How artificial intelligence will change how we work

For a number of years, James Manyika led much-discussed research into the impact of AI and automation on jobs as chairman of the McKinsey Global Institute. A board and committee member across many nonprofit, academic, and government institutions, he brings both a research background in artificial intelligence and robotics and years spent working with businesses at McKinsey.

Since early this year, Manyika has been at Google as senior vice president for Technology and Society, where he focuses on how Google can have a positive impact in areas including AI, the future of work, and sustainability. We reached out to him for his latest views on how work will change. Here is a transcript of our conversation, edited for clarity:

What impact do you think AI and automation will have on jobs?

Automation has already had and will continue to have a big impact on jobs. But it's always important to think about what the nature of that impact actually is. The way I would summarize it is this notion of jobs lost, jobs gained, and jobs changed—because I think all those three things will happen. How much they happen is entirely a function of sectors, of job occupation categories, of the progress we'll see in AI, robotics, and other automation systems. But all three will happen.

Often there's particularly a concern around this notion of a 'jobless future,' that all jobs are going to disappear. I don't think that's going to happen because in most of the research that I've done and if you look at the most up-to-date research, they roughly conclude something similar: at least for the next two or three decades—which is the period over which some detailed work has and can be done—what most of those studies will show, including the ones that I was involved in, is that yes, there will be some jobs lost or that will decline. That's because some of the constituent activities in those jobs will become increasingly automatable.

There are some occupations where they have a majority of their constituent tasks that are easy to automate. The good news based on our research—and others come in similar ranges—is that those jobs that will be lost are probably closer to 10-ish percent of occupations. Some of those do involve a lot of people who work in those occupations. So if you're working in those occupations, you'll see those jobs start to decline in terms of the number of them.

But then you also have jobs gained. Jobs gained is where we will actually see job growth as a result of technology and automation. Where do those jobs gained come from? They come from a few different places. Some of those are occupations that simply just didn't exist, but will exist because of the impact of technology. This is, by the way, a historical thing that has always happened throughout history, that new jobs appear even as technology automates and changes jobs. So there are those jobs gained in that sense.

There are also jobs gained in the sense that the demand for many existing jobs will go up. This is kind of a weird thing. I can illustrate this with an historical example that may be useful. The historical example is think of the bank teller. The bank teller—if you had looked back in 1970—what does a bank teller do? A bank teller before the broad adoption of ATMs would basically spend their time counting money, either to take it from you or to give it back to you. But with cash machines or ATMs, that activity goes away. So at the time, there were actually quite a few academic papers that were predicting, 'Oh my goodness, bank tellers are gonna disappear.' Because why are we going to need bank tellers because there are now these machines?

But what's interesting, if you look across the United States from roughly the period of about 1970 to about when studies were done about 2012, the number of bank tellers actually grew in the US economy. How did that happen? What basically happened is the number of bank tellers per branch did decline, but the number of branch banks went up dramatically because it became easier to have many, many more branch banks. And the demand for branch banking went up dramatically. That's why the number of bank tellers actually went up.

So you'll see some of that also happen. And then the other source of new jobs growth is going to come from mostly demographic shifts. We know that demographic shifts are going to change the amount, for example, of care work in the economy. There's a whole set of categories of work—teachers and others that are much, much harder to automate—and demography and other things will actually increase the demand for those occupations. So you'll also see growth from that.

Then, technology drives productivity growth. We know that whenever economies grow, labor markets get tight and demand for work goes up. So you put all those pieces together and the jobs gained category could actually be quite substantial. Most analysis that has been done seems to point to the fact that the net of the jobs lost versus jobs gained comes out positive. How positive it is is a function of the country economy you're in, the sector you're in, and the geography that you're in. So it may be that in some sectors, the jobs lost part would be larger than the jobs gained. In other sectors, the jobs gained would be far, far, far larger. But at the national level, it looks as if it comes out positively.

The third part is the jobs changed category. In some ways, this may be the more important category actually because we know that many more jobs will change than will be lost or gained. What I mean by that is when we've done research—and others have done this too—where we looked at all the job categories that the BLS (Bureau of Labor Statistics) tracks, which is 800-plus occupations, something like 60% of those occupation categories have something like a third of their constituent tasks that will be automatable in the coming decades. That tells you that actually many more jobs will change than will be lost. If many more jobs are changing, then how do humans and workers adapt to working alongside powerful machines? That's where some of these questions about skills and so forth becomes important.

In quick summary, I think the impact of AI and automation is this notion of jobs lost, jobs gained, and jobs changed.

Journalist Kevin Roose and others have written with some alarm about the pandemic labor market tightness and remote work having accelerated the replacement of people with machines in some workplaces. Do you share that observation about those things being accelerants and do you share their alarm at the nature and speed of the changes?

