Key Issue for AI is More Emphasis on Task Creation or on Automation: Nobel Laureate for Economics

TMTPOST – Simon Johnson, the 2024 Nobel Prize winner in economics, has noted that whether to focus on new task creation or automation when applying artificial intelligence (AI) would lead to two diverging outcomes: shared prosperity or widening income gaps.

Johnson made the comments during his keynote speech delivered during the T-EDGE Conference organized by TMTPost Group on December 7.

There are techno-optimists and techno-pessimists when it comes to AI; optimists believe in amazing effects of AI such as universal basic income for everyone and the latter are doomsaying: people’s lives would be ruined with the elimination of jobs and the middle class would be wiped out entirely, Johnson said when summarizing a state of discussions on the impact of AI on the economy.

“Our view is somewhere between these two extremes,” said the economics professor at the Massachusetts Institute of Technology (MIT) and the co-director of Shaping the Future of Work initiative at MIT, adding that productivity growth rate would be on trend in the United States and that would not afford universal income in the country and a pro-worker approach can reverse the existing growing income gap since the 1980s between those at the upper end of wealth distribution and those at the bottom.

“We think that over the next 10 to 20 years, U.S. productivity growth will remain roughly where it is, so roughly on trend,” said Johnson, who coauthored the book Power and Progress: Our 1,000-Year Struggle Over Technology and Prosperity, which delves into the significance of inclusive institutions to shared prosperity.

Johnson was awarded the Nobel Prize in Economic Sciences in October for studies of how institutions are formed and affect prosperity, along with Acemoglu and James A. Robinson.

Looking at the Industrial Revolution, he explained that technological advancements did not necessarily bring about wage raises and there must be a productivity boost and creation of new tasks as a precondition. While machines replaced workers in Britain about 250 years ago, “the demand for labor” in other sectors that were not automating went up and these could be activities complementary to the automating sector, he said, referring to them as “new tasks.”

"I think what we need is a recognition of the process of skill acquisition and figuring out which skills are going to be in demand in an AI-driven economy," Johnson advised on how to tackle the challenges posed by AI.

“AI presents us a moment of choice… All technology decisions involve someone deciding whether to make a technology favorable to one group or another,” the professor noted. He went on to elaborate on how AI can favor cognitive workers by increasing their capabilities in writing and coding and benefit workers with less education as well, citing the example of the U.S. transition from the predominantly agricultural economy to the industrial nation in the 1890s.

“But it could also favor people with less education. And we have a term that we coined and that we attempt to use and explain in all context, which is pro-worker AI,” said Johnson.

While Johnson believed that the government could do a great deal to move the path of technology in a pro-worker direction, he was not upbeat that the pro-worker AI would be a priority for the upcoming Trump administration.

He also predicted global inequality is going to increase although it can be reversed through global cooperation.

The T-EDGE Conference, with the theme of All-in on Globalization and All-in on AI, featured a galaxy of innovators, entrepreneurs, investors and senior officials from all over the world, attracting hundreds of millions of online and offline viewers.

The following is the edited transcript of the keynote speech delivered by Simon Johnson, the Nobel Laureate in Economics and a professor at MIT, at the T-EDGE Conference on December 7.

Hello, everybody. It's a great pleasure to be with you. My name is Simon Johnson. I'm a professor at MIT's Sloan School of Management, and I'm co-director of the Shaping the Future of Work Initiative at MIT. I'm going to talk to you on how artificial intelligence is likely to affect the economy of the United States, the economy of Europe, but also the economy of the world.

I think there's a lot of really interesting questions to explore there. We know some things, but there's a lot of things we don't know. And I will try to be very clear and honest with you about what we don't know.

This is a summary slide, because I would like to get my main points across to you and reinforce them in the rest of my talk.

And the state of discussion, at least in the United States, is really one of polar extremes at the moment on AI as well as on other things. Some people are very convinced that the technology will have amazing positive effects. So sometimes these people are referred to as techno-optimists. And one version of techno-optimism, of course, is that nobody will need to work again, and universal basic income can easily be afforded for many people.

