand = Value added x Labor share

Value added increased more slowly because of slower growth in productivity. In addition, labor’s share of value added declined in manufacturing, construction and mining. That was because labor displacement due to automation accelerated, while labor reinstatement due to new task creation slowed down.

Deeper explanations for these trends are harder to agree on. One puzzle is why the accelerating automation hasn’t done more to raise productivity. The authors point out that productivity gains from automation depend on the actual superiority of machines over humans for a given task. A rush to automate because a company gets caught up in a wave of technological enthusiasm or receives a tax break on new equipment may not be that helpful.

This analysis clarifies that automation will reduce labor demand when the productivity effect is not very large. Contrary to a common presumption in popular debates, it is not the “brilliant” automation technologies but those that are “so-so” and generate only small productivity improvements that will reduce labor demand. This is because the positive productivity effect of so-so technologies is not sufficient to offset the decline in labor demand due to displacement.

Favoring machines over human workers may also result in a lack of investment in education and training:

…[T]here may be a mismatch between the available skills of the workforce and the needs for new technologies, which could further reduce productivity gains from automation and hamper the introduction of new tasks—because the lack of requisite skills reduces the efficiency with which new technologies can be deployed.

The future of work

By considering a variety of things affecting labor demand, Acemoglu and Restrepo avoid a simple preoccupation with a positive factor like the productivity effect or a negative factor like the displacement effect. “Our evidence and conceptual approach support neither the claims that the end of human work is imminent nor the presumption that technological change will always and everywhere be favorable to labor.” It will be the balance of the various effects that will matter.

The authors don’t make specific policy recommendations, but some very general prescriptions follow from their analysis. We shouldn’t be afraid to automate tasks where machines have a clear advantage, since the gains in value added are too good to pass up. But we should also enhance the value of human labor, both by giving the workers who remain in automating industries their share of the gains, and also by investing in the education and training workers need to perform new tasks that humans can do better than machines. Where private employers don’t find it profitable to nurture and reward human labor, government must play a strong role for the general good.

As it stands now, too much displaced labor is crowding into low-skill, low-tech, low-wage work. That is a waste of both our human and technological resources. A jobless future where robots work, their owners get rich, and most people live off public assistance is also a dismal prospect. If we are to have a hi-tech economy, let it be one where the average worker can reap the benefits in creative work, good pay, and rewarding leisure.


Effects of New Technologies on Labor

January 4, 2019

Previous | Next

David Autor and Anna Salomons, “Is automation labor share-displacing? Productivity growth, employment, and the labor share.” Brookings Papers on Economic Activity, Spring 2018.

Daron Acemoglu and Pascual Restrepo, “Robots and jobs: Evidence from US Labor Markets.” National Bureau of Economic Research, March 2017.

I have been interested in automation’s effects on the labor force for a long time, especially since reading Martin Ford’s Rise of the Robots. Ford raises the specter of a “jobless future” and a massive welfare system to support the unemployed.

Here I discuss two papers representing some of the most serious economic research on this topic.

The questions

To what extent do new technologies really displace human labor and reduce employment? The potential for them to do so is obvious. The mechanization of farming dramatically reduced the number of farm workers. But we can generalize only with caution. In theory, a particular innovation could either produce the same amount with less labor (as when the demand for a product is inelastic, often the case for agricultural products), or produce a larger amount with the same labor (when demand expands along with lower cost, as with many manufactured goods). An innovation can also save labor on one task, but reallocate that labor to a different task in the same industry.

Even if technological advances reduce the labor needed in one industry, that labor can flow into other industries. Economists have suggested several reasons that could happen. One involves the linkages between industries, as one industry’s productivity affects the economic activity of its suppliers and customers. If the computer industry is turning out millions of low-cost computers, that can create jobs in industries that use computers or supply parts for them. Another reason is that a productive industry affects national output, income and aggregate demand. The wealth created in one industry translates into spending on all sorts of goods and services that require human labor.

