What can history teach us about technology and jobs?

What can history teach us about technology and jobs?

We can learn from past shifts in employment to improve the future of work.

This is not the first time the world has experienced significant shifts in employment due to new technology. History tells us that in the long run technology is a net creator of jobs. But with artificial intelligence (AI) and automation’s rapid advances, could this time be different?

For the New World of Work podcast, Peter Gumbel, senior editor at the McKinsey Global Institute, spoke with Susan Lund, partner at the McKinsey Global Institute and coauthor of the MGI report Jobs lost, jobs gained: Workforce transitions in a time of automation, and Richard N. Cooper, the Maurits C. Boas Professor of International Economics at Harvard University, about what we can learn from past employment transitions, first out of agriculture and more recently out of manufacturing.

Podcast transcript

What can history teach us about technology and jobs

Peter Gumbel: Welcome to the latest in our podcast series on the new world of work. This is Peter Gumbel, from the McKinsey Global Institute, and today we’re going to be talking about what history can teach us about issues of technology and employment. And here to discuss that are Richard Cooper, who is the Maurits C. Boas Professor of International Economics at Harvard University, and Susan Lund, who is a partner at the McKinsey Global Institute, based in Washington, DC. Thank you both for being here.

I will, first of all, ask Professor Cooper what the lessons really are, because it looks at least from a very superficial point of view that technology has tended to create more jobs than it destroys. Is that true? And if that’s the case, why is that?

Richard Cooper: Well first, the historical perspective. This issue, which currently preoccupies people, goes back a long way. Technology has been changing for at least seven centuries, since the horse collar in Europe. It reached a new stage with the industrial revolution, starting in the 19th century. And roughly once a generation, we have a near panic by some people because technology is destroying jobs. And it’s true that new technology often destroys existing jobs, but it also creates many new possibilities through several different channels. Think of cloth making back in the early 19th century or automobile making in the early 20th century. All kinds of supplementary industries, all kinds of new possibilities. And don’t forget demand. If goods become cheaper, speaking in general terms, people want more of them. So, there may be a much bigger demand for the output even though the productivity has gone up and, per unit output, fewer people are employed.

For all of those reasons, higher incomes, greater demand, supplementary activities to the activity that’s being focused on, we’ve discovered that total employment has increased over the years in spite of concerns, roughly once a generation, about the loss of jobs created by new technology.

Susan Lund: To put some numbers on this, we’ve looked at the productivity growth and employment growth over different time periods in a variety of different countries. And what we find is that since 1960, in the United States, for instance, both productivity and employment have grown in individual years 79 percent of the time. And in only 12 percent of the years did we see productivity growth with employment declines. And when you look over a longer time period, say three years, five years, out to ten years, you see that the number of times that employment actually falls while productivity goes down literally to zero, in the case of the US, when you look out at a ten-year period. Indeed, in a five-year period or a three-year period, 95 percent of the time you see productivity and employment growing for all the reasons that Professor Cooper just explained.

We see the same pattern in other countries as well. In China, for instance, when you look at individual years, you see employment and productivity both growing in 77 percent of the individual years since 1960, but in 98 percent of the ten-year rolling periods. In Germany and Sweden, we see the same pattern, albeit somewhat lower due to less turnover and other rigidities in their labor markets. But it’s very clear from the evidence that, in fact, as productivity grows, you don’t see fewer jobs; you see more jobs.

Peter Gumbel: Susan, in the report that you have just put out, called Jobs lost, jobs gained: Workforce transitions in a time of automation, you cite an interesting case of the Ford Model T as being an example of how this actually plays out in practice in the workplace.

Susan Lund: The Ford Model T is a good example of what Professor Cooper described. Over a six-year period, the number of Model Ts produced per worker tripled from eight to 21. So, that’s productivity. But at the same time, the price of a Model T dropped by more than half, from $950 per car in 1909 to $440 in 1915. As a result, demand to purchase automobiles just soared. And so, rather than declining, employment in the automotive industry soared and the number of people employed went up. It was because consumers demanded more of the goods.

Peter Gumbel: OK. Professor Cooper, perhaps you’d like to jump in on that. So, how actually does it happen that autos create net jobs?

