人工智能、工作的未来以及 21 世纪社会契约面临的挑战，与詹姆斯-马尼卡（James Manyika）合著

WENDELL WALLACH: Welcome to the inaugural webcast of the new and exciting Carnegie Council Artificial Intelligence & Equality Initiative. I'm Wendell Wallach, a Carnegie-Uehiro Fellow, a scholar at Yale University's Interdisciplinary Center for Bioethics, and a senior advisor to The Hastings Center. I will be co-directing this project with Anja Kaspersen, a Carnegie Senior Fellow who recently stepped down as the director of disarmament affairs from the United Nations in Geneva. Through this Initiative we will explore whether artificial intelligence (AI) can be deployed in ways that enhance equality or whether it will exacerbate structural inequalities and create new forms of inequity.

This is a multifaceted subject. We will build, of course, on existing empirical research, but we are more interested in developing a comprehensive understanding of the ways emerging technologies are impacting equality than in answering specific wonkish questions. For example, whether a technology that automates jobs falls within a specific definition of what is or is not AI is far less important to us than the fact that AI systems are or will impact every aspect of modern life and are amplifying scientific discovery and the deployment of biotechnologies, nanotechnologies, and other forms of innovation.

When most people think about this subject area, the first thing they think of is technological unemployment. "Technological unemployment" is a term that was coined by the British economist John Maynard Keynes in 1930, and with it he wished to refer to the longstanding Luddite fear that each form of technology would automate and destroy more jobs than it creates. Over 200 years, even though that fear has come over and over again, this has not occurred. Nearly every technology has created more jobs than it destroys, and yet many people think it's different this time, different because artificial intelligence can replace cognitive skills and not just manual labor.

When I think of technological unemployment and who might be best to present this to our audience, the very first person who came to mind was James Manyika. James is a senior partner at McKinsey & Company, chairman of the McKinsey Global Institute, and a member of McKinsey's board of directors. Though I have met many of the leading scholars in this field, I know of no one who can talk about it as comprehensively or has thought about it as deeply as James. He and his team have produced a series of reports for McKinsey specifically about the future of work in the United States and in the European Union, but also about how technology is impacting the social contract, the explicit and tacit understandings between citizens and their institutions.

Without further ado, I would love to turn this over to you, James.

JAMES MANYIKA: Wendell, thank you very much for having me join you for this webcast. It's a real pleasure and a delight. I have enjoyed all the times we have had a chance to speak about these issues.

I thought I might start with my own brief background into AI. Many, many years ago I did my Ph.D. in AI and robotics at Oxford University, and I remember at the time we were starting to think about how new ways of developing AI might make a profound difference. We were working on neural networks, we were working on what was then at the time deep learning, reinforcement learning, and how you could use these neural networks to actually make profound progress. This was back in the early 1990s. Since then I have ended up doing what I do, where I spend most of my time looking at the intersection of these technologies and their impact on business and the economy.

AI itself has come a very long way. As you know, the term itself was coined in 1956 at a conference that John McCarthy and others had in Dartmouth, and they were trying to find a term that was different enough from one of the terms that was being used at the time called "cybernetics," so they thought to call this "artificial intelligence."

They went through a period of profound excitement in the early 1960s and so forth about the possibilities and that led to, as you may remember, films like 2001: A Space Odyssey and so forth. After that followed a period that was considered a period of AI "winter" or "winters," where not a great deal happened.

Then progress picked up again starting in the late 1980s and the early 1990s with the work of Judea Pearl and others who had started to use Bayesian systems to think about how do you do learning differently, and work by people like Geoff Hinton and others trying to use neural networks as a way to teach these systems to learn. So they made profound progress, and now here we are, and everybody is very excited about the possibilities. We have had announcements coming up fast and often about the breakthroughs.

But one of the things that has been quite striking has been this question of: How does this all intersect with the economy, and what are the impacts on the economy? In thinking about the impacts on the economy, first and foremost the progress has been astounding in terms of the applications we all get to use, the technologies we all live with now. In fact, AI is more pervasive than we realize, everything from the recommendation engines—we talk to our machines, we talk to our phones. It's already here.

So the impacts for us in terms of utility but even in the impacts for business are quite profound. We have done research, Wendell, where we have looked at what could the impacts be on business, and the impacts are profound in terms of increased productivity and utility, cutting across all sectors of the economy, all kinds of business, in sales and marketing, in operations; it is pretty profound. So the impact for business is relatively clear, which is it's quite profound and impactful.

When you look at the economy as a whole, the impact is also expected to be quite significant, partly because as you know the way economies grow is largely a function of increases in labor supply and growth in productivity. In fact, this is the kind of work that people like Bob Solow and others won their Nobel Prizes for, where essentially what you want is to have productivity growth coupled with increases in labor supply. When the two work together, they give you gross domestic product (GDP) economic growth.