There's no question that Covid has had a big impact on work. What was before Covid a small—in hindsight now—narrow experiment about remote work became an economy-wide experiment. We've learned a lot of things. But one of the things that gets missed—and at MGI we did some research on this, as others have done too—is if you actually look at all the occupations in the economy, we forget that the number of occupations where the work can be done remotely is anywhere between 20% and at the high end, maybe 30%, or a third. It's not 100%. We can debate whether it's the 20% range or the 30% range, but it's not 100% because what happens is there's still another two-thirds of people whose work requires that they actually show up someplace because it's in some physical place. So we should always keep that in mind that often when many of us are talking about the remote work phenomena, we're talking about ourselves. We're not everybody by any stretch.

But Covid did accelerate digitization. It did accelerate the possibilities of remote work in some ways for the better. It made it possible for people to work from places that they couldn't. One of the things I should have commented on where we're talking about the impact of automation is there's a geography to it. In much like the sectoral impacts are different, the geographic impacts are different. But one of the useful correctives that remote work does is it actually makes it easier for people to work from many different places. So that actually helps in that sense.

So I don't know if I'm as alarmed as perhaps those articles suggested, but it does suggest that to me there are some new adaptations you are going to have to do, quite frankly, to cope with the world where there is more remote work, there is more hybrid work, there are people working from different places. One of the things that I don't think we've evolved enough is how we adapt work workflows to all these changes from automation, digitization, remote work. We haven't done anywhere close to enough work on how organizations and companies adapt and workers adapt to that. Skills are only a part of it. I think there are much, much more things to be done.

Over the last few years the power of labor has increased in workplaces, partly because there was a tight labor market. Labor organization is showing a bit of a resurgence, among other things. What does the future of automation that you're imagining mean for the power of the individual worker or the collective worker?

There's no doubt in my mind that it's important that workers feel empowered. Because when workers are empowered, they're more innovative, they're more productive. And quite frankly, in all of these settings, having the engaged collaboration of workers and organizations, is very, very important. I had an interesting experience when I co-chaired the California Future of Work Commission and my co-chair was a labor leader. It was a commission that included business people, people from labor, from academia, from all kinds of settings. It was an amazing reminder for me over the two years we worked together about the interesting ideas and innovations that come when people collaborate.

In the case of automation, for example, I actually think we are more likely to build better systems that can complement work and workers if, in fact, workers are involved in thinking through the design of those systems collaboratively. One thing that is also important, by the way, and this is a long trend that we need to think about, is that we know that if you took a hundred-year view, labor income's the share of the economy has been declining. Labor as opposed to capital. That's not because anybody's chosen to do anything bad, by the way, it is simply because if you take a factory 100 years ago, most of the inputs were labor inputs to get any unit of output.

Today, if you're building a factory—it doesn't matter where in the world—your inputs are both capital, meaning equipment and machines, and people. So in that world where we have to think about the two inputs that are important to contribute to our output, we have to have some collaboration between the machines provided for through capital and labor and work. So workers need to work collaboratively with organizations if we're going to get this to the right outcome.

Going back to the outlook for automation—with jobs lost, jobs gained, and jobs changed—what should individuals and leaders of organizations be doing now about these changes?

It starts by recognizing what the changes are, and then what needs to be done about them. Even though I'm less worried about a jobless future, at least not for the next two or three decades per the jobs lost, jobs gained, jobs changed. But I am worried about a few things that we need to work on, which takes me to your question. So what are those things? The skills challenge is an important one. The reason the skills question is an important one goes back to jobs lost, jobs gained, jobs changed—as some occupations decline and others grow. There may be some shifts of workers from one activity to another, from declining occupations to growing ones. That often requires different skills, different capabilities, just to make the shift from the jobs lost to the jobs gained part of the equation.

The jobs changed part also has some imperatives that come with it, which is if people are shifting the mix of occupations, even as they work alongside machines, they probably need new skills. If I go back to my bank teller example, as a bank teller shifted from spending all the time counting money, counting money seemed to be the only skill that they needed. They needed to be able to do problem solving to help the customer think about financial planning and other things. Nowadays, they might still be called a bank teller, but what they're doing is very different. That probably required some skills and some training.

So for all these reasons, even as the net comes out fine in terms of jobs available, there's a lot of skills adaptation and learning that are going to be so important. The responsibility for that is with both organizations and companies as well as workers to empower themselves. That's what I'll call the skills imperative of the reskilling imperative.

