May 2019. GrowthPolicy’s Devjani Roy interviewed David Deming, Professor of Public Policy, Education, and Economics at Harvard Kennedy School and Harvard Graduate School of Education and Director of the Harvard Inequality and Social Policy Program, on higher education and social mobility, the labor markets of the future, and solutions for income inequality. | Read more interviews like this one.
Related Links: David Deming’s faculty page | Harvard Inequality and Social Policy Program | CLIMB (Collegiate Leaders in Increasing Mobility) Initiative | NBER research page | Google Scholar page | Twitter
Growthpolicy.org. How should we promote economic growth?
David Deming: We should promote long-run economic growth by investing in the education and basic skills of our population. At the turn of the twentieth century, only one nation on earth had begun the transition to mass secondary-school education—the United States. One hundred years later, secondary education is near-universal in the U.S. and the rest of the developed world. In the U.S., we hear a lot of hand-wringing about whether everyone really needs to get a college education. I imagine—although I do not know for certain—that the same concerns existed about high school a century ago. There is no natural law which specifies the amount of education someone needs to be a productive member of society. People are living longer, and work is becoming more complex and knowledge-based. As the knowledge frontier moves outward, we need to invest in helping all of our citizens get there.
Growthpolicy.org. Where will the jobs of the future come from?
David Deming: I don’t know, and neither do you! If there is one lesson we’ve learned from the study of technological change, it is that the future is very hard to predict. The job-destroying functions of technology—self-driving cars, automated bank tellers in an earlier age—are easy to see. The job-creating functions of technology are subtler, longer-run, and ultimately more important. In 1830, the average work week (not including home work) was 70 hours. It had dropped to 41 by 2002. In the short-run, technologies such as the Jacquard loom automated the manufacture of textiles, putting the “Luddites” out of work and causing social unrest. In the long-run, the Luddites’ descendants have become knowledge workers and buy their clothes at a fraction of the handmade cost. In 1924, 87 percent of U.S. housewives spent more than 4 hours per day on housework. By 1999, that figure had fallen to 14 percent. We have robots in our homes already. They are called dishwashers, ovens, and microwaves. In the long run, inventions like these were the engines that powered the incredible rise in female labor force participation around the world, and ultimately, prosperity and growth. Most of the jobs that exist today—video game designer, occupational therapist, IT consultant, Uber driver—would be incomprehensible to our ancestors. Trying to predict the exact jobs is a fool’s errand. Yet we know they are coming. We should embrace change by creating a set of educational and labor market institutions that can help students and workers pivot more easily to the jobs of the future. This means flexible labor market arrangements, including health care and other benefits that are not tied to one’s employer, and policies that give a measure of economic security to workers who want to get retrained and learn new careers. Broadly speaking, we want a society that provides insurance against the risk of future job loss, and that embraces technological change rather than running or hiding from it.
Growthpolicy.org. What should we do about income inequality?
David Deming: We should first recognize that inequality is not just about economics—it is a political issue as well. Objectively speaking, all of us are better off than our grandparents. I ask my class every year whether they would trade places with someone who was in the top 1 percent of the income distribution 100 years ago. Few of them say yes. Growth makes everyone better off. Inequality is the result of uneven growth. It is a problem of distribution, and thus fundamentally about fairness. Who deserves the spoils of progress? Talent is allocated more efficiently and rewarded much more handsomely than it was even 50 years ago. Wilt Chamberlain and Michael Jordan were both the greatest basketball players of their era, but only one of them sells sneakers in China. On one hand, Michael Jordan “deserves” to earn the return on his talent that comes from globalizing his brand. On the other hand, Wilt probably deserved it too, but just got unlucky on the timing. On yet a third hand, the ability to sell sneakers in China depends on peace and normal trade relations between two of the world’s great powers, and MJ certainly doesn’t deserve sole credit for that! So I think the “what should we do” question is inherently normative. There is no mathematical formula that tells us who deserves what share of the windfall from progress. When nearly all that windfall goes to a small group of people, everyone else understandably gets upset. They think the system is rigged, and they are probably right. And the resulting turmoil threatens the whole endeavor. We need to come together and figure out a better way. And that sounds like politics to me!
Growthpolicy.org. You teach a popular course at Harvard Kennedy School: “SUP-206: The Causes and Consequences of Inequality.” You are also one of the founders of the Collegiate Leaders in Increasing MoBility (CLIMB) research initiative. Broadly speaking, what are some of the policy solutions you recommend for addressing rising inequality in the U.S. in the area of education? Why, in your opinion, is higher education the solution for social mobility today?