In the time we are at now, with the population starting to age, the key ingredient of what is going to drive economic growth going forward is going to be productivity, and productivity has been sluggish. Technologies in general, but AI in particular and all these technologies that drive automation, are in fact expected to be one of the key contributors to productivity growth at a time when we can no longer rely on labor supply growth because of aging populations and so forth. So the potential contributions to economic growth are quite profound.

Then you get to the big question, which is where you started: But what does this mean for work and employment? The fear about technological unemployment has been around for a very long time. Here let me go into some detail because we have been doing some fairly deep studies over the last several years to try to understand exactly what the impacts of these technologies on work would actually be.

I should say, by the way, we are not alone in doing this. Many, many others have done similar research. There is the Organisation for Economic Cooperation and Development (OECD), researchers at Stanford and the Massachusetts Institute of Technology, and a few others have done this work. Let me describe what we have done, which is in some ways analogous to what many have done.

We started by thinking about jobs not so much in terms of what is going to happen to a welder or what is going to happen to an accountant, but rather to start with the actual activities. If you think about work, the work you do, Wendell, what I do, or what any of us do, it typically is a combination of many, many tasks. So we took the task-level view, and on that we looked at something like 2,000-plus tasks and activities. There is a great data set called the O*NET data set and a few others where you can actually understand the constituent activities that make up any occupation. We then looked at all those tasks and tried to understand with the potential for technology how much of that could be automated as these technologies make progress.

Having looked at the potential rates of automation along with what those technologies could do, we came to a few conclusions. If you look at the activity level, in a country like the United States, 50 percent of the constituent activities that make up work could in principle be automated with technologies that have already been developed today and are probably on tap to be deployed in the next couple of decades. That is 50 percent of the activities. I should emphasize that this is activities and not jobs. Keep in mind again that jobs are made up of a combination of activities.

When you apply this back to actual occupations—and we did this using the Bureau of Labor Statistics data—what you then find is that there are only about 10 percent of occupations that have close to or more than 90 percent of their constituent activities that are automatable, and that is important to keep in mind. These are whole occupations where all the activities could in fact potentially be automated. That's about 10 percent of them. There is another 60 percent of occupations where about a third of their constituent activities can be automated.

What does this all mean? I think where it nets out is something as followed, and we actually wrote about this in one of our reports. We called it "Jobs gained, jobs lost, and jobs changed," and let me describe each of these.

I will start actually with jobs lost. It is the case that there will be occupations that will decline in terms of how many are required because as automation progresses we will have some job declines. That is hard to dispute. We are also going to have some jobs gained. You pointed out that throughout the history of technology one of the things that happens is that as technology advances the demand for some occupations grows, and new other occupations are created.

A good example of this, by the way, historically is if you look at what happened with bank tellers. It's a wonderful historical example. If you looked back in 1970, you might have said, "Well, with the advent of ATMs we are no longer going to need any more bank tellers," because it was the case if you looked at 1970 or thereabouts that something like the vast majority of activities that bank tellers did was mostly to count your money, either to take it from you or to give it back to you. You would not be wrong to think, Well, if all or most of that is being done by machines, presumably we are going to need fewer bank tellers.

What actually happened was that between 1970 or thereabouts up until the early 2000s the number of bank tellers actually grew, and the reason they grew is because the demand for retail banking went up, and we opened more branch banks, and so forth. So the demand for certain activities, even as they change, can actually lead to job gains.

So there will be jobs gained, and there will be jobs lost. There will also be jobs changed. The jobs-changed part is quite important because as I mentioned earlier we found that something like 60 percent of occupations have some portion—about a third—of their activities that can be automated. So, as portions of people's activities get automated, what remains changes in terms of what kind of activity it is.

Again, the bank teller example is quite illustrative. While the bank teller may have used 80 percent of their time in 1970 counting money, today they don't do that very much. They actually spend their time doing other things because the occupation itself has changed.

When you put all these pieces together—jobs gained, jobs lost, and jobs changed—do we end up with technological unemployment? We found that at least for the next several decades the answer is no. In fact, there will be more jobs gained than there will be jobs lost, and an important part of what we need to think about are the implications of that still.

If you ask me, Wendell, do I worry about a jobless future over the next several decades, the answer is no. Not that we should ever stop worrying about that—we shouldn't—but is that at the top of the list? No, it isn't.

So what is at the top of the list of things that I worry about when it comes to AI and work? There are at least four things, and let me describe what they are:

The first one is the fact that because of the jobs-changed nature of what I just described and what we just talked about, we are going to need to develop an incredible ability to adapt skills. That's why the conversation about re-skilling and skills adaptation is absolutely critical. Why do re-skilling and skills and education matter? They matter for a couple of reasons. One reason is because as some jobs decline and others grow, workers are going to need to be able to have the skills for the jobs that are growing and the jobs that are changing. So the skills and education is a critically important part of the adaptation that needs to happen. That's number one.