On the other hand, there's another view, which I think we can reasonably call techno-pessimism, which says, hold on. If you're going to eliminate all these jobs, what you're probably going to do is ruin people's lives. You're going to hollow out the middle class entirely. And this is going to create major political and social problems.

And it's possible that what we might get is what's so-called excessive automation, which I'll talk about a little bit later, which shifts the power balance within companies towards the owners of the companies and top management of the companies. But it doesn't necessarily boost marginal productivity or the productivity of that additional worker very much. And in that scenario, the people who make the decisions could implement a lot of automation through AI.

But if national and global productivity doesn't go up by much, you can't afford universal basic income. So then we're talking about a lot of poverty for a lot of people.

Now, our view currently and of course, our views evolve as we get new data. But our view is somewhere between these two extremes. We're not taking an extreme position.

We think that over the next, say, 10 to 20 years, U.S. productivity growth will remain roughly where it is, so roughly on trend.

We can discuss other countries when we come to the question and answer session. I think we have less data. But there are some things we could say about Europe, for example, and some things about emerging markets. We think, though, that even though the effect of this macroeconomic overall growth level will be fairly modest, we are likely, on our current trajectory, to get a lot more job market polarization. So a lot more gains at the top and a lot more people getting pressed down towards the bottom.

Again, in the United States, I think this potential picture is clearer. But we can see some hints of this elsewhere also.

This job market polarization will widen income gaps. That's absolutely a feature of this kind of phenomenon. And we expect that global inequality is going to increase, both if we define it within countries, but also, of course, across countries, which is a very important consideration for all of us. The key point from our research and all of our engagement with industry and all of our policy analysis and engagement with policymakers is it doesn't have to be this way.

AI presents us with a remarkable moment of choice. All technology involves a choice. All technology decisions involve someone deciding whether to make a technology favorable to one group or another.

AI, because it's so potentially influential, particularly for cognitive workers, people who work with their minds, it has a particularly pointed potential to go rather more in the direction of favoring people at the higher end of the income distribution. But it could also favor people with less education. And we have a term that we coined and that we attempt to use and explain in all context, which is pro-worker AI.

We are advocating for artificial intelligence that will boost the productivity of workers who did not go to college, for example, in the U.S. or didn't go to four-year college. And we think there are many ways in which AI could develop technically in ways that will boost the productivity and therefore the potential pay of those lower income, less educated, and less well-to-do workers. So more on that in a moment.

But of course, the key question is going to be, in the United States in particular, who makes the decisions on which technology path you take, and why do they make those decisions?

Now, I'd like to say that while this is a very short presentation, it does build on quite a long book that I published last year with Daron Acemoglu called Power and Progress, Our 1,000-Year Struggle Over Technology and Prosperity. Daron is my long-time colleague at MIT. He's an Institute professor, which is a very high honor. And Daron is one of the many winners of the Nobel Prize for Economics in the Economics Department.

I, as I think you know, have been very fortunate to work with Daron and Jim Robinson. And I was also awarded a share of the prize this year. There are many big statements and massive claims about AI, what AI will do to human capabilities. One leading entrepreneur in the space has said we will all be gods, which is an interesting statement we could discuss. Our work is much more, I would say, mundane and much more focused on jobs, because we think that a huge amount of the AI impact will run through jobs. Although, we're also very concerned about and we work on the impact of AI on social media, the impact on information and disinformation, the impact on how we make decisions in our societies, because that's really important for the technology path that we take.

So before I get to AI, I just want to make a couple of points about our existing moment. Again, focused on the United States, which matters a lot for the world economy, but it particularly matters for AI because of where AI is being developed. And the key point here is that we have had a lot of job market polarization already since the 1980s in the United States and in other industrial countries.

So if you look at the picture on the right here, this is the change in real weekly earnings for men in the top and women at the bottom since the 1960s. This is a picture that draws on work by Daron, my co-author, and David Autor. David is the co-director with Daron and me of our Shaping the Future of Work initiative at MIT. And what you can see in this picture on the right is that there's a dark blue line that rises quite nicely. That's the earnings of highly educated people. No problem there. In the middle, it's been less impressive for people with mid-level education. But look at that bottom line for the people with the least education in the data. Their real wages have barely moved since the 1960s. That's a remarkable and very unfortunate outcome given how much technology has changed and how much, quote unquote, "progress" we've had during this period of time.

