Jerry Jacobs, a 2018-19 CASBS fellow and University of Pennsylvania sociologist, discusses labor dynamics with regard to automation, job replacement, and historical statistics.
The much discussed paper on automation and the workforce by Frey and Osborne
"Will the Robots Take Care of Grandma?" Jerry Jacobs
"Economic Possibilities for Our Grandchildren" John Maynard Keynes, 1930
"Smarter Than You Think" A series examining advances in AI and impacts on society (including pieces by host John Markoff)
Explore Bureau of Labor Statistics data yourself
Center for Advanced Study in the Behavioral Sciences
@casbsstanford on twitter
John Markoff: Welcome to Human Centered, a series of short conversations with researchers at the Center for Advanced Study in the Behavioral Sciences at Stanford University. The Center was founded in 1954 to encourage interdisciplinary research focused on the most pressing societal issues of our time. Each year, a range of scholars, scientists, and government officials come to spend a year studying contemporary societal problems. My name is John Markoff. I'm a science and technology writer and former reporter for the New York Times. In these conversations, we've set out to find interesting projects at the Center that shed new light on the way we think about society. This month I spoke with Jerry Jacobs, a University of Pennsylvania sociologist who is exploring Bureau of Labor Statistics and has found evidence that is contrary to Silicon Valley's conventional wisdom that everything is moving faster and faster. We discussed the controversial research of Carl Frey and Michael Osborne, the Oxford University researchers who have argued that as many as 47% of all US jobs may soon be at risk of automation. The numbers, Jacobs found, just don't support that trend. Why don't I start by asking you to describe the— I guess the original Frey and Osborne work was 2013, wasn't it? Didn't it— it's been a while now. But what kind of impact did that have on your community when it came out?
Jerry Jacobs: Well, it's had a— Frey and Osborn's piece estimates that 47% of jobs are at risk of automation. This has generated a tremendous amount of interest. It's kind of an irresistible headline for a journalist. Lots of people have adopted their methodology or adapted their methodology. And so you see pieces like this, you know, 47% of jobs in Finland are gonna— maybe it's not 47%, but some number like that— are gonna disappear in Finland. They have a version that says a billion jobs are gonna disappear around the world, or maybe it's 5 billion, I forget what number, but a huge number. And then you see article after article in specific fields. The retail apocalypse, you know, shopping malls are all going to close, and so some huge— so people will come up with an estimate of how many jobs are going to disappear, how quickly. You see headlines that say your next lawyer is going to be a robot, and on and on and on. And so For me, Frey and Osborn have been very influential because they have a very simple and powerful idea, which is if you understand the tasks involved in a job and you have a sense of how, how easy those tasks are to be automated, then you can come up with an estimate of the likelihood that that job is going to, to to disappear. And that just seems— it makes so much sense. I mean, elevator operators, you know, that job disappeared because that task could be automated. And so if we think about all the other tasks that people do, and if we can get a— you know, if we have a scientific estimate of the probability of their being automated, then that's a very powerful idea.
John Markoff: You were not new to this field when their report came out, their initial report. This was already your background is in sociology of work. Had you been looking at the BLS numbers before they—
Jerry Jacobs: I didn't enter the field as a result of their work, but I did think it was interesting because they're not sociologists who spend a lot of time with occupational data. And I thought of, you know, all the sociologists who are immersed in that work. And it was interesting that this set of predictions emerged outside it. And so I think that for me, my goal is in part to debunk overly rapid estimates of how many jobs are going to disappear. But I'm also trying to understand the reception of the argument. So, the polls I've seen suggest that Frey and Osborn have won the debate. As far as the public is concerned, most people believe that we're in the midst of a major wave of technological innovation, that it's going to have a major disruptive effect on the job market, that many jobs are going to disappear, but also people feel that it's not going to be my job. It's going to disappear because people have a sense of, well, my job's a little too complicated, there's too many nuances, there's— it's not— it's— my job's not that easy to automate. So there's a personal understand— on the one hand, there's a personal understanding that this is more complicated than it may seem, but there's a broad acceptance of the idea that this is, uh, this is something that we're all going to have to recognize.
John Markoff: But that's so interesting because doesn't that kind of get at the task replacement versus job replacement sort of duality. I mean, maybe people are more sophisticated in a sense because they realize what really goes on in their particular domain where they actually understand that it's not just one task but they're doing a multiplicity of things.
Jerry Jacobs: Right. I think that's right. The way I've been thinking about this is that task replacement is kind of, you know, automation is like the last step in a sequence series of events. So if your job is Taylorized to begin with, if we break the job down into many different little tasks and then we have one job per task, then you're in a situation where your job really may be vulnerable. But most jobs are not completely Taylorized.
John Markoff: Plus your job would probably be soul-killing if it was.
Jerry Jacobs: Oh, well, but that's the That's the other part of the story. So that when you go to a robot conference, the roboticists will say, "Hey, there's a shortage in this, you know, there's a lot of turnover in this field, and, you know, we're bringing the robots in because it's hard to keep people." Or they say that, you know, "This is a crummy job anyway, and we don't, you know, people don't want to do it, and so we're helping everybody out." by automating it. But in a sense, that's— there's a whole bunch of steps that have made the job kind of a crummy job to begin with. And so you kind of have to back up and say, well, what if we made the job more interesting? What if we incorporated an automated element and mixed it up with a bunch of other tasks and other work that needed to be done in a more complicated way? But that's not how the technologists are often looking at it.