The second thing that is going to be important is: How do we manage these work and occupation transitions? As we know, some occupations are going to be declining and some are going to be growing, so how do we enable or make it easy for those transitions as people move from one occupation to another to work as smoothly?

This, by the way, is one of the things that I think the world endures well through globalization. We didn't manage the workforce adaptation transitions very well. The scale of that in the case of globalization was not anywhere close to as substantial as it will be with automation. So number two is to think about these work transitions and how we enable and make it easier for workers to do that.

The third thing that I worry about are the implications on wages. While I said at the beginning that there will be enough work, at least for the next several decades, and that's not what I worry about, I do worry about the wage implications, however, and I worry about that for a couple of reasons: One is that we know that many of the growing occupations as well as the very hard-to-automate activities—think a teacher, think a child care minder, think those kinds of activities—have tended to be occupations that at least the current labor market dynamics don't reward very well. Those have tended to be the lower-wage occupations, unlike many of the occupations that are easy to automate; think accounting, for example. Those have tended to be many of the middle-wage jobs. So the effect, even if there is enough work, of the shift to these growing activities like care work and also to these occupations that are harder to automate, could in fact have a downward draft and impact on wages, and I worry about that.

I also worry about the impact on wages when you think about even when we are complementing work because one of the things that happens even if you don't replace a job but you complement some portion of it, two things can happen. In some cases that job becomes even more valuable because complementing it with technology makes it incredibly more productive and incredibly more valuable, so the person doing that work gains a lot in impacts and wages and so forth. But at the other end of the spectrum, it can also happen where the technology complements the value-adding, if you like, part of that occupation, and what remains behind is much more commoditized, even though a human is still doing it. That could have the effect of also depressing those wages.

So this question about how we work through our wage-impact dynamics even when there is enough work is something we should think about, especially at a time when we are all quite concerned about inequality and so forth.

The fourth thing that I worry about and that we need to think about is how we actually redesign work and the workplace itself, because we know that whenever people are working alongside machines and machines become part of our work environment, how that work is organized is fundamentally important. I think we have all come to learn through this COVID-19 crisis another reason to think about how we organize for work, which is for health and safety. In a world in which we are going to have to rethink the workplace, how work flows happen, not just because of technology, at least for the moment anyway but also because of COVID-19, the redesigning of work and the workplace is also an important thing for us to think about.

These are all things to think about in this transition. So while I don't worry as much about do we face a jobless future, I think these four things are things we are going to have to work on if this is going to end up in a wonderful place.

That's a quick snapshot on this work question. I am sure there will be questions about that.

I should mention two other things we have been thinking a lot about when it comes to the future of work. One is, how all of this plays out in particular places and geographies. We have done some deep work looking at the United States and Europe.

As part of this work, I didn't know before but now I do how many counties there are in America. I think it's something like 3,149, because we went county by country trying to understand the labor market dynamics in each of those counties in terms of how labor markets are working today but also to understand what might be the impacts of technology and other forces going forward. We did something similar in Europe as well just to understand the dynamics. I will give you some interesting highlights from the U.S. work.

What was fascinating about that is that even at a time—this is before COVID-19, of course—when we were celebrating how jobs were growing since the last recession of 2008 and how job growth was great, what was striking about the job growth, though, is how concentrated it was in this sense. When you go county by county, we suddenly realized that counties that make up roughly about a third of the U.S. population were generating slightly more than two-thirds of the job growth, and in fact about 25 cities and a few dynamic growth hubs—some of them were small counties—were generating most of the growth.

So if you are in this roughly third of the country, you would have seen quite profound and impressive job growth. If you were in the second-third group, you saw mostly flat job growth. If you were in the other third, you still saw job declines, and we called some of these counties "archetypes of distressed Americana" and so forth.

The point about this is that not only are the impacts of how the labor markets work different for different people in terms of where they are on the socioeconomic scale, they are also different by geography and place. The way labor markets work both today and going forward is profoundly impacted by skill-wage profile and the dynamics of geography. Raj Chetty and others have also catalogued much of this work, looking at the differences by geography. We have done the same and also looked at what this might look like going forward. Similar dynamics turned out to be the case in Europe when you go through different micro-regions within Europe.

So these dynamics are profoundly important, which then takes us to a topic I think we will discuss in conversation, which is: What does any of this mean to how we might think about the social contract?

Let me stop at that point, Wendell. I'm sure this has probably already generated some questions and things we can get into in conversation.

WENDELL WALLACH: Great, great. We will definitely get to what this means for the social contract, but let me focus on a few things that you have said here.

So it is not the overall number of jobs being created that concerns you, it's not about work, but the impact on equality falls into three different areas, basically demographics, geography, and where you are in the workforce, who you are in the workforce. Who you are in the workforce, if I have it correctly—it's largely that we are hollowing out the middle-management jobs, and people are either being pushed to higher-paid tech jobs, presuming they have the skills or they are re-skilled to take those jobs, or they are being pushed down to service jobs where their overall income is lower. Have I got that correct?

