CEOInterviews.AI
Start App
Get an email when Daron Acemoglu gives a new interview
One email per new interview. Unsubscribe in one click.
Daron Acemoglu
Economist and Nobel Laureate, MIT

MGAIC Symposium: Opening Plenary: The Future of Work in the Age of AI

📅 Nov 17, 2026 Massachusetts Institute of Technology (MIT) 30 MIN 4 VIEWS 27 SEGMENTS · 1 SPEAKERS
Speaker Daron Acemoglu talks about how AI's effect on jobs is not predetermined, and how we should steer toward pro-worker AI that complements humans rather than automating them away. To learn more about MGAIC and the projects and research it supports please visit the MIT Generative AI Impact Consortium (MGAIC) website: https://genai.mit.edu/ The MIT Generative AI Impact Consortium (MGAIC) brings MIT researchers together with industry leaders to explore how generative AI can spawn transformative solutions for real-world challenges. Cross-disciplinary and open-source, MGAIC’s goals are to en...

What Daron Acemoglu said

Written from the verified transcript and checked against it. Every figure links to the moment it was said.

Daron Acemoglu, an MIT economist and Nobel laureate, argued that the future of work in the age of AI hinges on whether AI is used for automation or for creating new tasks that complement human skills. He highlighted that poverty, informal employment, and underutilized human capital are endemic in developing countries, with poverty rates over 75% in some regions and informal employment as high as 75% in parts of Africa. He contrasted successful East Asian growth, which generated jobs, with slower-growing countries like Brazil and India, where employment-to-population ratios stagnate. Acemoglu noted that US inequality has exploded since the 1980s, driven by automation, and that the US tax system is actually more redistributive than Western Europe's. He emphasized that AI's direction is a choice, and that pro-worker AI, which creates new tasks, offers a path to shared prosperity, but developing countries face challenges with digital infrastructure.

Key takeaways

  1. Developing countries are already exposed to AI, and without the right skills and infrastructure, they risk being left behind.
  2. US inequality has exploded since the 1980s, driven by automation, and the US tax system is more redistributive than Western Europe's.
  3. There is growing interest in pro-worker AI, with more attention in the last year and a half than in the past 20 years.

Numbers and commitments

FigureWhat it refers toTypeAt
75% Poverty rate in some parts of the world metric 2:14
30% Youth unemployment in many countries metric 3:30
75% Informal employment in parts of Africa metric 4:57
50% Informal employment in several parts of South America metric 4:57
40%, 50% Employment-to-population ratio in slow-growth countries metric 10:14
1980 to 2013, 2014 Period of inequality increase studied timeline 18:26
60%, 70% Real wage declines for weavers during Industrial Revolution metric 20:39