The point is that technological innovations have both direct effects on local or industry-specific employment, and also indirect effects on aggregate employment in the economy as a whole. The direct effects are more obvious, which may explain why the general public is more aware of job losses than job gains.

A related question is the effect of technology on wages, and therefore on labor’s share of the economic value added by technological change. Do employers reap most of the benefits of innovation, or are workers able to maintain their share of the rewards as productivity rises? Here too, aggregate results could differ from results in the particular industries or localities experiencing the most innovation.

The historical experience

American history tells a story of painful labor displacement in certain times, places and industries; but also a story of new job creation and widely shared benefits of rising productivity. Looking back on a century of technological change from the vantage point of the mid-20th century, economists did not find negative aggregate effects of technology on employment or on labor’s share of the national income. According to Autor and Salomons:

A long-standing body of literature, starting with research by William Baumol (1967), has considered reallocation mechanisms for employment, showing that labor moves from technologically advancing to technologically lagging sectors if the outputs of these sectors are not close substitutes. Further,…such unbalanced productivity growth across sectors can nevertheless yield a balanced growth path for labor and capital shares. Indeed, one of the central stylized facts of modern macroeconomics, immortalized by Nicholas Kaldor (1961), is that during a century of unprecedented technological advancement in transportation, production, and communication, labor’s share of national income remained roughly constant.

Such findings need to be continually replicated, since they might hold only for an economy in a particular place or time. In the 20th century, the success of labor unions in bargaining for higher wages and shorter work weeks was one thing that protected workers from the possible ill effects of labor-saving technologies.

Recent effects of technological change

Autor and Salomons analyze data for OECD countries for the period 1970-2007. As a measure of technological progress, they use the growth in total factor productivity (TFP) over that period.

They find a direct negative impact of productivity growth on employment within the most affected industries. However, they find two main indirect effects that offset the negative impact for the economy as a whole:

First, rising TFP within supplier industries catalyzes strong, offsetting employment gains among their downstream customer industries; and second, TFP growth in each sector contributes to aggregate growth in real value added and hence rising final demand, which in turn spurs further employment growth across all sectors.

To put it most simply, one industry’s productivity may limit its own demand for labor, but its contribution to the national output and income creates employment opportunities elsewhere.

With regard to labor’s share of the economic benefits, the findings are a little different. Here again, the researchers find a direct negative effect within the industries most affected by technological innovation. But in this case, that effect is not offset, for the most part, by more widespread positive effects.

The association between technological change and labor’s declining share varied by decade. Labor’s share actually rose during the 1970s, declined in the 1980s and 90s, and then fell more sharply in the 2000s. The authors mention the possibility that the newest technologies are especially labor-displacing, but reach no definite conclusion. Another possibility is that non-technological factors such as the political weakness of organized labor are more to blame.

The impact of robotics

Autor and Salomons acknowledge that because they used such a general measure of technological change, they couldn’t assess the impact of robotics specifically. They do cite work by Georg Graetz and Guy Michaels that did not find general negative effects of robots on employment or labor share in countries of the European Union. That’s important, since many European countries have gone farther than we have in adopting robots.

The paper by Acemoglu and Restrepo focuses on the United States for the period 1990-2007. (They deliberately ended in 2007 so that the impact of the Great Recession wouldn’t muddy the waters.)

The authors used the definition of robot from the International Federation of Robotics, “an automatically controlled, reprogrammable, and multipurpose [machine].” Over the period in question, robot usage increased from 0.4 to 1.4 per thousand workers. “The automotive industry employs 38 percent of existing industrial robots, followed by the electronics industry (15 percent), plastic and chemicals (10 percent), and metal products (7 percent).”

Adoption of industrial robots has been especially common in Kentucky, Louisiana, Missouri, Tennessee, Texas, Virginia and West Virginia. As Thomas B. Edsall titled his recent New York Times column, “The Robots Have Descended on Trump Country.”