JAMES MANYIKA: Yes. The slight modification I would make is it's not so much managerial middle jobs, but it's really middle-wage jobs. Some of those are managerial and some are not, but there is a whole range of middle-wage jobs that have been declining, and it's what economies and we have been calling "labor market polarization," where the top end does well and the bottom end does well in terms of employment but doesn't do as well in terms of wages and income.

WENDELL WALLACH: So we are definitely getting an exacerbation of inequality just in that fact alone, and we don't have the normal progression in wages that we had presumed, though it was never quite true.

JAMES MANYIKA: The progression in wages was true at a time. I remember a couple of years ago, back in 2016, we did a research project where we looked at wage progression in advanced economies, starting from the 1960s. You would have seen in that data set across most advanced economies that progressively each generation got better off for a long time. One of the things that we found in 2016 was that in the last decade or so—and that's why we called the report "Poorer than their parents?"—that's when we started to see wage stagnation, if you like. It is a very polarized profile, depending on where you are in the social income scale. So this wage stagnation question—never mind the inequality of it, but just the wage stagnation—is a profoundly important and relatively recent phenomenon, at least in many advanced economies.

WENDELL WALLACH: You also emphasize that wage stagnation or even the declining of jobs may be very specific to regions. In fact, that was one of the things I was most taken about by your reports, the ways in which you broke down what would happen in the United States county by county and region by region and the difference between what was going on in high-tech centers or agricultural centers or regions where the demographics were largely tilted toward the aging.

JAMES MANYIKA: It was quite striking. As you went county by county, there were a few patterns. Some of those where the growth was spectacular were either a few large cities—not all of them but some of them, because you had the Detroits and other places in there too that aren't doing as well. You also had some very dynamic small growth hubs. Some of those, for example, happened to be small local towns where fracking activity, for example, is going on. So if you were in that locale, the dynamism there was pretty high.

You also had some other dynamic small hubs where there was the location of a large data center or some economic activity that had located itself there, so the dynamism was quite high. There were all of these kinds of profiles and patterns. It's a little bit more than simply saying cities versus rural. You have to understand actually what were the economic dynamics going on in that particular locale to see the differences.

WENDELL WALLACH: Yes. The other thing that jumped out at me when I looked at your reports was the differences between the European Union and the United States. Maybe you can speak a little bit more about that, but I particularly noticed the factors on aging and migration patterns between the United States and Europe would actually be significant in terms of how this all plays out, or at least how we anticipate it to play out based on existing trends.

JAMES MANYIKA: That is correct. I have to say one of the things that was surprising—and many of us had not quite appreciated the extent of it, because we always think of the United States as a very mobile workforce where there is a lot of movement. That was historically true, but that mobility has declined. We looked, for example, at how many people were moving in a couple of ways, moving city, moving county, and moving state, and we found that had slowed down dramatically in the last couple of decades compared to what it used to be. In Europe dynamism was surprisingly trending up more so than it was in the United States, so on dynamism there was quite a difference.

In Europe, one of the other differences was that the nature of the dynamic hubs or micro-regions that were growing had a slightly different profile than the ones in the United States. They tended to be not so much high-tech but a few industrial manufacturing hubs because manufacturing has stayed relatively robust in parts of Europe so you had dynamism in those places. Then the aging profile was different. In fact, in Europe one of the striking conclusions that we had was that Europe may actually have a shortage of workers in the very near term in terms of the demand on labor that you could expect over the next decade or two, which is quite different than the profile you would see in the United States, partly because the demographic profiles are quite different.

The other thing we should perhaps talk about also, Wendell, is this issue of the social contract. I think it is quite an important part of this. In fact, the report itself is actually called "The social contract in the 21st century." As you know, the social contract is a deeply philosophical concept that has been around for many, many years, but we took a very economic view to it and examined this across 22 OECD countries. We were trying to understand at least how this had changed in the 21st century across these 22 countries. We were looking at it in three ways: how it had changed for individuals as workers, individuals as consumers or households that consume products and services, and third, from the point of view of individuals as savers who are anticipating a future. The results were quite interesting. I will summarize them quickly.

From a work standpoint, the story of the 21st century is a mixed one. It's great in terms of the number of jobs that have been created. It's interesting that despite even the recession of 2008—and of course, all of this is pre-COVID-19—job growth in these countries, including the United States, was unprecedented. So jobs were very, very high.

What was more mixed were the questions around: How did wages do in that period? Mixed. Things like work security and fragility had gone up in that period, and part of the reason the work and income fragility had gone up was because in most countries, including the United States, most of the net new jobs that had been created were of an alternative variety. What do I mean by that? "Alternative" in terms of either contingent work, part-time work, gig work, and those different varieties of work compared to, say, traditional full-time jobs. So that was the work story.

