Daron Acemoglu on Artificial Intelligence

Listening to the ongoing debate about artificial intelligence, one could be forgiven for assuming that the technology is either a bogeyman or a savior, with little ground in between. But that’s not the stance of economist Daron Acemoglu, professor at the Massachusetts Institute of Technology and the author, with Simon Johnson, of the new book Power and Progress: Our Thousand-Year Struggle Over Technology and Prosperity. Combining a cogent historical analysis of past technological revolutions, he examines whether a groundbreaking new technology “augments” the status quo, as opposed to merely squeezing out human labor.

“[M]y favorite term is ‘creating new tasks’ because I think it really clarifies what the quote unquote augmenting needs to take the form of,” he tells interviewer David Edmonds in this Social Science Bites podcast. “It’s not just making a worker more productive in tightening the screws, but it’s really creating new jobs that didn’t exist.” And so, he explains to those perhaps afraid that a bot is gunning for their livelihood, “Automation is not our enemy. Excessive automation is our enemy.”

This is not to depict Acemoglu as an apologist for our new silicon taskmasters. Current trends such as the consolidation of power among technology companies, a focus on shareholder returns at the expense of all else, a blind trust in companies to somehow muddle through to societal equilibrium, and a slavish drive to automate everything immediately all leave him cold: “I feel AI is going in the wrong direction and taking us down with it.”

His conversation doesn’t end there, thankfully, and he offers some hopeful words on how we might find that modus vivendi with AI, including (but by no means only relying on) “the soft hand of the state in tipping the scales one way or another.”

Acemoglu is an elected fellow of the National Academy of Sciences, the American Philosophical Society, the British Academy of Sciences, the Turkish Academy of Sciences, the American Academy of Arts and Sciences, the Econometric Society, the European Economic Association, and the Society of Labor Economists. He is also a member of the academic-cum-policymaker group of economic movers and shakers known as the Group of 30.

Besides Power and Progress, his books include the popular bestseller Why Nations Fail: Power, Prosperity, and Poverty written with James Robinson. Acemoglu has received a number of prizes, including two inaugural awards in 2004, the T. W. Shultz Prize from the University of Chicago and the Sherwin Rosen Award for outstanding contribution to labor economics. He received the John Bates Clark Medal in 2005, the Erwin Plein Nemmers Prize in 2012, and the 2016 BBVA Frontiers of Knowledge Award, as well as the Distinguished Science Award from the Turkish Sciences Association in 2006 and a Carnegie Fellowship in 2017.

To download an MP3 of this podcast, right-click this link and save. The transcript of the conversation appears below.

David Edmonds: According to the doomsters, artificial intelligence is going to wipe out most of our jobs: Social Science Bites will be brought to you by an advanced chat bot. But Daron Acemoglu, co-author of Power and Progress, and a professor of economics at MIT, says such an AI dystopia is not inevitable. Daron Acemoglu, welcome to Social Science Bites.

Daron Acemoglu: Thank you. It’s my great pleasure to be here. Thanks for inviting me.

Edmonds: We’re talking today about the economic impact of artificial intelligence. But before we get there, let’s take a step back and talk about technology more generally. There’s a common view that, despite worker resistance to technologies throughout history – most famously, one thinks of the Luddites – that in the end, technology benefits almost everyone. Is that how technology works? Is it true that it’s almost always advantageous to all of us?

Acemoglu: Yeah, I think you’re right about the common belief, and probably journalists, economists and technologists are equally to blame for this view. I think the world we live in is, comparatively speaking, a fortunate one. We are much more prosperous, healthier, comfortable than those who lived 300 years ago, and we are surrounded by technology, and there is a general techno optimism around us. And all of that implies a particularly strong default position for most people, that technology is an amazingly useful tool, and just let it rip, and everybody will ultimately benefit, and that ultimately means something different to different people. And yes, indeed, that’s why ‘Luddite,’ for a while, and even today, has been used as a term of abuse. So, if you are questioning technology, you’re a Luddite, and it’s not a good thing to be.

So, at the risk of sounding like a neo-Luddite, which I’m not, I do see the benefits of technology. But I would also say that who benefits from technology is always up for grabs, and there is no guarantee that technologies in general will benefit humanity. I mean, think of the knowledge to make much, much more powerful bombs or much, much more deadly biological weapons. That’s knowledge; it expands what we can do. But on the whole, it wouldn’t help humanity greatly. And there are many technologies in the age of Luddites, like exactly the weaving machines that the Luddites were exercised about, which tend to be good for business owners, factory owners, but not so good for workers whose wages are cut. So, it’s all up for grabs.

