As the next generation of robots arrives in the workplace, will they enable workers or replace them? According to MIT’s Daron Acemoglu, one of the most frequently cited economists in the world, this distinction is the difference between technology that raises workers’ wages versus tech that reduces overall employment and stifles wage growth.
Daron Acemoglu is a professor of economics at MIT, a frequent contributor to Foreign Policy Magazine, and co-author of the book Why Nations Fail: The Origins of Power, Prosperity, and Poverty. He joined me to discuss his recent research and what technological innovation means for the future of work.
Some of your recent work has looked at how the increasing use of machines will affect workers, and your conclusions aren’t too optimistic. In particularly your paper “Robots and jobs: Evidence from US labor markets” shows how robots have a negative effect on both employment and wages in some communities. This finding has been treated with some alarm and I see people using it as evidence we’re headed toward a jobless future, the robots are going to gobble up all the jobs, etc. And one reason I think there is this alarm from your paper — aside from the fact that we’ve seen the rise of new technologies at exactly the same time that we have a weak recovery after a recession — is that the standard Econ 101 perspective might be different than your finding from the paper.
So first I want to give a review of what that standard perspective is. I got a report from a consulting firm that was very optimistic about our future with robots, AI, and jobs. Here is the happy ending the consulting firm gave me:
New automated technologies will boost productivity considerably over time. This will generate extra income, initially for the owners of intellectual and financial capital behind these new technologies, but eventually feeding into the wider economy as this income is spent or invested in other areas. This additional demand will generate more jobs and increase incomes especially in sectors that are less automatable. So, the historical evidence suggests that automation will eventually lead to broader, similar overall rates of employment, and higher average real incomes across the country, though there might be a different skew to the income distribution.
So that is a fairly happy ending despite all the concerns of the rise of the robots. To what extent do you agree with this happy ending scenario?
I think that’s something we don’t know about. I think there is a lot of comfort in thinking that there is an inexorable link between productivity, wages, and employment, and everything is going to be fine; but, there are so many unknowns. I would interpret my research here, and it really goes perhaps even against my own priors, as concerning but not alarmist.
Alarmism would be: We’re losing jobs so quickly that within the next generation’s lifetime, we have some major problems if we don’t do something about it. That’s not the case. The rate of job loss from automation or this particular branch of automation, robotization, is relatively small over the last 20 years or so, perhaps less than half a percentage point of the population, (the working age population). So, we’re not talking about huge numbers.
I think what the result signifies is important to put in a broader context, and that was your question. The way that Econ 101 thinks about productivity, employment, and wages, is that whatever increases the productivity is going to translate into wages relatively quickly. And we used to chastise Keynes, one of the greats of economics, for predicting in 1929, before the Great Depression in a speech that he gave, that rapid technological changes would reduce demand for labor and lead to technological unemployment, or much shorter working days or working weeks for people. And that hasn’t come true obviously in the intervening 90 years.
But, the conceptual point that new technologies could reduce the demand for labor, rather than increase the demand for labor, is a possibility and it’s not a crazy idea. There are many conceptual reasons why that could be the case, and I think what makes it more likely is precisely new technologies that automate — meaning replace tasks that were previously performed by labor. And it’s not a crazy thing because there are other examples in which this happens. Almost one hundred years from the beginning of the British Industrial Revolution to the middle of the 19th century did see an unprecedented rate of arrival of new technologies that totally transformed how production, especially textiles but also in other industries, took place. But we didn’t see pretty much any change in wages, and that led to what is sometimes called the living standards paradox.
I think that is another example in which many technologies that replace workers in important tasks in the economy do not necessarily translate into higher wages and higher labor demand immediately. And I think the important point here is a concern but not alarmist, because I think how we deal with these technologies, and whether we can turn them into things that essentially lift all boats very much depends on our institutional and educational responses to them.