The consumption story of people as consumers was also very striking. It is actually a story of two halves in a way that has a profound impact on the social contract. The good half of that consumer story was that the cost of things like cars, white goods, electronics, and furniture had dropped dramatically over the last 20 years. In fact, they had become incredibly cheap in large part quite frankly due to globalization. These are highly tradeable, highly competed products, and in fact their costs had plummeted compared to inflation or whatever you want to index it to. So this is a wonderful story from a consumption standpoint. That's the good half of the consumption story.

The really challenging part of the consumption story across all these countries was how a few basics like housing, education, and in some countries health care and, depending on where you are, transportation, the costs of those things had gone up dramatically, far, far exceeding inflation or consumer price indices. In fact, they had gone up so much that for a few in the bottom half of the income scale, half the income deciles, these have come to consume huge portions of your income, in some cases the majority of it. So the story was about how the cost of housing and education and health care—they vary a little bit which country you're in; in the United States the big ones were housing, education, and health care—had gone up spectacularly, far, far exceeding income gains or growth for many, many households. This was a very hard story. It was more dramatic than we expected.

Then, if you look at individuals as savers, this was also mostly a tough story, because what happened is that many more people were in debt, many more people were saving a lot less. In fact, on average across all these 22 countries—it obviously varies from country to country—you could say that in all the different ways that people save, whether it's their own savings or pensions or any of the mechanisms by which people save for the future, the average person had savings that would cover about 10 years of retirement, when in fact most people are going to be retired for an average of about 20 years, so there was half there for retirement and for the future.

The thing about savings and the levels of debt by the way was interesting because what we found was that generally levels of debt had gone up for the bottom half of the income scale. The size of their debt had gone up, but the profile also of who was in debt had changed.

For example, in the United States we found, surprisingly, that the people who were indebted were younger and also often educated, which is different from what you might have seen in previous eras. This was quite striking.

And across all these, Wendell, there was a sense that a lot of institutions had actually retreated, as it were, so the kinds of supports that might have been provided by either governments or institutions of one sort or another, private or whatever the case, were becoming less and less in most countries. So this is the way in which some of the social contract had changed.

Just to summarize, if you looked at the 21st century so far, across all the OECD countries we looked at, there was a group of about 150 million people for whom this had turned out very well, and if you were in that 150 million your typical profile was that you were relatively well educated, high-wage, working certain sectors, and living in certain locales, this was actually great; whereas if you were in the other roughly 500 million where you could have been characterized by the fact that you were less educated, low-wage, lived in particular geographies, often female, very often a minority, and so on, this wasn't great.

WENDELL WALLACH: Right. It's clear if you get into the weeds of it, the role of citizens as wage earners, as consumers, as savers, those are all intricately bound up with each other and affect each other.

What I did notice about that report was that it was written before COVID-19. Interestingly, not all those reports were written before COVID-19. You took COVID-19 into account in considering the future of work in the European Union. That just tells me that you and your team are trying to think through what is the impact of COVID-19 on trends that you saw up to COVID-19 and trends that you anticipate we will witness over the next few years, so I would love for you to share some of your thoughts on that.

JAMES MANYIKA: We are starting to get our arms around the impacts of COVID-19 on labor markets. Here are some of the highlights, and we have actually started to write about some of this, Wendell. The patterns for Europe are quite similar, but I remember off the top of my head the numbers for the United States, so I will describe those. The European ones are directionally similar, although slightly better.

If you take the United States, something like about a third of the workers in the United States are economically vulnerable to COVID-19 work impacts. What do I mean by "economically vulnerable?" This third consists of people who have either already been laid off or have been furloughed or are experiencing reduced work hours or reduced wages for some combination of reasons. That's a much larger number than the unemployment data suggests, but that's vulnerability. That vulnerability is a function either because of the shutdowns we have had so that people can't go to work or because demand in that sector has dropped dramatically—think hospitality, for example, which has been very, very hard hit, or think about some parts of retail, for example.

So that's roughly a third. That is a very large number. And when you look at that third in the United States, close to 80 percent of it—I think the number is something like 77 percent precisely—of those people have tended to be people who are in low-wage jobs. In other words, these are people who earn less than $40,000 a year, which is quite striking, the vast majority of them, and also they are disproportionately people of color because of the occupation profiles of who has typically been in what occupations in what sectors, so they tend to be disproportionate to people of color, which is quite striking.

The other thing that is also striking, Wendell, if you read the popular press, you would think everybody can work from home. Well, that's not true. In fact, when you look at it, various people have tried to estimate who can actually do their work from home, can their work be done remotely and from home? Our estimate suggests it's about 20 percent. I have seen estimates as high as 33 percent or 30 percent, but no one is saying 100 percent. So even if you take the upper end of that and say, "Well, a third of the people can work from home," there is still another two-thirds who in order to do work have to show up somewhere in some physical place.