And in addition to the squeeze on the bottom, there's also been another squeeze in the middle, which is fewer-- we've lost a lot of middle skill, middle education jobs in the middle of the income distribution. So many people have been pushed down.

Now, why do I talk about this? Why do I begin with this? Because I think that the question to focus on is whether AI is going to reinforce this polarizing trend or rather reverse it.

One CEO in the United States said to me -- and this is his term, not my term. I don't like this term. None of us should like this term. But it is perhaps descriptive and tells you how a CEO is thinking about it. He said that AI will eliminate what he called cut-and-paste jobs, which means cognitive work that involves a routine element that AI can handle fairly easily at this point.

Now, I'll talk more about the details and the contours of where we think AI is at the moment.

But obviously, the rate of change is very fast. The amount of capital being deployed is enormous. The pull of talent, not just in the United States, where many people leave universities, including faculty, to go work in the AI sector now, but also globally, people being pulled towards this AI creation sector focused in the United States. So AI capabilities are going to increase dramatically in this space.

And I think we have to expect what's delicately called, for example, by McKinsey, employment transitions in the U.S. to continue at a fairly intense pace. And employment transitions in places like Europe may well increase, perhaps back to the pace that they saw during the COVID pandemic, which was pretty intense.

Now, the discussion of these issues and the thinking about the technology, of course, runs through all the forums that we have available to us in our various societies.

The problem is that AI, at the same time as impacting jobs, is also impacting social media. It's impacting digital advertising. It's affecting mental health in ways that are not healthy. It's affecting children. And it's really undermining our ability to make democratic, inclusive decisions, I think, in all the different contexts that we do that around the world.

Now, in addition to those concerns about inequality in the U.S., I think we can have and should have concerns about what happens globally. Obviously, what we've seen over the past 40 years is some countries, including China, rising up quite dramatically and quite impressively in the GDP per capita statistics. Now, personally, I hope that this continues. I hope that many people around the world continue to share in prosperity and have higher productivity.

But if the impact of AI is on cognitive workers, and if that also spreads to many manufacturing jobs, which it hasn't yet, but there are many smart people working on that extension, then I think it will become much harder for countries at the middle level of global income to continue to grow.

And there may even be some downward pressure through the effects on global supply chains.

There may be some de-skilling of jobs. Automation may replace manual work in a wide range of activities. We're not there yet. But if we're looking at 10 to 20 years, I think that is a pressure that's going to be there.

And of course, I'm going to talk about what we can do to counteract that, because none of our agenda should be interpreted as doomsaying or passive. On the contrary, we're trying to make and provide you with an honest assessment that can enable you, private sector, government, everyone in between, to think about policies and approaches and strategies that will counteract the trends that you don't like here.

So to understand that, we should take a little bit of a look back in history. And this is what we do in our book, Power and Progress, and ask the question, when technological transformations have positive effects, when they generate shared prosperity, how is that possible? How does that happen? Particularly because ever since the Industrial Revolution began 250 years ago, and perhaps before, many new innovations and a lot of machines have been focused on automation of work. What is automation? It means simply that we are replacing what humans do with a machine. Obviously made by humans, obviously operated by humans, but you need fewer humans to run that machine than all the people who were previously doing the manual work.

So if innovation of the modern variety requires and involves and runs really through automation, replacing workers, how is it possible that wages go up?

Well, the answer, of course, is you need the demand for labor to go up in other sectors that are not automating. And these can be new activities. These can be activities that are complementary to the automating sector. The key word that we focus on, and this is in David Autor's work, for example, is tasks. Who has what kind of task? Any job is a bundle of tasks, and you can strip jobs down to their underlying tasks. We, most of us, have between 20 and 30 tasks in the jobs which we're paid money. So who is getting new tasks, and particularly new tasks that require expertise? Because it's the expertise you get paid for.

Now, if you look back in history, you can see some remarkable moments, including, I would emphasize, in the United States in the 19th century, which started out as a relatively less educated place. We took in a lot of immigrants in the 19th century who had no education. Many of them didn't even speak English. And they were put to work in or they found work in factories where the employers and managers figured out how to make best use of their talents and to boost the productivity of workers without much formal education.