The British economy, in the middle of the 19th century, started a breakneck pace of institutional transformation, things that we had not seen before. You know, a welfare state of sorts, taxation, unions, and it also started investing in skills at a very high rate by historical standards. I think those were not unimportant in turning this wave of technologies into something that increased wages for the average working man of Britain. So, that’s why this is the right time for having this conversation about what these new technologies are doing, what we expect them to do, and how we can work toward making them work for us.
I have mentioned technology, you have mentioned technology, but one of the points that you make in your work is that not all technology is the same. And in fact, you differentiate between what you like to call “enabling technology” versus “replacing technology.” Can you talk about the race between those two things, and how that race is changing?
Absolutely, I think that is exactly the crux of the matter. Enabling technologies are essentially what we focus on in basic economic analysis. Those are the things that make the workers more productive in tasks and functions that they are already performing, and perhaps even also expand those tasks. They tend to increase wages and labor demand because they are making workers more productive. When you look at the details, it’s more complex because when you make workers more productive, sometimes you need fewer of them in some activities, and they can get reallocated to other activities. But at its root, this type of technology is directly making workers more productive; but this is very different from replacing technologies.
What would be an example of an enabling technology?
An example of an enabling technology that I like to give — but I think every technology has some grey areas — is the computer assistive design machines, or new spreadsheets. Those take some very specific types of workers. For computer assistive design machines, it would be design workers, and this type of technology increases their productivity. Now they can design things much more precisely, they don’t have to waste time doing repetitive tasks, and it boosts their potential capabilities. Spreadsheets, the same thing. If you are a mid-level or supervisory worker, or essentially a non-routine clerical worker, spreadsheets, early on in the 1980’s, really expanded your capabilities quite a bit.
In contrast, replacing technology directly displace workers from the tasks they were previously performing. A good example would be ATMs; ATMs took away a whole range of tasks that bank tellers were performing, so now we don’t need the bank tellers for those tasks. Now it doesn’t mean that ATMs don’t have the capability or the potential for raising wages, they could do it through the optimistic scenario that you outlined in the beginning of our conversation. Now ATMs make the business of bank transactions much more productive, and as those become more productive, we want to consume lots of other things. We want to perhaps now consume advice from bankers about financial planning, and we also might want to consume more coffee and more cars, and things like that, because the banking sector has become more productive. And that increases labor demand in other activities, and that might be sufficient to offset the negative effects of displacement, or even surpass them.
What I hear about most of all, and what you also focus on in the paper, is robotics and artificial intelligence. Are those — and again technologies in some ways can be enabling and maybe also replacing — but are those technologies fundamentally replacing technologies that are going to result in job loss and probably stagnating wages for some people?
I think the odds are in the favor of these technologies being more on the replacing side. If you look at robotics, what that technology does is it automates and increases the ability of machines to do relatively more complex tasks than manufacturing workers were performing before. AI is very much geared toward applying machine learning to big data, to unstructured data, in order to replicate human judgment and human decision-making in a variety of things. And according to the experts, we are getting to the point where we can use AI to perform tasks that were previously done by accountants, financial planners, paralegals, definitely customer service, and the clerical occupations.
As these technologies advance and spread, if I was just going to look at the big economic numbers that get released every quarter, I might see GDP go up, I might see productivity go up, but that wouldn’t necessarily translate into incomes going up or good paying jobs being created then.
Exactly, I mean, obviously these technologies have a great potential for increasing productivity. Automation is about performing a variety of tasks and functions much more cheaply, and that will translate into greater GDP, greater riches for the nation. But it will not raise labor demand and it might harm a great many workers. Now, the puzzle, of course, is that over the last 15 years, or perhaps slightly longer, productivity numbers have been doing badly.
Right, I’ve been writing a lot about this productivity paradox, in which we certainly read about these great new innovations, whether it’s AI or robots — now everyone has a computer in their pockets, Silicon Valley is going crazy with startups — so it seems like there’s all this innovation and technological progress. But when you look at our productivity numbers, it looks like total stagnation; they’re flat-lining, and they actually started to weaken before the Great Recession. So, how do you think about that productivity paradox and how do we solve it?