If you read the news, you would think everybody is spending time on Zoom like we are and working this way. That's actually a minority of people, so we have to think about these work effects which are quite profound. And we see similar patterns in Europe too. The numbers are obviously slightly different because of the mix of the economy and the labor markets, but this is what we are seeing as we go through COVID-19.

WENDELL WALLACH: COVID-19 seems to have speeded up a number of trends that existed beforehand. One of the trends that jumps to mind for me is that the percentage of productivity gains that went to wages versus going to the owners of capital, which really includes any of us who own stocks, but obviously it's disproportionate depending on who you are in the digital economy. The percentage going to wages has been decreasing, and therefore for the owners of capital there is a clear-cut value in not only getting rid of worker headaches or ongoing benefits to workers, but also that all productivity gains then go to increasing productivity gains, profits as they can be shown to stockholders. I wonder if this has come up at all in your reflections as far as COVID-19 goes and whether this is creating an opportunity to automate jobs more quickly than might have been otherwise.

JAMES MANYIKA: It is the case that we have seen a lot of the trends accelerate. In some ways, I think of COVID-19 as having done two things when it comes to labor markets: one is to perhaps accelerate the trends, and the other is also to more expose the trends that were already not so much in everybody's attention.

What has been accelerated? Clearly automation. If you hear what most business executives are talking about, there is a clear trend—we did a survey, for example, to try to understand how executives are thinking about automation and so forth, and one of the things that came up is an inclination to accelerate automation.

We have also seen an inclination to accelerate more varieties of work. Think about how many of us are living today, which is we now have things delivered to our houses and so forth because we can't go out, so we have seen some jobs related to delivery and logistics also start to grow in a much more accelerated way than they might have before. So there has been an acceleration.

But I want to come back to your capital-versus-labor question. I think this question should always be discussed in two parts, Wendell, because there is clearly one part which says that of course when there is more technology and more automation the owners of capital tend to do better, and the returns to capital go up if you happen to own assets that relate to that capital. That is certainly the case, and we have seen that in the stock market appreciations and all of that. So there might be something that is not so good about that.

But there is another part of this capital-labor question. In fact, we wrote a paper on the labor share of income, where the capital-labor question has been changing for a very long time. Think about a factory in 1900 versus a factory today. In a factory in 1900 almost all the inputs into that factory to get an output of anything were almost all labor, whereas if you take a factory today, yes, some of the inputs of labor, but some of the inputs are capital. In other words, the factor inputs we need to get a unit of economic output, that mix has been changing over time, mostly due to technological progress, and we do that because it drives productivity, and we need productivity growth.

So the question is: In a world in which the inputs we need to get our economic output are shifting in that way, how do we make sure people can still participate in the gains of that output? We have seen, for example, companies that are very high in intangibles and high in capital are the ones doing well.

For most people, the way they participate in the economy is primarily through their labor, and yet if the economy is generating output based on capital and intangibles, how do people then participate in those gains? So you have started to see people talk about, should we think about these issues of predistribution and other ways to have people participate, more stock ownership, and more asset ownership, because if we can no longer participate purely through labor or the labor portion is declining, how else can people participate in the gains of the economy? This is quite a profound question as the economy progresses because we certainly are not going to use less technology going forward, so we are going to have to think about ways to grapple with that.

WENDELL WALLACH: Clearly you have raised so many different topics here. We could dive into them, and they beg so many secondary and tertiary questions that we are not going to get to in this hour, but I did want to turn to one other subject before turning this over for some Q&A with those who have joined us today, and that is the Global South in particular.

We have talked already about in the European Union and in America it is not a question of overall job numbers necessarily or productivity gains, but it is an exacerbation of the polarization and who is affected and who will not be affected by the impact of technological deployments.

But now we have this question about: What is that going to mean for the poor nations, the indigenous communities, the marginalized communities that are not necessarily in Europe or America? Will that unevenness be so strong that, for example, large portions of Africa might be devastated? Certainly there are countries in Africa that have smart cities and that are leapfrogging ahead, but there is this broader concern about what is going to be the international impact here on underserved populations.

JAMES MANYIKA: I'm glad you asked this question, Wendell. It's an important one. Again, I find that it's a mixed story. Let me start with the positive and then the challenging parts of the question for the Global South.

The positive story is as follows. It is the case that inequality between the Global North and the Global South has narrowed in the last 20 years. It has, and that is largely a story of more people coming out of poverty in China, in India, and even in Africa. For example, if you look at the people who are now considered in the consuming class, meaning that they earn enough above subsistence in order to purchase consumer goods—cars and those kinds of things—that number is growing even in Africa, India, and China. In that sense, inequality between nations, between the Global North and Global South has improved, and I think that should be celebrated. It is largely a function of those countries plugging into the global economy, becoming part of the global economy, and so forth. That's a good story.