Acemoglu and Restrepo classified localities–technically “commuter zones”–according to their “exposure” to robotics, based on their levels of employment in types of jobs most conducive to robotization.

Their first main finding was a direct negative effect of robotics on employment and wages within commuting zones:

Our estimates imply that between 1990 and 2007 the increase in the stock of robots…reduced the employment to population ratio in a commuting zone with the average US change in robots by 0.38 percentage points, and average wages by 0.71 percent (relative to a commuting zone with no exposure to robots). These numbers…imply that one more robot in a commuting zone reduces employment by about 6 workers.

The workers most likely to be affected are male workers in routine manual occupations, with wages in the lower-to-middle range of the wage distribution

In the aggregate, these local effects are partly offset by “positive spillovers across commuting zones”–positive effects on employment and wages throughout the economy. With these spillovers taken into account, the estimated effects of robotics on employment and on wages are cut almost in half, dropping to 0.20 percent and 0.37 percent respectively.

The authors state their conclusion cautiously, as “the possibility that industrial robots might have a very different impact on labor demand than other (non-automation) technologies.”

Summary

While there is little doubt that new technologies often displace labor in particular industries and localities, the aggregate effects on employment and wages are less consistent.  Historically (late 19th and early 20th centuries), employment and labor share of income held up very well. For developed countries in the period 1970-2007, Autor and Salomons found a mixed picture, with robust employment but declining labor share after 1980. With respect to robotics specifically, Graetz and Michaels did not find declines in employment or labor share in the European Union, but Acemoglu and Restrepo found some decline in both employment and wages in the U.S.

It seems fair to say that the jury is still out on the effects of automation on the labor force. It may be that automation has no inevitable effect, but that it depends on how we as a society choose to deal with it. We shouldn’t assume a world of mass unemployment and widespread government dependency on the basis of recent, preliminary results from one country. Authors such as Thomas Friedman, who are more optimistic than Martin Ford about the long-run effects of new technologies, have yet to be proved wrong.


Viking Economics (part 3)

June 28, 2017

Previous | Next

Even the wealthiest, most economically developed countries in the world face serious challenges in the years ahead. An important question for the future is whether a more egalitarian social system is an advantage in dealing with these challenges. If so, that makes the Nordic model even more relevant to current policy discussions.

Globalization

In much of the developed world, globalization has benefited capital more than labor, as global corporations profit by offshoring work to cheaper sources of labor. Nordic countries have a long history of global trade. Countries like Norway “lacked the extensive land, abundant resources, and large population that enabled countries like the United States and Germany to generate robust, internally driven economies.” But Nordic countries also have a strong commitment to high employment and good wages. Can they sustain that in the global economy?

One way of reconciling an openness to foreign trade with a desire to protect domestic workers is a policy of “flexicurity,” a Dutch concept that has become central to economic policy in Denmark. “The Danes changed the social contract between the state and the workforce. Instead of guaranteeing workers their existing jobs, the government would guarantee workers ongoing support and retraining so they could get new jobs.” By providing the universal services of education, training and a strong social safety net, Nordic countries help their workers cope with a world of enhanced foreign competition.

Although immigration is a controversial issue almost everywhere, Nordic countries have also had the confidence to extend economic assistance to newcomers. Both Norway and Sweden have about a 14 percent foreign-born population, even a little higher than the US’s 13 percent. Norway will support immigrants for a year while they learn the language and culture and receive job training. “Norway is ranked number one among the twenty-seven richest countries for its policies on migration: acceptance of asylum-seekers and refugees, open borders to immigrants and students from developing countries, and friendly integration practices.”