I think this is one of these questions that is so puzzling that nobody has the answer to it. But I think it is important to get some facts on the table; back to like you said, this is not about the Great Recession, this is not about a cyclical phenomenon, this is something that we’ve been living with. Into the 2000s and perhaps even in the late 1990s, there was a pickup in productivity in the 1990s but even then, we’re not in the age of very rapid productivity as in the 1950s or the 60s.
Now, the second thing is, people point out the mismeasurement of productivity. I think that’s definitely something to be taken into account and we need a much better measurement. But it’s not as if the Bureau of Labor Statistics and the Bureau of Economic Analysis are idle. They are doing a great job in terms of trying to measure productivity as well as we can. And I think the evidence is, we’ve always mismeasured productivity to some degree and there isn’t a compelling reason to think that it’s gotten much much worse, that it explains all of this slowdown.
So, I think the big issue is that new technologies are coming, but it takes time for us to use them in the most productive way, and that’s for two reasons. One is, we need to adapt a whole host of supporting institutions in order for the productivity of these new technologies to be realized. If our organizations don’t work well with these new technologies, if the workers don’t have the right skills for these new technologies, that’s going to hold back the gains that we can make from them.
But the second one is that, in fact, the technologies are not sufficiently pervasive — yet. If you imagine our cars, the amount of new technology, high-tech gadget that are in our cars, it’s just mind-boggling. But you cannot drive our cars any faster than our grandparents did. Why? Because if you want to drive much faster, you need non-congested roads, much better tires, and there are going to be a whole host of other institutional limits for our ability to drive much faster. So, at this point, we can pack better and better software into the cars — perhaps our air conditioning can get better, perhaps our stereo system can get better — but the thing that we really want from our cars, which is get us faster from A to B, that’s not changing.
Now, the next wave can start transforming that, now you can have many more innovations together with the organizational adjustments, institutional adjustments, and we can go to driverless cars. And that can suddenly change the equation because even if the cars cannot go faster, the cost of doing that sort of driving is going to go down a lot. And that’s going to start changing. So, what that suggests is that there is a cumulative aspect to all of these innovations. And it may well be that we are still in the middle of this cumulative aspect.
I have seen numbers saying that you have productivity inequality, where you a smaller number of leading-edge firms which seem to be very productive and very innovative, and then everyone else is getting left behind. Could it be that we are innovative and we are productive but it’s limited to superstar firms, and either there’s not enough competition so other companies aren’t taking up these innovations, or they’re hard to figure out, and these other companies just figured it out sooner? Is the future already here, just not evenly distributed?
I think that’s very important. I think the business of innovation is probably becoming more unequal and the phenomenon of superstar firms, as you’ve mentioned, is a real one. The US economy has become much more concentrated than any time for which we have data to measure concentration. There’s a recent OECD study which shows that if you look at the top firms in various sectors, and this is tracked over several OECD economies, their growth rate seems to be very similar to what it was in the 1980s and the 1990s. But if you look at the next tier of firms, the follower firms so to speak, they have slowed down.
So that again highlights the inequality in the technology treadmill or the inequality in the innovation capabilities of different firms. And that may have institutional reasons, but I think that it might also be a consequence of the nature of these new technologies. Once Google starts dominating all information, it’s very difficult for another search engine or for another company who wants to use another similar innovation for AI or for customer service to catch up with it, because Google has this superb war chest in the form of terabytes and terabytes and terabytes of data that nobody else can access.
Just to jump back for a second to this idea that you have two different kinds of technological progress, replacing and enabling, it reminds me very much of some work done by Clayton Christensen over at Harvard, in which he talks about the different kinds of innovation. You have efficiency innovations, which is sort of classic; it’s a new technology, it replaces workers, overall it just reduces the cost of making existing products. Then he has what he likes to call sustaining innovations, which replace old products with new, such as a Toyota Prius replacing a Camry or something. And then he has what he calls empowering innovations, which are brand new products and services creating new jobs. He gives the example of Ford Model T or the Sony transistor radio. Could it be that the technological progress we’re making is just more of the sustaining or efficiency kind of innovation, and we’re just not generating enough of this enabling or what he calls empowering kind of innovation?