Edmonds: But the Luddites objected to these mechanized looms, and as a result, we all have cheap clothing. You’re not suggesting that it would be better if clothes were made by hand?

Acemoglu: No, absolutely, I’m not suggesting that clothes should be made by hand, nor that we should give up diggers and excavators and go on using shovels. But some of the Luddites may have had the view that, yes, break all the machines and let’s keep on doing things the way we have been for ages. By the way, which is also not quite correct, because the weavers were doing so much better in the early 19th century, precisely because spinning had been mechanized, and that had increased the demand for weaving. But that’s a different story.) But there were others within the broad Luddite tent that were saying, “No, let’s try to find a way of introducing the new technology, but in a way that’s good for workers as well.” And I think that’s the emphasis that is very important and sometimes gets lost.

Edmonds: It’s difficult to imagine what that would mean in terms of the mechanized looms, other than perhaps introducing it more slowly so that workers have time to adapt.

Acemoglu: Yeah, I think that’s one possibility, but that’s not the one that I am actually about. That’s not the right perspective. The right perspective is: can we also do other things with technology at the same time so that we introduce new tasks, new capabilities for workers? So, the world would be a very different one in the early 19th century if, instead of just rapidly throwing out weavers because the power looms took their jobs, at the same time you also introduced other technologies that could re-employ these workers, especially re-employ them at relatively high wages.

And if you look at periods in which the UK economy or the American economy have generated rapid, robust wage growth, such as, for example, the three decades that followed World War II, those weren’t characterized by lack of automation. In fact, automation was quite rapid. There were a lot of machines, just like the power looms, that took tasks that workers used to perform. But what the secret sauce of these periods was that at the same time as that automation was going on, employers also introduced new technologies that increased the contribution of workers to the production process, and that’s why they were actually willing to hire the workers and pay them high wages.

Edmonds: And you talk about three decades after World War II, something happens at around 1980, late 1970s, where the benefits of technology stop being evenly distributed. Can you tell me how dramatic a phenomenon that is and exactly what explains it?

Acemoglu: Absolutely, that’s right, and nobody knows the exact date, but sometime in the late 70s, around 1980. And it’s a little more than what you just said, because again, in line with my emphasis a second ago, there are really two aspects to this beast. One is that the institutional structure that is about sharing of the benefits from technology change so that the sharing of those benefits alters. But even more importantly from my point of view, the direction of technology changes, that we start putting all of our eggs in the basket of computers and digital processes that automate work and sideline workers. So, it’s the combination of these two that’s very important. We shouldn’t just focus on unions and deregulation. Those are very important, they have to be a central part of the story, but it’s not just them.

Edmonds: And this is the distinction you make between augmenting and replacing technologies.

Acemoglu: Correct, and my favorite term is ‘creating new tasks’ because I think it really clarifies what the quote unquote augmenting needs to take the form of. It’s not just making a worker more productive in tightening the screws, but it’s really creating new jobs that didn’t exist. And if you look at it today, the majority of the jobs around you are actually things that did not exist 80 years ago. Even what blue collar workers do, if you go to a factory floor, you’ll see that about 50 percent of what they do didn’t exist 80 years ago, because the technical knowledge that they bring to the work process could not have been done with machines of the sort that were available in the 1900s or 1920s. So, it’s this combination of new technologies, workers upgrading their skills and their capabilities and their tasks, and at the same time, we have an institutional structure that doesn’t let the largest firms and the most pernicious managers get away with taking all the money for themselves or a few shareholders.

Edmonds: Give me an example of a technology that’s been augmenting, that has sort of helped workers become more productive and hasn’t put them out of jobs, and a technology that’s been replacing.

Acemoglu: Well, I think, and I don’t want to create the sort of very sharp dichotomy. First of all, even automating, replacing technologies could be very important and very welcome, provided that they are combined with other things. So, I’m not against the automation at all. In fact, automation has been going on for 300 years, and hopefully will go on for another 300 years.

And secondly, even the most augmenting technology might actually do a few displacing things on the side. I think a very good example of a technology that really expanded what workers could do was the railways. So, the railways created a much better communication and transportation structure and led to many new tasks for workers: conductors, mechanics, new repair people, new engineers. So those workers were absolutely critical for the functioning of the earlier railways. But even beyond that, railways expanded what regular workers could do, because now they could get more varied forms of inputs for their businesses, and their output could be sold more widely.