Chapters

  1. 0:00Global poverty and joblessness
  2. 4:57Where good jobs come from
  3. 8:49Economic growth and employment
  4. 12:43US inequality and automation
  5. 21:34AI's automation agenda
  6. 25:00New task creation and complementarity
  7. 29:15Developing countries' challenges
  8. 30:12Hope for pro-worker AI
Daron Acemoglu 0:01 ↗
Good morning. It's my distinct pleasure to be here with you, and I think this is one of the most important topics we could be studying and thinking about. And I am really delighted that MIT is throwing its considerable weight in taking these issues seriously. So I was very happy to accept the invitation when the organizing committee and Anantha asked me to speak about the future of work, and that's a topic I have studied a lot. And my focus, the focus of my research and a lot of talks I give, is the future of work in the United States and in other developed nations. And as Anantha kindly said, with the support of MIT, David Autor, Simon Johnson, and I run a James and Kathleen Stone Foundation program on shaping the future of work. That's a lot of our work is focused on these issues of the future of work in the United States.
But then, MIT is a global community, and the issues of the future of work are actually even more vexing when you look at the entire world. So that's what I decided to do. So I'm going to talk about the future of work in the age of AI, but with a focus on developing countries as well as the United States. And in a nutshell, poverty, joblessness, low employment rates, underutilization of human capital, informal employment are endemic in the developing world. And if you look at experiences where living standards have improved significantly in the developing world, they coincide with economic growth, but especially economic growth that creates good jobs and employment opportunities. And I want to just talk a little bit about that very quickly to give you a sense of the severity of the problem and the lessons from a few success stories.
And if you look at a variety of measures, you see that the situation we have in the United States, which I'll talk a little bit about in terms of inequality and lack of good job creation, has been a very important problem. But around the world, there are even more severe problems. If you look at a measure of poverty of living on less than $8 a day, some parts of the world still have over 75% poverty, which is striking. And even Latin America, which is relatively prosperous and middle income, a lot of the countries are around 30%, 40%, 50% in terms of poverty rates. So poverty has been a very stubborn problem, and it continues to be so. Inequality, however you measure it, is high and rising. If you looked at this map 30, 40 years ago, some of these countries could have been at sort of moderate levels of inequality. Now they have very high levels of inequality, including the United States, parts of the developed world, Latin America again, Africa, and parts of Asia as well.
And the same is true if you measure inequality, and I'll talk about this in some of my later, more detailed numbers. If you look at the share of the bottom percent of the population, you see the same thing as well. And even more concerning, both for social stability and just as an indication of the difficulty that these economies are having, youth unemployment is very, very high. There are many countries in which youth unemployment is around 30%, and even in many European economies that are generally prosperous and able to still have productivity growth, youth unemployment is around 20%. And in Latin America, again, youth unemployment is considerably higher. And when you look at employment-to-population ratio, meaning working-age population, people between the age of 20 and 64 and what fraction of them work, you'll see that, again, around the world, there is a huge amount of underutilized human capital. A lot of people who could be working are not working. Some of that could be for country-specific factors, such as in the Middle East with female employment, but a lot of it is lack of employment opportunities.
And another aspect of that you can see is that when you look at where people work, a lot of it is informal employment. Informal employment is characterized by firms that are outside of the formal sector. They don't have any kind of protection for workers. They pay very low wages. They have irregular hours and often very dangerous or unhealthy working conditions. Informal employment is as high as 75% in parts of Africa and about 50% in several parts of South America. So all of these are sort of related to lack of good job creation. So the question is, where do good jobs come from? And in a nutshell, the basic economic framework is fairly informative about that. You need to have demand for human skills and expertise. You need to have supply of these kinds of skills and expertise so that employers can actually employ these workers, and then some process institutions that bring workers and firms together in a relatively efficient manner.
In some places, you can have employment being held down because those institutions don't work. And there are a few places in which skills are lacking or haven't kept up with the changing times. But again, the global evidence we have is that a lot of the picture that I've shown you is caused by lack of demand for human skills and expertise. There just isn't enough demand for the kinds of skills that workers have coming from employers. So economic growth has to be of the form that generates these kinds of demands so that it can bring people up together with the process of economic growth. And economic history gives us ample examples of how this happens. For example, demand for new skills and new tasks that workers perform has been essential for American economic growth, starting in the 19th century, but especially accelerating in the 20th century with the new factories and new work practices, both in the manufacturing and non-manufacturing sector.
And what you see, for example, in the new Ford factories that were formed as indicative of the large manufacturing move that the United States was making, is that they were combining new machinery. Electrified factories that organized the factory floor differently and used much more advanced equipment than the companies were using at the end of the 19th century or the very early 20th century were central, but it was also critical that these factories were actually requiring a lot of human expertise. And a lot of it was new expertise, and that's why many of these manufacturing firms were also at the forefront of training these workers. And it was also very important that at the same time as this demand for human capital of a specific form was building up among the employers, we also have that the US supply of skills was keeping up with it with the high school movement and workers generally getting new skills either on the job or off the job. And the US system was relatively efficient in bringing firms and workers together.