David Deming: The course is broad and covers much more than just education. But since education is my area of expertise, I do spend quite a bit of time on it. I have come to believe that college is the most important source of inequality in U.S. education, and the place where we should focus our policy efforts. It took me a while to come to this conclusion. There is a strong counter-narrative which suggests that early life experiences are the most important, and that long-run inequality in skills gets locked in at a very young age, perhaps even before kindergarten. The science behind the importance of early experiences is rock solid. But you have to consider the institutional context. In a child’s early years, connections with a small group of trusted adults are of primary importance. Abuse, neglect, and environmental trauma have terrible and lasting impacts on young children, and we should root it out whenever possible. Beyond these extreme cases, it is hard to legislate your way toward better parenting. I am not saying it is impossible or not worth trying, just that the path to impact in the home is not so clear to me. I do think we should continue the push for universal pre-K, and possibly extend to earlier ages over time. On the other hand, schools are a natural delivery mechanism for social policy intervention. And we have simple levers that we can pull, like providing more funding and distributing resources more equitably. Even though most public schools in the US are funded by local property tax revenue, there is little or no difference in per-pupil spending between rich and poor school districts. This is because of school finance reform policies (nudged along by court orders) that redistribute local property tax revenue across districts in a state to equalize access to school funding. Interestingly, while economic inequality has been increasing for several decades, inequality in academic achievement has not followed the same pattern. In fact, the distribution of age 17 test scores in the US compressed between 1978 and 2012. This is due primarily to gains at the bottom of the distribution and no change at the top. Moreover, several recent quasi-experimental studies of school finance reforms suggest that spending has a causal impact on achievement. Put that all together, and it makes a strong circumstantial case that policy choices we’ve made in K-12 education have acted as a force against rising inequality. Things look really different in higher education. Selective private colleges like Harvard spend about 6 times more per student than less-selective four-year colleges, and the differences are even greater for community colleges. They are starker still for the 53 percent of youth aged 25-29 with no postsecondary education at all. They receive zero postsecondary resources. On top of that, there are dramatic and troubling differences in access to selective colleges by family income. The probability of attending an elite private college is 77 times higher for children in the top 1 percent of the family income distribution, compared to families in the bottom 20 percent. So for all those reasons, I am really focused on higher education as a key driver of economic inequality in the US. I think a mix of institutional improvement and larger-scale policy reform is necessary. The good news is that our CLIMB college partners are willing participants. They recognize the problem and want to do something about it. So I am optimistic that we can work together and have a real impact.
Growthpolicy.org. Your 2017 Quarterly Journal of Economics paper, “The Growing Importance of Social Skills in the Labor Market” is one of the most cited in the research on the economics of education. What are your predictions on the impact of Artificial Intelligence (A.I.) on the labor markets of the next fifty years, particularly in terms of wage and employment growth? Will the current trend in the commercial deployment of A.I. impact skill obsolescence among older STEM workers, a topic you've also studied?
David Deming: I want to restate my response to the question above—predicting the future is hard, and I’m probably going to be wrong! But I’ll try. My QJE paper that you mention is called “The Growing Importance of Social Skills in the Labor Market.” I think you can attribute the paper’s influence to the fact that I am carefully documenting something that everybody already knows. The workplace has become more team-oriented and more reliant on interpersonal interaction. Jobs requiring social interaction have become more numerous and are paying relatively higher wages, and this trend started around the time of the computer and IT revolution of the 1980s. Any workplace task that is predictable or can be fully specified ahead of time is a candidate for automation. As technology has progressed, we have continued to expand the set of tasks that are “routine” and can be done by machines. We all engage in thousands of social interactions every day, and we do it mostly effortlessly (to be fair, some of us are better than others). Conversation is “routine” work for humans. Interestingly, it is devilishly hard to program a machine to have a two-minute unstructured conversation with a human being. So this is a workplace task that is easy for humans and hard for machines, and thus a key source of our comparative advantage over the robot overlords. Other tasks are not like that. We think of chess grand masters and mathematicians as the smartest among us. Yet many of the tasks they perform are trivial programming problems. Calculations and formal logic are hard for humans, but easy for machines. Some people think that Machine Learning techniques can overturn these relationships by mining datasets of human responses, predicting what humans are likely to do in a situation and then mimicking it. For example, we read emotions from people’s faces intuitively. With enough training data in which humans tell the algorithm what emotion is being expressed by a face, machines can “learn” to read faces as well or better than the average human. Yet the machine doesn’t know why you are angry or sad. It’s not reading any of the context around your emotions. And this context is necessary to develop the right response. To get that, you have to really put yourself in the minds of other people. This is arguably the most human skill, and it seems robot-proof to me. At least for now. The more general lesson is that adaptability is a key source of human advantage in the job market. You can develop a machine to do any one thing better than a person. But people are general-purpose—we can pick up new skills and job functions, and switch between tasks easily over the course of a day or even an hour. The newest technologies need humans to tinker with them and figure out a best use. As these technologies become more mature, they get automated or replaced by something better. Your key advantage over the machine is your ability to go with the flow.