Economist Daron Acemoglu, , who was awarded the Nobel Prize today, delivered the following speech at the conference “The Divided Society After November 3rd,” held on November 23rd and 24th, 2020, in collaboration with the Centro Studi Americani and the Italian Academy at Columbia University. To watch the full panel click here.
I am going to discuss the question of automation and the prospect of a new society. We have a huge gap in status and capability inequality that has emerged over the last 30 years, and I am going to argue that the technological and economic trends have been at the root of many of the fault lines and have exacerbated those cultural backlashes [that the United States is currently experiencing, ed.]. In particular, I will emphasize the role of automation, but let us start with something tangible, that in my opinion, does not receive enough attention, which is that if you look at overall labor demand in the US economy and the picture is similar in many other advanced economies, there is really a sea change about what we experienced in the four decades after World War II and what we have experienced over the last 30 years and especially over the last 20 years.
These images reflect what economists call the wage bill of the private sector, which is total employment times average wages. Figure 1.1 is a remarkable picture that shows very steady growth of about 2.5 percentage points faster in the private sector wage bill than population every year for about 40 years, which translated, of course, into more than two percent real wage growth for American workers. Figure 1.2 shows that over the last 30 years, the same quantity grows more slowly than before and then flattens out and there is no change from around 1999 to today. So, the private sector is not generating anything approaching the labor demand growth that we experienced in this economy. This has various implications, not least the decline in the labor share in the economy; capital captures much more, has greater power over labor, but also the huge increase in inequality that has happened around the same time, and is causally linked to this slowdown in labor demand. That indicates that there is something quite deep that is opening, like a caste structure in the US economy. One way of seeing this is by looking at changes in real earnings by gender and education.
In figure 2.1 we have men and in figure 2.2, women, with different colors corresponding to different education groups. In the 1960s and early 1970s and, similarly, in the 1950s, the picture seems to be that of a “rising tide that lifts all boats;” that very rapid growth of labor demand translated into commensurate real wage growth for all demographic groups. Remarkably, the postgraduates and the high school dropouts experienced the same real wage growth, but then two notable things happen. Firstly, both for men and women, sometime starting in the late 1970s, early 1980s there is this big fanning out, the gap between the top and the bottom widens, but also, perhaps as consequentially, the real earning of men with low and middle education drops sharply. The red, for example, the high school dropouts or the orange high school graduates, even those with some college (two years of college, community, or associate degrees) and those without postgraduate degrees is quite flat. It is really a labor market in which the post-graduates, those with very specialized skills, are benefiting. For women it is a little better because they are still catching up during this period, but the increasing inequality is very similar, both for men and women. There are huge social costs of inequality and falling real incomes of low education men, and I think some of the cultural backlash and polarization is related to this. But it is also raising the risk of a new caste society, not just because inequality is increasing, but because the status of people who do not have those specialized skills has also fallen very sharply during this period.
Such a complex phenomenon has many roots. Globalization is part of it, as well as the disappearance of institutions protecting low wage workers. But more important, both for the phenomenon that I have demonstrated so far and its ramifications into a new caste society, is that the technological changes are largely underpinning these developments. And one way of thinking about this is that in contrast to what many economists typically go back to, which is “technology is always good and it will generate gains that are going to be shared”. We can think of different types of technologies having very different implications for labor demand and very different implications for different groups, in particular, automation. Those types of technologies that replace work tasks previously performed by middle skilled workers, low skilled workers, and manual workers; those are going to have very different effects.
In the work that I have done over the last decade, I have also looked at how this is changing over time. And this evolution, again, is remarkable because in figure 3.1, during those four decades where there was rapid growth of labor demand, there was a lot of displacement, shown by the dashed line: workers losing out relative to capital, but it was broadly counterbalanced by what we call reinstatement: new technologies, new tasks that are creating new opportunities for humans. So much so that the sum of those two is the thick blue line in the middle, hovering around zero showing that the two are broadly counterbalanced.