In many developing economies, all three of these channels have been a problem. Institutions bringing workers and firms together aren't working as well. The supply of skills needs to be upgraded. But again, as I said, demand for skills is the real culprit. So the question is, how is AI going to impact that? Well, before we do that, let me just give you just a few more pieces of evidence to bolster the case that in the past, when we have had successful cases, it has been associated with economic growth. So here is employment-to-population ratio, the thing that I pointed out as one measure of the underutilization of human skills and talent. You see, if you look at it in the rapidly rising countries, for example, the ones in East Asia, they were already pretty high in terms of employment-to-population ratio at the beginning of the growth process in the 1960s and '70s for many of these countries. But then you also have very rapid growth of the employment-to-population ratio. So economic growth generated jobs, and people took these jobs. Inequality fell. Poverty fell during the process of economic growth. The same, whatever measure you look at: Gini coefficient, poverty, or the share of the bottom percent of the population.
On the other hand, if you look at low, slow-growth countries, and I just picked a couple here, but they are fairly indicative, several of them in Latin America such as Colombia, Brazil, Mexico, from North Africa, Egypt, India, Pakistan, and so on, they have lower employment-to-population ratios, and generally the employment-to-population ratio hovers around 40%, 50%. It doesn't grow even as the economy is growing because the growth rate is not fast enough and it's not generating these demands. You see, the only exception there is Nigeria, and that during that period where the red curve is curving up is exactly when the Nigerian economy picks up a little bit. So the employment creation is very closely linked to economic growth. Poverty declines even in some of these economies, but the decline is slow, and it starts from very high levels, and the same thing for inequality. Here is just a summary of all of these things. If you look at growth and income growth of the poor, they're very strongly aligned.
But the problem is this past teaches us some lessons, the connection between economic growth and poverty reduction and job creation. But it is predicated on the kinds of technologies, in particular the fact that many of these countries were growing in a way that leveraged their workforce's skills. So this is, again, very well illustrated by the rapid-growth countries from East Asia. They started growth in low-value-added, low-skilled manufacturing, and then went up the value chain, and the supply chains became more complex, etc. So using the human resources they had was a critical part of economic growth. So the question will become, in the age of AI, can developing countries grow fast and create jobs? And will that growth actually going to create jobs for the kinds of skills that their workforces have? And actually, this is, as I'm going to argue, pretty closely related to the same question about the future of work in the United States and other advanced economies.
So to make that connection, let me actually look at the picture of inequality. Since poverty is not as major a problem in Western Europe and the United States, let me quickly look at inequality and... what happened? One of my figures is missing. OK. Actually, I was going to start from here. Sorry. This figure somehow got the order, got mixed up. OK. So here is a summary of the income growth and inequality in Western Europe and the United States. On the left, you see the huge increase in inequality in the United States. The red curve is the share of the bottom 50% in income. It's going down. And the bottom 50% now earn much less than, significantly less than what the top 1% do, whose share has increased very steadily.
If you look at Western Europe on the right, of course, there is heterogeneity within Western Europe, and it's not entirely informative to bundle all of these economies into a single category. But for comparison, this is what sometimes people do, and I'm just following that. You see, there is an increase in inequality in Western Europe too. Both the bottom 50% share is declining and the top 1% share is increasing, but it's much less than what you see in the United States. So the question is, why is this happening? And there is actually a political issue that many people discuss as well. Let me skip to the third panel there on the right, which shows that the US-Europe contrast is true at every percentile of the income distribution. And many arguments emphasize the role of the welfare state or redistribution between the US and Europe. Actually, it turns out that's a relatively small part of the story about inequality.
So here, what I'm showing you is the same thing pre- and post-tax. So you can see both Europe and the United States, and the pre-tax and post-tax figures are entirely parallel to each other, both for the bottom 50% and top 1%. So it's not that somehow there was a lot of inequality in Europe in earnings that was taken up by the tax system, or that there was a lot of inequality added in the United States. The US problem is that in the labor market, we are not generating enough income growth for low-wage workers. And that actually is very related to why Latin America is not generating enough jobs for low-income workers, and they end up either unemployed or in informal employment, or why Pakistan or India are having the same problems. In fact, even the many stories that we have about US-Western Europe comparison in terms of redistribution aren't entirely accurate.