So that’s the kind of thing that expands the capabilities, both for workers and often for firms as well. The quintessential examples of pure automating technologies are the ones like what we have just discussed. Power looms that start mimicking what skilled weavers were doing in the 1810s, 1820s, 1830s. Robots that can now do assembly, painting, welding tasks that previously required workers. And software systems that have taken over many of the office tasks that people used to do. And in all of these cases, again, good riddance to some of those. Some of the ones that have been replaced by machines were pure drudgery. It’s good to see them go, but only if we can replace them with something. So actually go to workers on the factory floor and ask them, would you want to do the things that robots do now? They would say, “No, we’re quite happy because robots are doing the most dangerous, physically most demanding things that now we don’t have to do.” But ask the workers who have been laid off because of robots, and they say, “Yes, of course, we want those jobs back, they used to pay very high wages.”

So, it’s really that we want to do the automation, but not let all of those people go out in the call try to find ways of using technology in order to create new tasks and new jobs for these workers as well.

Edmonds: But those workers would have been trained for the factory floor. It’s not obvious how they would necessarily find work in a much more enhanced technological world.

Acemoglu: That’s a good point, and some of that may not work seamlessly. So, there might be transition problems, but at least their children would have opportunities.

But moreover, the work that’s available in many places is multifaceted. So if you are a low skilled blue-collar worker, but we at the same time introduce lots of other high-tech blue-collar work, some workers would take those, and others may go into construction work, because once you are in a booming economy because you’re introducing new work, you will need more construction services as well.

So, in fact you see this exact same pattern in the – by and large – very successful transition that the US had out of agriculture. Half of the US labor force was in agriculture in the second half of the 19th century. In the course of about 60 years, that went down to 10, 15 perent and then down to 2 percent. So that’s a huge displacement of workers. And how did it happen? It happened because combined harvesters, later tractors and better machinery started taking over the manual tasks that workers produced to perform in agriculture.

So, imagine an economy in which we just do that and nothing else that would have been a disaster. In fact, we’ve seen that because there was large unemployment of agricultural workers in some parts of the economy. But fortunately, not out of benevolent planning, but just things worked out, the US economy at the same time also invested in new manufacturing technologies. That’s the period in which manufacturing boomed, and it became much more sophisticated. It started hiring clerical workers. It started hiring more technical workers.

Now it is true exactly like you said, many of the agricultural workers, which were the least skilled employees at the time, didn’t then go and become clerks doing inventory control in modern factories, but they could find other jobs, and their children could actually take those jobs. And again, there was something amazing at the time in the US economy, fortuitously. There was the high school movement, which meant that there was plenty of investment in high schools, so even children of people from low-income background could actually get a high school degree, which was necessary for doing many of these more sophisticated tasks.

Edmonds: But essentially, in the example you’ve just given, the market eventually sorted this out.

Acemoglu: Right.

Edmonds: Companies care about shareholder value in the capitalist world we live in. Do we leave it to the market and hope that these new jobs will emerge? Or is there a role for the state here?

Acemoglu: No, I don’t think that we should blindly leave it to the market.

But I would also say that I am not very sympathetic to the general statements like capitalism does this, or the market does that, because there are many, many more ways of skinning this cat than just one. There are many different ways in which the market can function. There are many different forms of capitalism. The world in which businesses shared some of the gains with their workers, and in the process also introduced new technologies that increased worker productivity, is a market. And so is the one that’s dominated by the tech giants, where they take over everything, that’s a market. They are very different. Milton Friedman’s world of – the only thing companies should care about is their shareholders and forget the rest – is a market. But so is the ideas of welfare capitalism that people were articulating in the first half of the 20th century, where market economy works better when there is a more cooperative relationship between workers and their employers.

Edmonds: Let’s move on to AI. What are your worries about AI? Do you think AI may be augmenting, replacing? Is there something we can do to ensure that it’s more augmenting than replacing?

Acemoglu: Well first of all, AI, or the suite of technologies that we refer to [as] AI, has a lot of promise. So, I am not an AI denier. But all of the concerns that we have talked about, multiply them by 10, and that’s the AI world that we are being offered right now. The problem with the market that we were talking about that I wouldn’t want to trust much is when it is under the control of a few companies. I’d much rather take the market when there is competition than when Standard Oil owns everything. Well, today we’re dealing with tech companies that are more powerful than Standard Oil.