Look at the last 30 years (figure 3.2) and there is a completely different picture: much faster growth of this displacement and much slower reinstatement. We are putting all our eggs in a single basket and the thick blue line is going south. This may sound abstract because here we are aggregating different types of technologies and we are leaving out perhaps some other aspects of technological change. But we can focus on the specific example of robots. Robots are the tip of the iceberg because most automation takes algorithmic forms such as in the in the service sector. But robots are something that can be concretely measured. Let us analyze what happened local labor markets, where there was faster adoption of robots.
Figure 4 demonstrates that even though robots are one of the most innovative manufacturing technologies of the last three decades, the gains have been extremely unshared. In places where robots have been introduced, and productivity effects appear to be positive when firms introduce robots, labor demand and wages have fallen. Figure 4 does not depict wages, it only reflects private employment, which is led by the industrial heartland of the United States such as those listed in the above figure, Detroit, Wilmington, Cleveland, Toledo, the heavy industry heartland of the United States. Those are the places, of course, where we are witnessing this cultural backlash as strong as in other places like Beaumont, Texas. This is part of where a lot of these cultural tensions are being seen. These places are not driving the relationship, the cloud of points is the same slope if one were to do a regression through it, but they are emblematic of the fact that automation, if not counterbalanced by other opportunities, is creating these tensions both in the labor market and in the social sphere. One can say that robots are a technology of the past and that the future is AI and AI is different to robots. This is correct, AI is a broad technological platform which can be used in myriad ways. It can be used for increasing human activity and empowering humans. But in practice, the problem is worse with AI. It is worse because of two aspects. First, the evidence so far, suggests that AI adoption has so far been driven by the same business model that aims to substitute algorithms for humans. But, it is also worse because by virtue of its greater scope, AI is going to deepen the status inequality, it is going to deepen the new caste.
Those who are going to have a voice, politically, socially, through AI-mediated interactions and those who are going to be discriminated against by machines and algorithms are going to be silenced, are going to be monitored much more effectively because of AI. So, AI is at the same time, even though it could go in a very different direction, is going in the same direction as robots, at least that is the early indication, but it is also going to change the democratic process that in the past, during similar episodes, was one of the most powerful ways of reversing trend tendencies towards creating new inequalities.
The question that will be critical is who will control technology? Who will control AI?
So far, the answer is that a handful of large companies are controlling technology and AI. McKinsey Global Institute estimates that two out of every three dollars spent on AI are coming from big tech in the United States and a few Chinese companies. So it’s a huge concentration of research in the hands of a small number of companies and it is their visions that is shaping it. Global competition is another factor that is pushing companies towards focusing on automation, but we are also making things much worse with our tax system.
Looking at the case of the US, figure 5 shows that labor has traditionally been taxed more heavily than capital. But over the last 20 years, labor taxes have been roughly constant at around 25 percent because of income taxes and payroll taxes. But taxes on equipment and software, the types of capital that are involved in automation, algorithmic or otherwise, have fallen sharply because of depreciation allowances, because many corporations evade taxes, and because of the cuts in the marginal income tax rates of the capital owners. So instead of 15 to 20 percent, capital is currently taxed at less than 5 percent. Today, many companies receive a net subsidy when they automate work and substitute machines in the workplace.
One argument that I find unconvincing, even if it were true, is that this is the path of technology and is going to create benefits and we endure its negative influences, such as inequality and disempowerment for some workers. Well, the truth is actually very different. Not only are these consequences very costly, but they are unnecessary and moreover, we are not getting much of a productivity gain from these technologies either. If you look at patent statistics, we have never generated so many new widgets as we are doing now; patent statistics are exploding.
If one at figure 6 we can see that what we are getting as an economy, or as the sectors that are adopting new technologies, is not much. The headline number that economists use, total factor productivity growth, in the 40s, 50s, and 60s that was as high as 3 percent, sometimes 2 percent. Today it hovers around 0.5 percent so, we are not seeing the fruits of this. In fact, that is part of the reason why private sector labor demand is so anemic. We are hit doubly: productivity is not increasing, and we are using all our creativity for automating work. We are not reaping any of the benefits that were automatically received in the 50s and the 60s.
The COVID crisis is creating many problems, but it will undoubtedly exacerbate the problems just articulated. More than 75 percent of companies say they are going to take actions towards further automation and of course, many of the occupations of low-wage workers are coming under increasing pressure.