There are many problems in the US tax system, but overall, it's actually not less redistributive than Western Europe. So here is the distribution of taxes by different income percentile groups. You see Eastern Europe, Western Europe, and the United States. The thing that is striking here isn't that the United States is actually less redistributive. In fact, those curves there are sloping up faster in the United States. So if anything, the US tax system is more redistributive. The problem in the United States is that US taxes are much less than Western Europe. So some of the social welfare state services are much weaker. But overall, the tax system is quite redistributive. So you might wonder, what's going on with transfers? Perhaps Europe is much more generous for the bottom of the income distribution in terms of transfers, and a little bit more so. So now, here, a steeper slope downwards means more redistribution to the bottom percentiles of the income distribution. Europe is a little more redistributive in terms of transfers, but not so much so.
So the bottom line is that the United States has been very unfavorable for low-skilled workers. Inequality has exploded in the United States, mostly because we haven't generated enough labor market opportunities for low-wage workers. So that's what we need to understand, and that's what we need to think about what will happen in the age of AI. So here is a first step towards that. Let's look into the US inequality trends a little more. And on the left and the right, I show you where that inequality started coming up. And you can see it started in the 1980s for different education groups. Their income profiles or wage profiles were very much parallel in the 1960s. If you extend this to before, it was also so in the 1950s and 1940s, but then from the 1980s onwards, they start diverging. Even worse, low-income, low-education groups, especially low-education men you see there with orange and green, have experienced declines in their inflation-adjusted wages.
And then on the very far corner, that right picture is from my research with Pascual Restrepo, who's at Yale University, was a graduate student here at MIT. And what we show is that a lot of this inequality is related to automation, pre-AI automation, but automation, meaning that it was robots, software systems, and other machinery that displaced workers from what they were doing. And then as a result, those worker groups experienced wage declines or wage stagnation. And the reason why many of these low-education men are at the bottom of that curve of that figure is because many of them were employed in things like blue-collar jobs in heavy industries, and those were the jobs that were automated. So in particular, in that picture, we show for more detailed demographic groups distinguished by gender, age, education, and ethnicity, the overall change from 1980 to 2013, 2014, that period in which inequality increased on the vertical axis, and how much of the jobs that they used to do were automated subsequently.
And you see a very strong relationship between how much automation a particular demographic group experienced and how their incomes have fared. So that decline of that orange curve, for example, is mostly related to the automation of those jobs. So conceptually, what's going on is that when you have certain tasks and you automate those tasks, workers who are performing those tasks will have to be displaced. And the easiest thing that they can do is they can go to some of the remaining tasks that are still performed by humans. But that creates crowding in those tasks. And often, those are the tasks for which they're not trained. They don't have a comparative advantage, so their earnings actually decline. But they also push down earnings for other workers against whom they're competing. So as a result, overall, the labor share declines, and wages and employment are not doing so well as a result of automation.
And this is not just confined to the recent period. If you look at the beginning of the Industrial Revolution, we see exactly the same thing. The beginning of the Industrial Revolution is sometimes ignored in discussions of how today, in the age of AI, we're going to generate the second or third or the fourth or the fifth Industrial Revolution, depends on how you count, and everybody's going to benefit. But for about 80 years, many workers were not doing very well in the Industrial Revolution, precisely because of the same things. So some of the highest-paid workers, for example, who were working as weavers were displaced from their jobs as a result of power looms, or they were pushed into simple operator jobs, and they experienced something like 60%, 70% declines in their real wages. And that wasn't reversed for a very long time period. So the question is, what will AI do?
And the answer should be AI would probably do what we decide to do. Because AI is not just one thing. Even within generative AI, there are just so many different ways of developing generative AI, using generative AI. And we, as humanity, but especially the people who are working on AI, many of whom are represented here, and some of you may, in the future, work on the direction of AI, will shape the future of AI. But at the moment, there is a strong indication that a lot of what current AI models are trying to do is more automation. Under the banner of AGI, artificial general intelligence, what we're trying to do is make AI models mimic human skills. Mimicking human skills means that as they get closer and closer to humans, there will be more and more tasks that these AI models can do, and they will be able to perform these tasks reliably at some point. And you can already see that in coding, perhaps even in mathematics, a very big surprise for people.
And when they do that, employers are going to allocate these tasks to AI models, and the workers that are going to be displaced, like what I have shown you in the previous few slides. This is entirely consistent with the way that AI was conceptualized, going back to that brilliant mathematician Alan Turing's work, as well as the initial AI sort of push that came from the Dartmouth Conference, for which MIT was a participant. Also, the intention and the prediction of many of these brilliant scholars was that AI should mimic human skills, and that will happen very quickly. Now, it didn't happen quickly, because it turns out that it's actually hard to mimic human skills. Generative AI is a big step in this, but there is no guarantee that it's going to happen. And in fact, if you look at developing countries, they have a lot of skills, a lot of jobs that are already things that AI can do quite well.