Also, I contrasted Milton Friedman’s view of how corporations should work, which, by the way, is what our leading managers are taught every day in business schools and by management consultants, so it’s not just one academic’s view. That is very different from the welfare capitalism. So, if we believe that what companies should do is to just help shareholders and nobody else, that’s going to lead to a very different process. And AI today is driven by a belief that’s even more exaggerated than the Friedman view. Let AI destroy everything, and it’s OK if just a few companies collect all of the data and control everything, because they ultimately will know what to do, and then there will be some sort of trickle-down process for the rest of us.

And AI research has become obsessed with artificial general intelligence and automation, so it’s actually leaving a lot of low-hanging fruit about how it could be helpful to workers and to citizens and to consumers. And all of those are the reasons why I feel AI is going in the wrong direction and taking us down with it.

Edmonds: And when you say that the effect of AI could be 10 times the impact of technology in the past, what do you mean?

Acemoglu: Let me put it this way. Henry Ford would have been delighted if somebody offered him a technology that said you don’t need any workers, you can do everything with machines. But that wasn’t the technology he had available. And in fact, Henry Ford was a pioneer in not just electrifying the factory, reorganizing the factory assembly line, interchangeable parts, but he was a pioneer in introducing a lot of new tasks for workers. The things that workers did in the Ford factory floors multiplied, and he was a leader of introducing all this engineering, and many engineers playing more of an important role on the shop floor. So, all of these things were because he was dealing with a technology that did not permit what he may have been tempted to do, which is sideline all the workers.

The same thing is true of robots. You know, robots are amazing. They really cut costs in a few tasks, but they can only do a few things.

Now with AI, at least the ambition is a machine that can do everything, or at least pervasively. So, you will hear from tech titans about how AI can take over all of its jobs. AI can be integrated with robots and then start taking more blue-collar tasks, self-driving cars and AI, all the transport sector. So, the ambition is to do something that previous automation technologies could not even aspire to do.

Edmonds: It sounds like you believe there are two levers to a rosier world. One is a market lever, that’s more competition, and one is a state lever, which is more control in some ways. Can you say a bit about both of those? If I’ve got that interpretation right.

Acemoglu: Right, I would say competition, definitely. But I would also say a more balanced distribution of power in workplaces as well. I don’t think we’re going to go to a better future unless workers are represented and can bargain collectively. And, critically, I don’t think we’re going to go to a better, rosier future unless we undertake a fundamental redirection of technology. And it is in that context that the state comes in.

The market could do a recalibration of technology, but I think the current incentives provided by the market, especially with the largest tech companies having reached such domineering positions, is not very likely. So therefore, a better regulatory system, as well as some sort of state leadership in the agenda for technology may be necessary.

But let me be clear, I do not ever advocate state control. By state control what I mean is that the state decides what we do. So for instance, the state says this is the good technology, this is the one you should adopt. No, I don’t think that ever works, and it wouldn’t be feasible in the context of the digital technologies that we’re working with or we’re thinking about. What is necessary is the soft hand of the state in tipping the scales one way or another. So, if there is a problem with fossil fuels, which there is, we are emitting a lot of carbon and frying the planet, what the state should do isn’t decide exactly how you should run the energy sector but introduce two sorts of policies. One is market-based policies such as carbon taxes, that’s going to be very useful. But that’s not going to be enough, so the state should tip the scales for innovation as well. So provide funding subsidies, new possibilities for green innovations, and not for fossil fuels, right.

Now, actually, you know, and again, it’s not like we start from a situation which is pure laissez faire, à la the French physiocrats, or Adam Smith. It’s already a very heavily regulated economy. In that regulated economy, actually the US government and some European governments heavily subsidized fossil fuel companies. So, what would be necessary is change that. Look at it in the US context now. Actually, the US government heavily subsidizes automation. How does it do that? Well, the tax system in the United States favors introducing machinery instead of people. If you hire people, you pay 30 percent tax. If you hire machinery, you pay zero tax. These are the kinds of things that we can easily change in order to alter the direction of research, the direction of innovation in the market economy.

Edmonds: So, you disincentivize technology that replaces labor, effectively?