The other thing that we need to think about is the wage effects. This one will be tough. We're currently going through a tight labor market, but if you take a much longer view of the next decade or two or so we'll need to think about wages and incomes. That's mostly because if you look at the mix of the jobs declining and the jobs growing, many of the occupations that are declining have been the source of middle-income jobs. And many of the jobs that are growing have tended to be ones that labor markets don't pay as well. In my view, we don't pay teachers what we should, we don't pay people who do care work what we should. So the mix shift itself will change and will have some wage effects. We're going need to think about how we make sure people are earning enough to do well and live and so forth.

The third challenge is that we are going to need to rethink the workplace itself, work flows themselves, how we organize how we do our work. Do things always need to work in the sequence that they used to work in? I don't know. Do workplaces need to look the same way as they used to? It's always amazing to me when you go to a large warehouse these days where people still work there, but there are also machines in the same environment doing picking and stocking and so forth. You can clearly see that the way people work alongside those machines is very different. The workflows are very different. So this question of how do we redesign work, the workplace workflows themselves, given that humans are going to be doing different parts of the workflows etc., I think that's fundamentally important and something that all organizations are going to have to think about.

How do we know the skills that people should be training for, where the gaps are for each individual, and how they get those skills? It's this systems question...

Because it's a system question, let me at least try to answer pieces of that system. When you look at the jobs, lost, jobs gained, jobs changed categories, the BLS and others have tracked what skills each of those occupations require. We know, for example, that the most labor-intensive activity in our economy is still physical activity. It still is actually a big, huge chunk of activity. But we know that that's going to go down. Even though it may still be a large piece, it will go down. We also know that cognitive skills and analytical skills are going to grow larger.

We actually published a paper and others have done this too, where you can actually see at that aggregate economy level, which skill categories are going to decline and which skill categories are going to go up in demand. You can look at it that way and think, if I'm a worker, I need to focus on those skills because they're going to be higher in demand. So there's that version of it. Then there's the very specific individual version where this is at the level of what do I need to learn day to day? That starts to look like very specific to your occupation, to your sector, to the arenas in which you're working.

But no doubt you know, many of those skills we have better credentials and not-so-great credentials for some of them. Credentials have been useful for things you can actually test for. But often tacit skills that you learn over time, etc., that you don't get in a college, are harder to credential for. We're going to have to draw up new credential systems to be able to pick up those skills.

The good thing is that there are now ways for people to demonstrate what they can do. Technology has actually helped with this. The fact that people have coding challenges where it doesn't matter whether I've gone to college and the fact that I can actually just code, that's great. One of the things that's been fascinating to me is if you look at the number of people who are now developers, which is a huge number and growing, and you look at research other people have done into where those people acquire those skills, something around a third of them had acquired those skills in college. Another third had acquired them through some self-help online. And another third had learned by just seeing other people do them. So there are many, many ways to acquire some of these skills. And at least now we increasingly have ways to be able to see whether people have these skills or not.

So I think credentialing itself is going to have to be dramatically different. You and I probably know an amazing chef and we know that they're amazing because what they do is amazing. But they probably don't have a college degree or a credential in the way that we think about those credentials. This is one of the things we're going to need to solve, how we credential these amazing ranges of skills, how we recognize them, or at least have a way to assess them.

Can you address the last part of it, how to deliver the skills workers need?

This is one of the grand challenges because we all talk about skilling and reskilling but the challenge is we don't have very many examples that are doing this at scale.

One of the questions that I find myself always inclined to ask, is when somebody tells me, 'Oh, there's an amazing skilling program happening here,' I'll say, 'Great. How many people are going through it?' And typically you hear maybe a thousand, maybe even 10,000, maybe 30,000, maybe even 50,000. You very rarely hear hundreds of thousands. You certainly don't hear millions. And yet the scale of the skilling challenge we're talking about over the next decade or two is of that scale.

The question for me on scaling is now how do we do this at scale? Part of that includes incentives for organizations, for companies. There's been some good work that was done at the Aspen Institute future of work task force, which highlighted that we haven't put in place enough incentives aimed at human capital, including skill building. With physical capital, people are allowed to depreciate their investments. We write it off. We have tax incentives of one sort or another. But if you look at the equivalent for investments in human capital, not as much. Some organizations do it and invest in people, but we don't do this as consistently and at the scale we should.

Years ago, when Google studied what made an effective team, psychological safety was one of the key ingredients of that. Given the AI and automation context we're talking about, is there a new key ingredient to effective teams today or on the horizon?

That's a much longer conversation, but this is one of the things where assistive AI can actually be extraordinarily helpful. The amazing capabilities that are now possible in some of these large models will be assistive to teams in extraordinary ways, both for individuals and individual creativity, but also for teams.

The handbook for this new era of business doesn’t exist. We’re all drafting our own as we go along—and now we’d like to start doing so together. You can sign up here to receive this briefing by email.