Richard Cooper: Well, back up a minute. Automobiles, the internal-combustion engine, destroyed a whole industry—the carriages and horses to pull them, right? There were lots of jobs lost as a result of this new technology, an internal-combustion engine on wheels. Lots of jobs lost. And, as Susan says, even within the automobile industry in the early days, productivity went way, way up, so output per worker went way up.

Why was there no net loss of jobs? Recall that Henry Ford paid very good wages to his people. That was partly for his benefit. He did not want the high turnover that manufacturing then had. He wanted to keep good workers. But he also said he wanted his workers to be able to buy his product. So, incomes of those employed rose; demand for the product, the Model T and other automobiles, also rose (Exhibit 1). And as Susan said, because of the increase in output, despite a large increase in productivity, employment in the auto industry went up. That’s not always the case. You can imagine employment in the industry in which the technology is improving rapidly going down or, as I mentioned with carriages, whole industries being destroyed.

A virtuous cycle of increases in demand, production, and income drives overall economic growth.

By the way, that did not take place at once. It took place over three decades. My grandfather still had work horses on his farm, as well as two tractors. So, it’s a gradual process. But think of the supplementary industries that were created by automobiles. Filling stations, automobile repair. And then, of course, the possibilities of living farther from work, enlarging your housing by moving out where land was cheaper than it was in the central city. And then vacations. With automobiles, you could go to many more spaces.

Peter Gumbel: So, it sounds like what you’re saying is that at least some, or even many, of the jobs that are created by technology are ones that perhaps you couldn’t even have imagined before the technology existed.

Richard Cooper: That is correct. And that’s a central problem with public discussion of this issue. The jobs that are lost are tangible. There are people in them. We can sympathize with the people who lose their jobs and so forth. There’s a human dimension. The jobs that are created or made possible have no one in them, and, in fact, we don’t even know exactly where they’re going to be until, you know, time goes on.

One example is the ski industry. Of course, skiing has been around for centuries, probably, but the ski industry is a post–World War II phenomenon in which areas of New Hampshire, Vermont, Maine, Colorado, Idaho, and Utah were opened up by the creation of the automobile and the rise of incomes that went with higher productivity. And so, there’s a whole new industry with ski instructors, people who maintain the ski slopes, snow makers, and the manufacturing of snow-making machines. None of that was foreseen until the industry started. And even after it started, it was not foreseen how big it would become. So, the recreational activities have been vastly opened up by the creation of automobiles and paved roads.

Peter Gumbel: Susan, over to you. In the report that the McKinsey Global Institute has just published, there’s also a case study of computers. And are you seeing the same things with computers that you also saw in the automotive industry?

Susan Lund: Yes. It’s a good example of the point that we’ve just made, which is that we can’t often foresee the new occupations that are going to be created from a technology. Consider the introduction of the personal computer and then, after it, the Internet, and now, mobile-phone-based computing and smartphones. In the report, we tallied up, at a very detailed level back to 1970 in the United States, the number of jobs that were destroyed. For example, we didn’t need as many typists. We didn’t need as many office-machine manufacturers. Secretaries used to take dictations of memos from executives, using shorthand, and then go type them up. That wasn’t needed with the advent of the personal computer. And increasingly, executives started writing their own memos.

We tallied up all the jobs destroyed in the US since 1980 as a result of the rise of personal computing and the Internet, and it’s about 3.5 million. We see declines in a lot of these occupations that were once large and today are very small or nonexistent.

But on the other side of the coin, we see millions more jobs created both directly for computer-hardware manufacturers and the input industries, like semiconductors. We see growth in computer-enabled industries. For instance, software developers, app developers, computer scientists employed in other industries, the whole software service industry. And then finally, we see computers have given rise to a whole range of occupations that couldn’t exist without them. Think about call centers and as a customer calling in to a call-center rep. If that person didn’t have a terminal in front of them to look up your account information, they really couldn’t tell you much. Right now, there are over three million call-center service reps in the US.

Historical precedents of technological disruption
Susan Lund explains what we can learn from past workforce transitions and how technology is a net job creator.

When we add up all the jobs created, we find that over 19 million jobs have been created as a result of the personal computer and Internet. We see a net gain of 15.8 million jobs in the US over the last few decades. And that’s about 10 percent of the civilian labor force today, is in an occupation or a job that’s a direct result of the introduction of this technology. Of course, this plays out over decades, but I think it’s important to remember in all the discussion of automation today that there will be growth of occupations and industries, that we can’t even imagine that over time will replace the work that’s being automated.