This is called, by the historians, the American system of manufacturing.

And it was a remarkable success. And it also generated technology that was shared around the world, not evenly, not all at the same time, obviously, with the consequence being

that the United States went from being an agricultural country with very little industrial production, which you can see in my graph here, to a country that led the world in industrial production by the 1890s. And higher wages in the United States followed from that higher marginal productivity, the productivity of the extra person you hire, combined with the rise of trade unions, which is an important part of the story.

Now, if you want to look around the world at success stories, I think Japan, to me, stands out as a remarkable transformation. If you look at real wages after World War II, Japan starts out much lower than the United States. Obviously, productivity rises for several decades. And then it translates into higher wages. So we don't expect wages to rise instantly as productivity rises.

But you're hoping to get some form of convergence in real wages as a result of the new task creation. Because remember, Japan also had, and actually, for a while, led the world in automation. So there's nothing in our work or our recommendations that should be interpreted as being anti-technology. We work at MIT, so we're not anti-technology. And we're certainly not anti-automation. I don't think you can stop automation. I don't think you want to stop automation.

The key is, do you also, in parallel, create enough new tasks? So let's turn to AI and think about what that looks like. So clearly, AI is going to be automation, involves automation. I don't think there's any argument about that. You are replacing people with, in this case, algorithms. And we're noting that this process of purely replacing people with machines does not necessarily lead to higher wages. For example, in the early British Industrial Revolution, there was a 60-year lag between big productivity transformations and higher wages. 60 years is a very long time.

I don't think we want to wait that long in this case. And we also should note that we always create, in all our economies-- this is best documented for the United States-- we always create new tasks. David Autor has done very good work on this.

Oh, and I should say that we'll circulate these slides and the underlined names and links. You can follow and look at the sources yourself. I think that's very important in the age of AI to trust but verify all speakers, including me.

So between 1940 and 1980, we generated enough new tasks to keep pace with automation and keep the demand for labor, including the demand for workers without a lot of education. We kept that extremely buoyant, and that real wages rose. Since 1980, we have not done a good job of that. Now, AI could be used to upskill workers. And there's some really interesting research here. And I don't have time to go through it all. But again, I have the links that you can look for yourself.

So ways in which, for example, ChatGPT. But obviously, there are competitors-- ways they can

help you improve your writing, ways that they can help improve customer service, including making workers with less skill happier in their job and making the customers happier, which is a pretty impressive combination. And we definitely know that, for example, GitHub Copilot can really help with writing software. And I think improvements in productivity for people writing code is one of the most established and most robust results that we have here.

Now, of course, even in that rather rosy, positive scenario, there will be skill gaps. And there's some very interesting work that's been done to identify these gaps. And again, I recommend to you the underlying studies that I link here.

I think that the key point is that we don't know exactly what skills will be needed. We need to push very hard to encourage people to acquire those skills. We need to track which skills are really paying off for people in terms of what jobs they can get, and then feed that information back so people can make better choices.

I don't think any top-down recommendations from me, my co-authors, the government, anyone is the answer here. I think what we need is a recognition of the process of skill acquisition and figuring out which skills are going to be in demand in an AI-driven economy. So that might be less than fully satisfying answer, but I think it's reasonable. And I think that's the base on which we should proceed.

Now, we can ask, where are the potential skill mismatches? Or if you want to be a little more pointed, who is most at risk? So again, if you break all jobs down into tasks, you can see what people are employed to do. We also know which tasks AI is good at, although, as I said, that's evolving very quickly. But at least in that snapshot, you can see some categories of people who are most at risk of losing their job, and that would include, for places we have data, again, more United States. We can see that women and younger workers are most at risk of losing their position. Now, I should emphasize that that is the negative impact of the automation.

These studies do not yet look at who's most able to move to new tasks. And to the extent that women and younger people are more able to grasp what's needed and to acquire the skills and to get the new opportunities, that can be a very important counterbalancing effect.