So the view that somehow AI is going to compete against American or West European workers, but not so much against developing country workers, isn't quite true. Developing countries are already quite exposed to AI if you take the current jobs that AI can do. So then the question becomes, what else can we do with AI? And Anantha already mentioned complementing workers, and I think there are many ways in which that can happen. But my work, especially with Pascual Restrepo and David Autor and Simon Johnson, has emphasized the role of new technologies and AI, not in just automation or doing things that just marginally improve what workers already do or how efficiently capital works, but what we call new task creation, meaning that you actually enable workers to do new things. And what's brilliant about generative AI is actually that it has this potential in ample quantities.
And when you do that, the picture I showed you earlier completely changes. So again, you will have the automation there. Workers are going to be displaced. But now, workers don't just have to go and compete for the same set of existing tasks. What I'm showing you in blue there is the new tasks that are added, which use new human skills, and workers are going to be able to now acquire those skills and find employment in things that don't necessarily compete with the old things, but also meet new demands, new goods, new services, new innovation. And theoretically, and some of the empirical work that I've done with Pascual Restrepo documents this, but I'm not going to go into those details, when you look at this, you see that the implications of this kind of new task creation is very, very different from just automation. So in particular, now, the labor share can increase, wages and employment can increase. So the way to create shared prosperity out of new tasks is much easier.
And new task creation is the heart of creating things that are complementary to humans. So is that possible? Yes. In fact, I would argue it's not just possible. In many ways, it's easier than the automation agenda. And the reason for that is that automation is trying to mimic humans. And if AI models, which are very capable and are becoming more capable, were exactly wired like the human brain, that would be perhaps a natural path. But the way that these models work, where they are trained, their comparative advantage are very different from humans. They are amazingly efficient and powerful in sifting through very large quantities of data and finding relevant things, drawing analogies, but they're not as good as humans in one-shot learning, experimentation, hypothesis testing, modeling, simulating different worlds in their brains. So when two things are different, of course, you can try to mimic one with the other, but it's not going to be the easiest path. When two things are different, the most natural thing is to use them in a complementary way.
It would not have been a good idea to use, OK, I'm going to reveal my age, use Michael Jordan to mimic Scottie Pippen. But what made the legendary Chicago Bulls teams work was that the two were complementing each other. So in the same way, AI and human skills can complement each other, and they can complement in creating these possibilities for new jobs for humans. By the way, while Alan Turing and Marvin Minsky and people like that are associated with the view of AGI and automation, there is a set of scholars who argued for this long before me, and they are even more strongly associated with MIT. The first person to do that was Norbert Wiener, a very important mathematician and an engineer, and J.C.R. Licklider, who was also at MIT. OK, Douglas Engelbart wasn't. He was on the West Coast. But they already articulated and showed how you could use computer technologies in these human-complementary ways.
And many of the important things that we depend on today came out of this agenda, including the internet, mouse, menu-driven computers, and things like that. And when we can do this, it's going to create a lot of opportunities for workers. But even more importantly, it's going to create a lot of possibilities for generating new goods and services, which the world depends on because of aging populations, because of climate change, because of changing preferences, rising middle class around the global world. All of these things are actually not just producing more of the same. They're about doing new things, and for doing new things, human talent and better tools, which generative AI is very capable of producing and supplying, would be absolutely essential. But another challenge for many countries, especially in the developing world, and I will conclude with that, isn't just the direction of research, but it's also digital infrastructure.
So in some sense, developing countries are between a rock and a hard place. They need the right kind of AI, and they have very little control over it. But they need the right skills and the right infrastructure, and unless that happens, they're going to be left behind in all of this race. So let me now sort of conclude, which is that I think there is a lot to be hopeful about the future of AI. And if we want to make the best of AI, what we need is to be on that bottom right corner, which is high adoption of AI, but in a pro-worker direction. And that requires both the infrastructural investment, but also the direction of AI to be right. If we do high adoption of AI, but if we just go into the automation direction, we may get some of the productivity benefits, but I think the societal implications would be much harder.
And the good news is that I think there are a lot of opportunities for pro-worker AI, and there is more interest in it over the last year and a half than I'd seen when I started talking about these things 20 years ago. And I think with people, like those in this room, becoming interested in this issue of how we can use AI in both a productive and a more human-complementary way, I think there is a lot to be hopeful about the future. Thank you very much. [APPLAUSE]

Cite this transcript

APA, MLA, BibTeX
APA

Acemoglu, D. (2026, November 17). MGAIC Symposium: Opening Plenary: The Future of Work in the Age of AI [Interview transcript]. Massachusetts Institute of Technology (MIT). CEOInterviews.AI. https://ceointerviews.ai/interview/3314624/

MLA

Daron Acemoglu. "MGAIC Symposium: Opening Plenary: The Future of Work in the Age of AI." Massachusetts Institute of Technology (MIT), 17 Nov. 2026. Transcript, CEOInterviews.AI, https://ceointerviews.ai/interview/3314624/.

BibTeX
@misc{acemoglu2026_3314624,
  author       = {Daron Acemoglu},
  title        = {MGAIC Symposium: Opening Plenary: The Future of Work in the Age of AI},
  howpublished = {Interview transcript, Massachusetts Institute of Technology (MIT). CEOInterviews.AI},
  year         = {2026},
  month        = {nov},
  url          = {https://ceointerviews.ai/interview/3314624/},
  note         = {Speaker-attributed transcript with timestamps}
}