Peter Gumbel: OK. Both of you have sketched out very clearly this idea that sectors decline and others rise at the same time of these long-term shifts. But I’m wondering, Professor Cooper, what does this mean for people with different skills? I’m assuming that the new jobs require different types of skills from the old jobs that disappear.

Richard Cooper: Well, I think that’s right. And it can go either way. Susan mentioned, in her example, people—secretaries, we used to call them—who take dictation. And that was a specialized skill, reading what you wrote down and being able to type it up. Now, we can have people who don’t have the skills to take dictation that can produce a product in some respects even better than before. Similarly, with bar coding and scanning in grocery stores, or actually all shops, it used to be that at checkout counters, the people had to be able to make change, which means they were comfortable with numbers and quickly performing addition and subtraction in their heads. Now, we can hire people who don’t require numeric skills at checkout counters. This is a case where technology actually reduced the skill that was required for a particular class of jobs.

On the other hand, coding programs and apps requires a high degree of programming skill. And that requires new skills, which barely existed 50 years ago. We had programmable computers, but they were large, IBM-type machines. And programming has become much more common, but it is a technical skill. In medicine, now we can do things that were not possible before, but the medical equipment requires specialized skills, people who know how to use the medical equipment and how to repair it.

Training has to adapt to generate the skills that are required by the new technology. Technology changes the mix of skills. It can produce jobs that have lower skills than previously, and it can produce jobs that require new skills. Therefore, our educational system and workers have to learn the new skills in order to use the new technology, both to use it in application and to repair equipment that’s broken down.

Susan Lund: In our research in our new report, we quantify how much work could be automated for 800 occupations within countries, thereby reducing employment in that occupation, but then at different sources of potential labor demand. So, what jobs might be created in the years to 2030? And when we look at the net of those two forces, we see that some occupations may decline from today’s level, while others are going to grow.


There are two important findings, I would say. One is that millions of people are going to need to switch occupations (see interactive[JCS8] above[FC9] )[FC10] . There are too many people in jobs that are declining, like assembly-line workers, manufacturing, and retail cashiers. Any kind of work involving a lot of data collection and processing, machines can do very well, such as a mortgage officer assessing somebody’s credit risk.

It’s going to be a big transition. Globally, up to 375 million people may need to learn an entirely new occupation (Exhibit 2). So, that’s going to be a huge challenge. It means that people in mid-career, with children, mortgages, families, and financial obligations, are going to need to have training. And this training is not going to be measured in years. It’s not going to be feasible for many of these people, or most of them, to go back for a two-year degree, so we’ll need to rethink how can we take people mid-career on a large scale and help them learn new skills to find jobs in the growing occupations.

Globally, up to 375 million workers may need to swtich occupational categories.

The thing that we find when we look at which occupations are growing and declining is that in advanced economies, disproportionately the occupations that are declining are those that require only a secondary degree or less, and the jobs that are growing disproportionately require a four-year college degree or more. The technical and two-year degrees sit somewhere in the middle of that. But what it means is that we’re going to need to rethink education, that not everyone needs to go to college, but as I said it’s going to be very important to give people opportunities to learn technical skills to get the jobs that are going to be out there in a reasonable time frame.

Richard Cooper: I would add to that, though, it’s important to keep in mind the time dimension. These technical changes en masse are not going to take place instantaneously. They will be spread over quite a long period of time. We have the history of the introduction of electricity, we have the history of the introduction of the internal-combustion engine, and now, more recently, computers.

It takes quite a while before everyone, particularly businesses, adjust optimally to the new technology. There’s a long period of overlap between the outgoing industries and activities and the incoming industries and activities. In the meantime, each of us, all of us, are getting older and new people are coming into the labor force. We need, as we think about retraining the labor force, to keep in mind that people are entering and retiring from the labor force also one year at a time. And so, it’s partly reshaping the new labor force relative to the outgoing, old labor force. And that can be painful or it can be easy, depending not only what the technology is, but also the pace of introduction of the new technology. The pace is very important in the social impact and economic impact of any new technology.

Peter Gumbel: Just to pick up on this idea of retraining and, particularly for people who are in mid-career, do we see in history any precedents of this being done successfully?