So all of these things are snapshots. All of this research is very dynamic, I would say, because the technology is dynamic. But this is what's in the data right now. And I think it's entirely appropriate that we're honest with ourselves about seeing this.

And then we track, going forward, who is better able to make the labor market transition with what outcomes, who gets higher pay as they change jobs rather than lower pay, rather than getting pushed down to the lower end of the labor market.

Now, there are some very optimistic macroeconomic forecasts about the impact of AI, both U.S. and globally.

And my colleague, Daron Acemoglu, I think has done a really good job evaluating these and attempting as best we can to link what we know about tasks and new task creation and the elimination of automation of tasks with the macroeconomy. He finds that the overall effect on total factor productivity, which is the standard economic measure of efficiency of use of resources, he finds it will be quite modest. I think in the policy space, I would recommend that you follow the writings and speeches of John Williams, who's the President of the New York Fed. His line, which-- and I've talked about this with him, and I think it's entirely reasonable-- his line is that we will continue to experience roughly the same productivity growth in the United States. AI will be part of what gives us that productivity growth. So AI is important in that sense, but it's not a miracle. It's not a massive boost. And therefore, schemes about extremely generous universal basic income, for example, seem rather fanciful because the whole economic pie is not getting that much bigger, or it's not accelerating in its rate of growth.

I think the other point to emphasize is that while there are upskilling opportunities,

as I just flagged-- think about customer service, for example, where existing workers are becoming more productive.That's good. Good for wages, good for labor demand.

But there are also plans-- I mean, companies are quite open about this-- to replace those workers, all of those workers, with artificial intelligence.

The technology is not ready for that, but how long will it take? I don't know. One year, two years, 10 years, very hard to say. Or maybe you just need a few humans

to oversee a huge amount of customer service that's automated. In any case, teaching the algorithms to better help workers is a step on the path to teaching and allowing the algorithms to take over from workers.

And that's why we really need to accelerate the creation of new tasks. Now, we've worked on this quite a lot in the policy space, including in the United States. I have to say that policy space is changing in the United States-- you know this from the newspapers-- on a daily basis right now. And we will see what the new government under Mr. Trump is interested in doing and willing to do. There's obviously big national security implications of AI.

It's also the case that the United States is attracting most of the AI investment in the world outside of China. We don't have-- in the United States, we don't have good information on what's happening in China. So all of these statements are without China. But what we know is countries, sovereign wealth funds, and private investors around the world are deploying lots of capital to the United States. One estimate is 95% of the capital, excluding China, is in AI foundational training models is being deployed in the U.S., 3% in Europe, and 2% in the rest of the world. Hard to know if that's exactly right. But if you think if you follow the talent and see where people are working really hard on these issues, that does seem to line up quite consistently. So we think that the government could do a great deal to move the path of technology in a pro-worker direction. And I can go through these specific recommendations that are rather US-centric.

But I think the U.S. is the center of a big part of the development of this technology, so that's appropriate. Honestly, I don't think under the Trump administration, these items will be a priority. And therefore, the conversation will shift to the private sector and what can be built with philanthropy, what can be built on the basis of private companies, the big tech companies, which are open to these ideas, I have to say. We have very good conversations with senior executives. But pro-worker AI is not, I think, honestly seen as the dominant strategy or the best way for them to make a return on the capital, which is what their shareholders and their creditors want. So it's an uphill battle, to be absolutely honest.

And I think we are going to see some unfortunate, heavy emphasis on the automation side of AI rather than what we would prefer, which is more emphasis on new task creation, new capability creation, extending what humans can do. But it's a big country. It's a big technology space. MIT plays an important role in helping people think about technology as well as helping people build technology.

And we are very engaged in these questions on a day-to-day basis, for example. And I'm happy to expand on that when we do our Q&A. So this is my last slide. Our book is Power and Progress. It's coming out in about 20 languages around the world. It's been well-received. And it's extremely heartening that I've been involved in many conversations in most of these countries about the future of technology. So I think it's a shared human moment to wonder and worry about whether technology will be helping us all or whether technology in this current iteration, in the next iteration, in this very powerful, important new iteration, whether it will be helping primarily just a few people and what that will do to income inequality, to jobs, and to the future of the global economy.

Thank you very much.