I think that’s exactly right. I think we are not creating the types of technologies that are going to use our available labor force. And I’m not sure the process versus product is exactly the right division. But obviously, many process innovations, exactly as you have suggested, would be of the replacing kind; but there are also process innovations like computer-assisted design or spreadsheets that increase our productivity, but do so in a way that enables existing workers to become more productive.
But it is also the case that many of the important products that we have created over the last several decades, over the last two decades in particular, have not added much to the bottom line employment figure. Think of Apple’s iPods and iPads and iPhones; they are amazing innovations and consumers have absolutely rewarded the company by purchasing billions of them. But, look at the number of employees working in the United States for producing these products — it’s very very small. So, there we have an intersection of new technologies that have a very heavy design component and the division of labor can be very finely broken down, and the labor-intensive parts of those products can be manufactured abroad. So, there is a sort of a parallel process to automation that’s increasing efficiency but it’s not really adding to the bottom line employment figure.
Is that a problem that can be fixed by policy? Can we rejigger things to create or change the balance? Move the needle more toward enabling and empowering kinds of innovation and technological change? It seems like we need more of that, but does that just fall like manna from heaven?
I think very much policy can play a role, but not an obvious one. Because I think there’s a lot of evidence that shows the exact composition of innovation and therefore the exact types of technologies that firms develop are responsive to policies and to financial profit incentives. The problem is, it is difficult for us to describe or recognize exactly enabling technologies when they are in the incubation period; so you cannot say we’re going to subsidize the enabling technologies. It was much easier for us to recognize green technologies, but even there, our track record of giving subsidies to green technologies is a mixed one. These sorts of fine incentives are very easy to game.
But I think there are other problems in the labor market and the innovation market that we can start thinking about. We implicitly subsidize production with machines relative to production with labor. If you buy a machine, you don’t have to pay payroll taxes, you can debt finance it, and that’s going to get a subsidy from the government. And the capital income that is generated is going to be taxed at a lower rate. If you hire a worker, you have to pay payroll taxes, they are going to be taxed at a much higher rate, you have to put up with lots of other costs coming from regulations, and other things you have to do when you’re employing workers. So that creates a very non-level playing field, so we’re essentially subsidizing firms implicitly for using machines rather than labor. So I think it’s probably a good idea to start thinking about how we can even that playing field a little bit. So I am definitely not suggesting a robot tax so that we can raise revenues.
That was my next question about Bill Gates and the robot tax.
I can comment on that, but I think what I’m suggesting is, let’s not subsidize robots and other machines implicitly while making it much harder for firms to higher labor.
I have heard people suggest though, when they look at those productivity numbers they say, “Well, gee, that to me looks like companies aren’t using enough machines, so maybe we need to dramatically raise the minimum wage, make workers more expensive, and then companies will, say, use more kiosks and they’ll become more productive and down the road, we’ll have higher growth and higher productivity.” So there’s also that flipside argument that I hear.
Well there is some of that going on, you know, certainly if you look at countries where wages are higher and workers are scarcer, you see more automation. So the US, actually, is a laggard in automation. South Korea, Japan, and Germany are far ahead of the United States in the use of robots and perhaps more worryingly, if you think that robots are part of the future, we don’t have any of the major robot producers in the world; they are all in Japan and Germany, so we are lagging behind there.
But I think taxing employment and workers further would definitely be at the very very very bottom of my list of good ideas. Because I think the problem we are going to be facing in the next several decades is how can we encourage people to be part of the labor force. And I think the United States does very well on that, relative to other countries, partly because of the earned income tax credit. So I think certainly the minimum wage has a role in making sure that some workers are not paid very very low wages, especially if you take workers who don’t have any other option so they might be subject to monopsony power. But I think jacking up the minimum wage so that you force firms to sort of automate more — I think that probably is going to be very distortionary.