Richard Cooper: We have many examples of 40-year-olds in the US entering the labor force successfully. And it involves retired military, because the military term, except for very senior officers, is typically 20 years for a career military officer. And so, every year, the US military turns out a number of people around the age of 40, maybe in their mid-low 40s or late 30s. And these people end up with jobs. Now, have they been trained for the jobs? Yes and no. Career military officers get lots of specialized training during the course of their military careers. Some of that specialized training is transferable more or less directly into the private sector. Think of nuclear engineers. Submariners who have trained in nuclear propulsion and then they go into the nuclear industry know a lot because they were trained in the military to do that sort of thing. But I’m sure there are other cases where a technology change is very hard on people in their mid-40s, because they either they find it difficult to retrain or we’re not set up institutionally to retrain such people.

Susan Lund: We also have successful examples from different countries in Europe today who were laid off and help them learn new skills and become reemployed quickly. And I would point to both Sweden, on one hand, and Germany. And they offer a similarly comprehensive suite of services to help displaced workers transition into new jobs, but they do it differently. In Sweden, it’s a private-sector-led model, and in Germany, it’s done through the government, through public labor agencies.

In Sweden, the worker-security councils are a system in which employers pay a small amount per worker into a private fund so that if the company downsizes and a worker is laid off, that individual goes to the worker-security council and they get a whole suite of services. And this is all privately run, but they get job retraining if they need it. They find out where are other job openings, what do they need to do to apply. It goes beyond simply providing income support to actually helping individuals find their next job.

Germany has had a very successful government-run system that operates in many ways very similarly. Reforms that were implemented in the early 2000s, the so-called Hartz reforms, have enabled Germany to reduce what was a relatively high unemployment rate, over 11 percent, now down to about 3.5 percent today.

In both cases, I think there’s a lesson that mid-career people can find new occupations and new jobs, but, if we see rapid automation, it’s going to take a more comprehensive and organized approach in many countries than we’ve seen so far.

Peter Gumbel: OK, well, we talked about skills, but now let’s just talk about wages. Perhaps I can ask you, Susan, what do we know about the effects of technology on wages?

Susan Lund: Well, we know that in history there have been times, including in the first industrial revolution, where we’ve seen technology produce large productivity gains, but wages have stagnated for a period. In the first half of the 1800s in the UK, you saw productivity continue to grow, but wages, for about 40 years, real wages were flat for a whole class of workers. In the US, more recently, since 1970, you’ve seen that real wages for goods-producing workers have stagnated even as productivity has soared. There’s no reason that all the gains from automation necessarily are going to benefit workers.

This is a puzzle that economists have been looking at. Why has the labor share of income been declining? When we look at our research at MGI, we just take today’s wages. What we see is that in the US, most job growth is coming at the high-wage jobs and low-wage jobs, and middle-wages jobs on net decline. So, the income polarization that’s been studied by other economists, notably David Autor, could in fact continue, our results would suggest.

Richard Cooper: Rather than talking about wages, I would rather talk about incomes, because incomes have a variety of sources besides the weekly or monthly wage. And I think one key going forward is who owns the robots, to be very concrete about it, as robots replace at least some workers in routine jobs. First, there’s an issue of who makes the robots. Of course, that may eventually be other robots that make them, but in the first instance it’s going to be wage earners who make the robots, but that requires new skills.

And second, as productivity goes up due to substitution of capital for labor, who owns the capital? That goes back to our institutional arrangements, how people are paid. Are firms “worker owned,” which can happen in in a variety of ways, and do workers benefit from the improved productivity generated by substituting capital for labor?

That’s a big issue and it varies from country to country. It differs according to the pension system and how the pensions are invested in fixed-interest securities or equities and so forth. It gets way beyond the workplace, but it ends up affecting people’s incomes.

Peter Gumbel: OK, well, we’ve talked about the rise of the automobile and how that has helped to generate the whole idea of family vacations, but of course one of the effects of technology over the years has been actually to increase the amount of leisure and to reduce working hours. Professor Cooper, perhaps you can talk us through that. How does that happen?

Richard Cooper: We should not measure our well-being just by looking at GDP or measured output per worker. We should look at the whole life. And of course, one of, as you suggest, one of the dramatic changes taking place over the last century has been a tremendous increase in leisure. That’s not just a decline of the working week from 60 hours 125 years ago to 37.5 hours in many industries today in the US, much lower in Europe, but also you have to think of paid holidays and paid vacations. You need to look at the whole year, not just the length of the working week.