Acemoglu: Well, I would say I would get rid of the excessive incentives for replacing labor, such as through the one in the tax system, and I would create new incentives for blue-sky technologies that can create new tasks, new competencies. But again, I want to emphasize you don’t want to completely disincentivize automation, because what you want for the future is a balanced portfolio of automation and new tasks. Automation is not our enemy. Excessive automation is our enemy.

Edmonds: How does that work in a global world, when the Chinese companies are possibly replacing all labor and competing with US companies who are perhaps keeping labor which is more expensive than technology?

Acemoglu: Well, first of all, US has a tremendous leadership position in many technologies. So, where it leads, others will follow. It’s not a unilateral decision. You know, European decisions, European regulations really matter, and Chinese ones do matter. But US has a lot of leeway.

Second, US is the one who’s behind in terms of regulation. The Chinese government, I do not like their policies, I do not like their objectives, I do not like their heavy-handed approach. But China shows how you can regulate AI. The AI sector is quite vibrant in China, but it’s very, very heavily regulated. The government decides who can play games at what time, the government is the main purchaser of AI technologies, and AI investment is booming in China. So I’m not advocating any of that, but I’m just pointing that out to say that the argument, well, the moment you do the smallest amount of regulation, China is going to surge ahead, is complete bogus.

And then the third point is that China has the fastest demographic transition world ever seen. Of course they’re going to introduce robots. Of course they’re going to introduce AI to automate things. And they should, given how quickly their workforce is dwindling. But that’s not an advantage, it’s a disadvantage. Their labor is becoming very expensive, very fast. So, you know, we are the odd one out in the United States, and to some extent in the UK, where there is plenty of labor that’s clamoring to get into good jobs, but we’re not leveraging the technology to create those good jobs.

Edmonds: The tech billionaires have a lot of power, particularly in the US, and they’re going to try and lobby and sway decisions in their own favor. Are you an optimist or a pessimist about the future you would like to see?

Acemoglu: I’m neither an optimist nor a pessimist, but I retain a small degree of hope. Humans are extremely versatile, changeable, adaptable and creative. I don’t think anybody who lived in the last decade of the 19th century or the first decade of the 20th century in the United States would have said that, you know, reforming that system was a walk in the park. It was extremely unequal. Companies in the railway sector, in the financial sector, such as JP Morgan, steel, Standard Oil, were hugely influential. And they had their own senators, you know, senators were in the pockets of many of these companies, especially because there were no direct elections for the US Senate. The federal government had no tools of regulation. They did not have a federal income tax. And the media was controlled by the robber barons as well.

But in that environment, it turned out to be a very, very powerful force to reveal the kind of abuses that these companies were committing, and it led to the coalescing of the populist movement together with a more urban middle class movement into the progressive movement. And that then completely remade politics in the United States, you know, influencing what at least three presidents did, and leading to many landmark legislations.

So, what was done then can be done now, but it’s not easy. So, I wouldn’t say I’ll put money that within the next five years, we’re going to have a landmark legislation in the United States. And indeed, the tech sector has real tight control, not just because they are the biggest or one of the biggest lobbyers in Washington, DC, but the tech sector completely controls media in the United States. Journalists are largely allied with the tech sector for a variety of reasons. They’re mesmerized by them; they get funding from them. Newspapers themselves have turned into social media companies to some degree. So, it really is very difficult to shift the balance against the tech companies in the United States. But you’re seeing the beginnings of that at the moment, actually, despite that difficulty.

Edmonds: Let me get a prediction from you. What do you think the unemployment rate will be in the United States in 10, 15 years’ time?

Acemoglu: Well, I wouldn’t want to make that prediction, because I think, as with many other things, it is not predetermined. It will fundamentally depend on what we do with technology. If we continue to use AI and other digital technologies to automate work and not create new tasks, I think there’s going to be a shortage of work, and that’s going to show itself via lower labor force participation, people doing more menial, meaningless jobs. And a variety of other things, low wages as well. On the other hand, I think AI has the capacity to create these new tasks, augment labor, and if we go down that path, there is no reason for future unemployment to be just like today. You know, around 3, 4, 5 percent, as long as the economy is functioning well. I think there is no shortage of potential work, potential good work that humans can do. It’s just a question of us investing in it.

Edmonds: Daron Acemoglu, thank you very much indeed.

Acemoglu: Thank you, David. My pleasure.

Edmonds: Social Science Bites is made in association with Sage Publishing. For more interviews, go to socialsciencespace.com.