There are some challenges, however, as you pointed out. Even in those places, that has not lifted everybody, depending on which part of those countries you're in, which part of those regions you're in, and even which countries. We know, for example, if you look at developing countries, many of them that are within the economic ecosystem of large successful countries have tended to do well. Think of a Cambodia and a Vietnam, which are now part of the Chinese ecosystem in value chains and supply chains. They have tended to do well. Think of Mexico and others who are part of the North. Countries that are around these larger economies that are growing and get plugged into those supply and value chains have tended to do well.

Countries that are not part of that and in fact are quite isolated have done terribly, and you worry about: What is going to be the path for those countries and those regions? Think of countries like Bangladesh.

Similarly, one of the things you worry about from a technological standpoint is that it used to be that the story for the path for how poor countries developed or grew—and you could even argue China itself is a version of this story—is that those countries typically took the path of industrialization. They would become a cheap source of labor for some other country overseas somewhere, and that was almost their pathway into the global economy.

Well, that pathway is not as available as it used to be as a way to get into the global economy, partly because of technology. There is less and less offshoring where people are chasing cheap labor these days. It still happens, but there isn't as much of it today as you might have seen in 1995 or 2000, for example. People do that locally or use technology to do that—3D printing and so forth. You have seen all the articles about "reshoring." So there is a little bit less of that.

There is also less of that available as I said if you are not plugged into other people's global supply chains. In fact, this is something that economists like Dani Rodrik and others have been writing about recently, which is this idea of "premature deindustrialization," where you could industrialize and run on that for quite a while if you are a poor country to get into the global system, but it looks like the point where that stops is happening earlier and earlier, and in some countries it is not available to them to begin with.

This raises a very profound question for those particularly poor countries and in fact the poorest regions of the Global South: What happens to them? In an ironic way it is not dissimilar to the question about particular micro-regions or counties in America, who are not part of the amazing story that you see at the national level but are actually in these small places. This is a version of that playing out globally, where these micro-regions, micro-countries, and in some cases large countries are just left behind.

WENDELL WALLACH: Maybe we will get a few minutes to come back to that, but in the meantime I would like to see if there are any questions from those who are listening in today.

I see something here from Kenneth Miller: "Was there any attempt to categorize the 2,000 tasks by impact on job satisfaction, and how replacing certain tasks may increase or decrease satisfaction?"

I think is also more broadly about, well, if we are polarizing, who is advantaged and who is disadvantaged, are we also polarizing satisfaction or not?

JAMES MANYIKA: Thank you for that question, Kenneth. It is a very important question.

We didn't look at that question, but I know others have. If you look at the literature and the research on satisfaction, it gets at some of these questions about satisfaction and even dignity. We know that in some occupations and some of those tasks when the automation of tasks is enhancing the occupation and making that worker even more productive, doing even more, the literature suggests that the satisfaction goes up, whereas in occupations where the technology is mostly doing the value-added portions of the work and what is left over is much more mundane, satisfaction is potentially lower, and arguably even the sense of dignity in doing that work is quite different. That's why as we get excited about technology complementing work, even if it doesn't replace it, we should think about these questions of satisfaction because they don't all play out exactly the same way for every occupation.

WENDELL WALLACH: Our next question is from Christina Colclough. Before I read her question, let me just mention that she is going to be one of our future webcast speakers. Christina has a remarkable understanding of the impact of AI on workers.

She writes: "Great session! Regarding skills, as more and more workers are on precarious contracts, who will pay for the needed and continuous training for all workers?"

JAMES MANYIKA: Thank you, Christina. That is one of the toughest questions because one of the things that has happened, partly because of technology, particularly because of issues I am sure you know well, Christina, like work fissuring, for example, and all these alternative varieties of work, one of the things that happens is that those varieties of work increase, and it breaks the structure of how worker training is typically provided.

First of all, there was already less of it anyway, but those trends like fissuring and alternative varieties of work further lower the possibilities of work training. So then it becomes a real question as to who pays for that. You may already know this, but it is quite striking. As you look across the OECD countries, the United States provides the least as a percentage of GDP of government supports for on-the-job work training. But generally, most OECD countries have been decreasing anyway the amounts of support for on-the-job training.

One of the things we perhaps could think about is should there be new incentives put in place to increase what you might call "investments in human capital," much like we have incentives for companies to invest in capital capital—tax write-offs and research and development tax credits, etc. Should there be something similar for companies and employers to encourage them to invest in building human capital?

WENDELL WALLACH: This already gives you a taste of some of the exciting topics we're going to get involved with in greater depth with Christina in a few weeks.

Another question is from Cordell Green. Cordell is one of the advisors to the AI & Equality project. Cordell asks: "What about the value of personal services such as tourism? Will the labor value proposition increase?"

JAMES MANYIKA: Personal services in general is an interesting category because a lot of personal services from a technological standpoint have tended to be some of the harder things to automate, so they are likely to persist. The challenge with personal services is that there is such a wide variety of them, some where the labor markets value that work and some where they don't.