I’ve long been in favor of a four-day work week, probably longer than eight hours, maybe nine hours, but have a long weekend. And we can organize society so that we can do that and still keep things functioning seven days a week. Made more possible today by computers. Western Europeans are well ahead of Americans in this growth of leisure, taking potential income in the form of more leisure, but Americans have benefited enormously over the last century from the same process.

Of course, that does not contribute directly to GDP, but it creates the possibility of many new activities, leisure-based activities, which do contribute to GDP.

Peter Gumbel: And the last question is perhaps the most complicated one, which is around automation. There’s a big debate that this time somehow the technologies that we see, artificial intelligence and advanced robotics, that they will somehow have a different impact on employment than technologies have in the past. Now, it’s a lively debate among technologists and economists, and it would be really interesting to get your take. Is it possible that this time it could be different?

Susan Lund: Well, it’s a complicated question. If you’re asking if this time is different in terms of the fact that some jobs will be destroyed and others created, no. Is it different in terms of the breadth of sectors of the economy that could be affected? No.

The one way that it could be different could be in the speed of the transition. There’s a lot of uncertainty today of how fast advances in automation and AI are going to take place, and even how fast companies will adopt them. So far, we see absolutely no evidence that companies are adopting the new technologies any faster than they ever have over the last 50 years or 60 years. So, despite all the advances that we keep reading about in the news about AI algorithms that can now diagnose pneumonia better than expert radiologists, they can win chess games, this is all great. But when you get down to the company level, companies are not moving any faster than ever at integrating new technologies into work processes.

It’s a huge investment for them, not only in terms of a capital investment, but also redesigning processes of how they make things, because a lot of the real gains from automation comes from rethinking how your business operates, not just applying technology to today’s processes. So far, we don’t think, in fact, that despite all the gains, that this time is going to be different in terms of the speed of adoption, but I would caveat that and say if, in fact, rapid advances in AI come about largely through machine learning and machines, you know, being able to progress themselves without human input, it could happen that businesses could pick up their pace of adoption.

And so, we may see a great workforce transition on the scale of what we saw from the agriculture society to the manufacturing society, and more recently from manufacturing into services, but on a more compressed time scale (Exhibit 3). Rather than many decades, it could be a faster transition. But again, today, I think it’s too early to see evidence that it, in fact, will be faster, but it’s one of the things I’m going to watch as we go forward.

History shows that technology has created large employment and sector shifts, but also creates new jobs.

Peter Gumbel: Professor Cooper, what’s your take on this?

Richard Cooper: Every period is different from its predecessors. So, of course, it will be different. The question is whether it will be different in a fundamental economic sense and social sense. And I agree with Susan. I am very skeptical about the pace of adoption being markedly more rapid than we’ve seen in past periods of technical change.

That is for a variety of reasons. Institutions change only slowly. It’s hard to turn an existing organization around, no matter what it is. And of course, leadership is old-fashioned, almost always by definition, and often it takes a generational change in leadership before you get the full benefit of any given technological change. My guess is that the next two decades will not, in this respect, be markedly different from what we’ve seen in the past. We will have big technical change, but it will be introduced gradually and not too fast for society to adjust to it with some pain, but that’s been the case for the last hundred years also.

What is the future of work? The New World of Work is a new podcast from the McKinsey Global Institute that explores how technologies like automation, robotics, and artificial intelligence are shaping how we work, where we work, and the skills and education we need to work. Featuring conversations with experts from the McKinsey Global Institute and thought leaders from the public and private sectors, this series will help business leaders, policy makers, and organizations understand what changes are afoot and how we can prepare today for a future that works.

To listen to more episodes from this podcast, subscribe to the New World of Work podcast on Google Play, iTunes, or Stitcher.

About the author(s)

Susan Lund is a partner of the McKinsey Global Institute and Richard N. Cooper is the Maurits C. Boas Professor of International Economics at Harvard University. Peter Gumbel is a senior editor at MGI.
More from the McKinsey Global Institute

An executive’s guide to AI

Article - McKinsey Quarterly

Why digital strategies fail

Article - McKinsey Quarterly

Strategy to beat the odds

Interactive - McKinsey Quarterly

Five Fifty: Strategy vs execution