And we did mention the robot tax, which is interesting because when Bill Gates mentioned it, he mentioned it both as a source of revenue to help with retraining, but he also specifically mentioned it was a way of slowing down progress and giving workers more time to adjust. I mean it makes a difference for truckers — which you always hear about now when you hear about automated vehicles — it matters whether we go to level 5, full automation in the trucking industry over 10 years versus over 25 years. So do you have any sympathy to slowing down the pace of automation to give workers more time to adjust?
Well what I would say is three-fold. First of all, I believe that if we are indeed, as I just have tried to articulate, subsidizing capital, then that’s distortionary, and that’s a bad idea. We should be having a level playing field. I think we need to go much deeper into this and look exactly at various different types of technologies and see if some of these numerically controlled machines or other automation technologies, as well as robots, are being introduced precisely because labor is being artificially made more expensive. And if that’s the case, there is an obvious thing for us to do.
Second, I totally agree with you; the issue is one of adjustment of labor. So if labor is suddenly caught unaware about what’s going on, and is thrown out of work because of automation, the costs of that are very high. Detroit is the case in point. But, we know self-driving cars and self-driving trucks are coming, so it’s not as if it is going to hit anybody as a big surprise. It’s just that US society does not have the institutions to prepare either the workers currently in this occupation, or even worse, the youth that will be graduating from high school or college in the next few years. We’re just not providing them with the human capital and training opportunities and vision to prepare for them so that they can work with the machines rather than try to do what the machines are doing.
So slowing down the progress, I think, yes and no. If the problem is we are totally unaware and this is hitting us as a surprise, there might be some adjustment process that might help us. But I think, at the end of the day, these technological changes are also our future; we want to have rapid technological change because that’s where the productivity gains are coming, and that’s where productivity gains are going to come from. So we don’t want to slow down the technological progress, we want to turn the technological progress and our skills so that they can work with each other.
We’re already going long, and we did not even get a chance, unfortunately, to get to Why Nations Fail, which I think probably has a lot to say to us these days in the United States. So I just want to end as I always do. I go online and I ask the Twitter-verse for questions, so here’s that question: “The thing that every middle-class parent wants to know is, dear God, what should my child major in when they go to college, given our technological progress and automation?” So what should the kids be majoring in and studying to deal with this future? Do they all have to be computer science majors?
I think that’s a great question and you know what the funny thing is? That parents are asking this question, but policymakers aren’t. Who in the current administration is worrying about such things, for example? Nobody. I think we don’t know the answer to that. We know certain things: We know that flexibility is a great asset in the current labor market and will become a bigger asset, and a more valuable asset in the decades to come. That we want students who can adapt to different circumstances, who can reconfigure themselves to use and deploy different types of skills, and that’s very important, and that’s not something that our high schools or even our colleges do a good job in. But if you go in greater detail, do we need workers who have numeracy skills? Do we need workers who have better software skills? Do we need workers who are better at teamwork?
I don’t think we know these things. Obviously, there is going to be a small elite group of workers, perhaps 2 or 3% of the US labor force, who are going to work in computer programming, designing these AI, big data machine learning programs, or designing new products, and those workers need to have all the computer science and all the technical expertise they can get. As well as the vision; I mean, it’s not just about knowing how to program. Steve Jobs didn’t become Steve Jobs because he was the best programmer. He had a vision, he had a way of conceptualizing the product that other people could not, so that’s actually a broad skill.
But let’s take somebody who’s going to work in the financial industry. Do they need to know much better programming so that they can seamlessly program the AI machines? Possibly, but probably not. Probably the AI technology is going to be developed enough that people who use it don’t actually need to reprogram it. But they may need to have a broader set of skills so that they can act as the conduit between these programs and the customers. They may need to have much better teamwork and soft skills in the new labor market. We just don’t know that because we have not been studying it.
So in some sense, this is reemphasizing what I was trying to say earlier on. The changes that are going to come in the next 20 years shouldn’t surprise anybody, but we are totally unprepared for them.