Tourism might be one of those where the proposition is valuable economically, and so people are prepared to pay for them, but we know of other personal services, if you like, that we may think are very important, but the labor markets don't value as much. Think about child care, for example. We all think it is one of the most humane things we can provide by humans, but labor markets for child care are not great. Think about teachers. We all think teachers are some of the most important people in our society, but we don't pay teachers very well. There seems to be in some of our labor market occupations where these very personal, very human activities that we value don't always get reflected in the way our labor markets value these things.

But I think tourism is one where we would value them, because there is a market for that, and people tend to want to pay for things when they're traveling and having tourism experiences.

WENDELL WALLACH: Presuming we can get out of hock and actually travel again soon.

Eugenio Garcia, who is a Brazilian diplomat, asks: "Do you see the prospect of AI and work being discussed by the United Nations to possibly adopt in the near future governance tools to mitigate risks in this area?"

JAMES MANYIKA: I think, Eugenio, you might be better placed to help us think through this. I think one of the things that is hard—and I know, Wendell, you are working on this question too—is that the question of governance for AI is very complicated because it spans not just these issues of AI and work, which may be some of the somewhat easier ones, but also issues of use and misuse, which gets into issues of national security and other things.

The governance questions for AI are quite complicated, and I think at the UN level the United Nations has tended not to be as involved or at least directly effective in labor markets, so even when it comes to the question of AI and work I don't know what role the United Nations could play in that, partly because that tends to operate at the level of national and, quite frankly, local labor market dynamics.

Where the United Nations could play a role perhaps is to think about principles, to think about embedding goals and aspirations when it comes to work and other things in things like the Sustainable Development Goals (SDGs) perhaps, and provide at least guiding principles. But when it comes to enforceable governance mechanisms, I am not so sure, but others who are much more savvy about how these diplomatic arenas work can perhaps develop a point of view.

Wendell, you know something about these things. You are thinking about them; I know that.

WENDELL WALLACH: I am, but rather than delve into them today, we have just a couple more minutes, so I am going to give you a closing question. Hopefully this is not the end of our involvement with you, but it is at least the end of today's webcast.

What are the ways in which you have been thinking that we can shape the future in regards to these issues around the future of work and perhaps more importantly the reinforcing or creation of even new elements of a social contract for the future? What would you suggest that we focus on in addition to some of the proposals you have already put forward?

JAMES MANYIKA: I would say we should tackle this with a sense of optimism. I think these technologies can have a profoundly positive impact on humans and on life in the 21st century. Think about the potential contributions to economic growth. We didn't even get to topics like the role that AI can play in advancing the SDGs, so there are a lot of positive things that could come out of this. I think the real question is how we make sure that everybody gets to participate and benefit from the profound impacts. So far we have not done too well in the 21st century. That's why we wanted to examine this question of the social contract.

But the good thing is that the ways our economies work is they don't work in their own right. We create the rules. We create the principles. Capitalism and market economics have always changed. If you look at the way capitalism was before the 1970s, it was quite different. So we can always revise the rules. We can always work to reform the rules. We can always work to make sure that these persistent tough challenges like inequality don't persist.

Those choices are up to us. Either as people who think about these things or as citizens or as people who participate, this should work the way we want while getting all the benefits that these technologies can provide. I think that is the work for us to all do.

WENDELL WALLACH: Unfortunately, our time has run out, even as I see some wonderful questions from colleagues such as Marek Havrda and Volha Litvinets and others.

But you have done a wonderful job of providing us with a broad canvas of the considerations, packed full of information and insights, and I can't tell you how pleased we are to have had you with us today.

I invite everyone else listening in to join us for future podcasts and webcasts, all of which will be made available on our website at the Carnegie Council.

JAMES MANYIKA: Wendell, thank you for having me. Much of what I described today I drew from the reports that we publish. By the way, these are all freely available. You can download them to get more details on some of the data and insights that I was summarizing here.

WENDELL WALLACH: Where should people go to download those reports?

JAMES MANYIKA: If they go to the McKinsey Global Institute website, or even if you go onto the Internet, simply search "MGI social contract," and it will probably take you to the report; if you search "MGI future of work in the United States," it will probably take you to the report; and "MGI future work in Europe," it will take you to the report. These are all freely downloadable, so you can get them.

WENDELL WALLACH: Great.

Thank you ever so much, James. I am so pleased that you could kick off this web series for us.

JAMES MANYIKA: Wendell, thank you for having me. It is always such a pleasure to talk with you. I always learn a lot when I speak with you. I am sorry we didn't get a chance to hear from you on some of these governance questions. I would have liked to hear you answer that question.

WENDELL WALLACH: Maybe we will get to them in other webcasts, but again, thank you.

JAMES MANYIKA: Thank you.