The argument that I want to make is that the way to build a new society is to first recognize that the future does not need to be fully automated. We can use many of these technologies in different ways, both in the workplace, to increase human productivity, and for democratic governance, for communication, for the privacy and protection of individuals against companies and governments, rather than increasing the power of companies and governments. But this can only be done if we have a new institutional framework to guide us. This is not easy at the best of times. It is made even more difficult because looking at history, typically the creation of new institutions, especially the new institutions that protect those who are weaker and less powerful; the disenfranchised or the disempowered, comes from a combination of two things: grass roots demands and government action that introduces regulations to protect them. These are not functioning together. We have the malfunctioning of democracy, and we can recognize the widespread collapse of trust in institutions that makes it very difficult for them to be used for anything for anything like this. And the erosion of expertise and autonomy within institutions is another exacerbating factor.
Essentially what we are facing, is a situation in which we need new responsibilities from the state, new responsibilities that have never been tried before, because even at the heyday of the welfare state, which was mostly concerned with macroeconomic management: social security, social safety net, and labor relations. However, today we need something much more than just regulation of technology, but regulation of globalization and international coordination as well. These are extremely onerous new responsibilities on the shoulders of the state. This requires the state to have the capacity, the expertise, and the trust to carry out those responsibilities, but it also requires that we keep the state in check once it has been so empowered.
This is what worried one of the most influential social scientists of the 20th century, Friedrich Hayek, who wrote The Road to Serfdom in response to the 1942 Beveridge Report which was a template for the UK version of welfare 1.0. He was worried that a new state that was administratively so powerful could not but turn into a totalitarian state. In the end, Hayek’s concerns did not come to pass and I want to provide one perspective about why Hayek was wrong and what small hope there may be in trying to forge this new institutional framework. I have elaborated this idea in my book with James Robinson, The Narrow Corridor. This is what we call the Red Queen effect, and this figure (below), which is emblematic of the theory of the book, and lends it its story, that we think of democratic institutions, new capabilities for society, new ways of conceiving and practicing liberty as evolving in this narrow corridor illustrated in figure 8, which is defined by a balance between the power of the state (its capacity and its repressive apparatus) and the power of society (its ability to withstand the demands of the state through collective action, through norms, traditions, and of course, democratic voice and participation).
The key here is that while outside of this corridor, one dominates the other, inside the corridor, there is a balance where each must run in order to keep up with the other. That is the dynamic that in some sense, Hayek did not fully recognize. He recognized that there is a conflict between State and Society and that was his big insight that resonated with many people making The Road to Serfdom such an influential book throughout the second half of the 20th century, but what happened in Scandinavia, for example, after the Workers’ Party came to power in 1932 in Sweden or 1935 in Norway, is that despite the fact that the State was becoming far more powerful and far more interventionist, democracy deepened. There was a greater participation in politics, and that is what we mean by the Red Queen effect. That is exactly what happened in much of Western Europe and in some bastardized form in the United States as well, and the fact that it was bastardized is because of the inequities that existed in social status and social participation, especially in the US South, but also because of the weakness of the federal government in the United States. On a hopeful note, perhaps there is a way of proving Hayek wrong again. Though it may not be easy, it requires what we have not been able to do so far, which is articulate new ways of thinking of how the state can play a useful role, find the right coalitions for it, but at the same time increase the power of society so that the society keeps the new and more interventionist and more powerful state in check as we develop.
Cover photo: Academy of Sciences permanent secretary Hans Ellegren (C), Jakob Svensson (L) and Jan Teorell, of the Nobel assembly at the Swedish Riksbank announces the Swedish Riksbank’s prize in economic science in memory of Alfred Nobel 2024, which goes to Daron Acemoglu, Simon Johnson and James A Robinson, during a press meeting at the Royal Swedish Academy of Sciences in Stockholm on October 14, 2024. (Photo by Christine Olsson / TT News Agency via AFP)
Follow us on Facebook, Twitter and LinkedIn to see and interact with our latest contents.
If you like our stories, events, publications and dossiers, sign up for our newsletter (twice a month).