That is not to say that conflict between immigrants and natives is nonexistent. Sweden has experienced youth riots in immigrant neighborhoods, especially during the period after the mid-1980s, when it was cutting public spending and allowing inequality to grow. In general, however, Lakey believes the Nordic model goes a long way to reduce social conflict. While American-style inequality “institutionalizes scarcity,” making people of different races and ethnicities compete for too few opportunities, the Nordic model:

generously funds agencies and programs that assist people who otherwise might lack opportunity. It seeks out barriers to advancement, such as burdens of childcare and dependent elders, and tries to alleviate those to free everyone to move ahead. By universalizing such programs, as well as health care, vacations, access to public transportation, and other enhancements that otherwise can become racialized for disadvantaged populations, the model carefully avoids setting categories of people against each other.

Of course, doing all these things is costly. But who can calculate the social and personal costs of our failure to do them?

Climate change

As I have argued elsewhere, environmental issues highlight the tension between private gain and public cost. Fossil fuels provide profits for producers and cheap energy for consumers, but their market success depends on not factoring in the social and environmental costs of climate change and other environmental damage. Renewable energy will become cheaper and more profitable over time, but the government may need to put a big thumb on the scale to discourage what is publicly dangerous and encourage what is publicly good as quickly as possible.

Because Nordic countries are more receptive to market interventions for the public good, they have generally been leaders in national and international action on climate change. Sweden, Denmark and Norway were among the first to impose taxes on carbon emissions, back in 1991. Denmark has been a world leader in wind power, because of national policies like incentives to form local wind energy co-ops. In 2013, Sweden was already getting over half of its energy from renewable sources, compared to an average of 15% in the European Union and even less in the U.S.

Norway is in an awkward position on climate change, because oil accounts for almost half of its exports. How much of the Arctic oil reserves it can actually develop without unacceptable environmental damage is a vital but unresolved question. On the other hand, Norway’s large public pension fund has divested from coal, as well as from Canadian tar sands oil. Norway also doubled its carbon tax in 2012 because the government wasn’t satisfied with the country’s rate of emission reductions.

Automation

Lakey does not discuss the potential impact of automation on employment, but it is a challenge that is receiving more and more attention. I recently reviewed Martin Ford’s Rise of the Robots, which warns of a “jobless future” for millions of workers whose jobs are vulnerable to automation. Ford and others envision an expansion of public welfare programs to support the jobless multitudes.

Lakey has described Nordic countries not as welfare states, but as “universal service states.” They place a strong emphasis on helping people to become productive citizens with good jobs. Does that make them more or less prepared to cope with a more automated economy?

In Parts 2 and 3 of my discussion of Ford’s book, I described my somewhat different vision of the future, emphasizing the transformation of work rather than just the elimination of jobs. I have no doubt that robots will take over many tasks that they can do more efficiently than humans. But as in the transition from farming to manufacturing in an earlier time, I would hope to see human labor shifted to new frontiers of economic activity, especially in the area of skilled personalized services. I would also like to see the extension of the twentieth-century trend of shortening the typical work week, which would have the effect of spreading the available work to more people. As the twentieth-century experience showed, fewer hours is compatible with high pay as long as workers have the skills and the technological support to achieve high productivity. That in turn depends on the development of human capital, which requires broad access to education, health care and other human services, industries that both create jobs and equip people to get jobs. Since the development of human potential is a public good that not every family is able to pay for, a strong public role in such areas as health insurance is called for. There is also a role for non-market work–labors of love if you like–which can flourish when people have the leisure to balance their work and family responsibilities and participate in volunteer work.

Although I hadn’t read Viking Economics when I developed these ideas, the Nordic model seems relevant to everything on my list. The same “flexicurity” policies that reduce fears of globalization can also reduce fears of automation. If you lose a job, you can expect help in finding and qualifying for a new one. The Nordic work week is already shorter than ours. The universal services model is more conducive to the development of human capital, and citizens are already accustomed to paying high taxes to support it. Finally, “Thanks to an economic model that fosters work/life balance, people have abundant time to volunteer in the community.” It’s a way of life that compares favorably to the American system, where workers cling to technologically and environmentally obsolete jobs like coal mining because they expect little help to become something new. We can do better.