John Markoff: So I wanted to get into your research, but before doing that, I wanted to ask you a general question of where you are with respect to John Maynard Keynes and the very famous article he wrote about the economic possibilities for our grandchildren. I think the simple thing is he argued at that point that jobs will disappear but not the overall amount of labor. And for so much of, you know, the last 30 or 40 years, that was considered to be common wisdom. Now I think that's under some attack.
Jerry Jacobs: Well, let me start with my summary of Keynes. So Keynes' piece was remarkable in that he was writing at the depths of the Great Depression. In fact, the recession had— the Depression hadn't quite bottomed out yet. Things were still getting worse. And Keynes comes along, one of the world's leading economists, and says 100 years from now we're going to be so much richer and capitalism is going to work so well. That we're gonna be in an age of affluence and in an age of leisure. And I thought, I mean, it is a remark, and he basically did a straight-line projection. If you had 2 or 3% real economic growth for a century, how much would that raise the standard of living? And the answer is it would raise it a lot. And he was fundamentally right about that, that a century— you know, we're not quite a century past his prediction, but he was fundamentally right that economic growth would be a sustained feature of the British case, but also the Western case, and he hoped eventually the world as well. People have been very critical of specific predictions that Keynes made. In particular, he predicted that we would be working 15 hours a week. And the basic idea was that he thought that the standard of living would be so high that people would prefer leisure over work. So that hasn't worked out. And it turns out— not only hasn't that worked out, but it turns out that the most educated and the highest paid people work the most. So in my research on working time, I found that poor people often don't have enough work and want to work more. And it's the affluent, especially the affluent dual-earner couples, who feel very pressed for time and often want to work less but can't figure out quite how to do it because their high-pressure jobs demand that they they both put in long work weeks. Again, this is all particular to the United States. It's— there's an element of it in other countries as well. So for me, the issue of working time is complicated because the work-family challenges of the dual-earner professional couple are quite different than the working-class person who actually needs the flexibility on their job to be able to go to a parent-teacher meeting or to take their kid to the doctor if they have asthma or something like that. And the working-class person's situation is different from those people who are kind of more marginally attached to the labor force because their challenge is getting enough, you know, getting past the part-time job to get a full-time job that has benefits and more stable income. So there's work-family challenges across the spectrum, but they're, they're all quite different.
John Markoff: Let me read something you wrote so we can plunge into it. Sure. Over 5 years, about 5% of workers would have to find new lines of work and new careers. This rate of disruption is significant. It represents a real challenge for those impacted as well as their families and their communities. And there's a good reason to appreciate the anxiety, uncertainty that this causes to many American workers, especially given our weak and fraying safety net. But this constant flux is also not entirely news. Indeed, the rate of change since 1970, and especially since 1990, is not faster than that observed in earlier periods in the US economy. Going back to 1870, the rate of change has been faster in most decades than that observed in the 2000 to 2010 period, and thus— thus far since 2010. Now that, to me, seems like it's news to Silicon Valley and probably news that they won't understand.
Jerry Jacobs: So going up through 2017, which is the most recent data that are available from the Bureau of Labor Statistics, we've had a fairly steady rate of occupations growing and occupations shrinking. Most of the time occupations don't completely disappear, but, but there is churn in the economy, but it's not getting faster. The churn in the economy seems— I don't want to say that this is kind of my opposite version of Moore's Law. Moore's Law is that we have an accelerating rate of growth. The rate of occupational change, structural occupational change, seems to be— it bounces around a little bit, but I just don't see any evidence that it's accelerating in recent decades. And to some extent, the earliest periods of industrialization urbanization, kind of moving off the farm, seems to be a bigger change than the rate of change that we're seeing more recently.
John Markoff: So the first lesson you have to take away is that there's not a necessary cause and effect relationship between information technology and the structure of the labor force. It's more complicated.
Jerry Jacobs: I like the way you put that. I think that's a very good point. The information technologies have become embedded in many different fields of work. So when a person comes to your house to do an estimate on a repair of your heating system or to put in a new heating system or cooling system, they're more likely to have a laptop computer with them than they would have had 5 years ago or 20 years ago. So before, they would have taken notes and they would have put those notes in a in a manila folder, and now those notes are recorded in an electronic form that can be shared with their coworkers and so on. But somebody still has to come out to your house, and somebody has to understand the nature of the heating and cooling issues in your house, and they have to spend some time talking with you. So that job is impacted by information and communication technology. But it's not replaced by the fact that the representative of the company has a laptop computer to work with.
John Markoff: Are there other candidates that you can see that would be social or economic or political forces that are at play that aren't technological that are playing on the structure of the workforce?
Jerry Jacobs: My estimates of occupational change incorporate everything that's affecting the occupational system. So technology is the topic that we've been discussing, but the fact is changing fashion can be responsible for the fact that we don't buy as many hats today. So culture is different. Some of it may be international, you know, so the manufacturing jobs don't necessarily disappear, but they're moved overseas. So I'm looking at all the occupational changes, whatever their cause. And even taking in all of those things, the rate of change doesn't seem to be faster than it used to be. If anything, it seems to be somewhat slower than it has been in the past.
John Markoff: The microprocessor was introduced in 1970. So, you know, machine learning comes along in 2010, but the microprocessor, which is like a cost reduction of computing has been out there playing in the economy for what, 40, 50 years now? 50 years, I think. And that's extraordinary that given the belief system in Silicon Valley, because the microprocessor should touch on everything. Now there is this theory about lags, and does that— could that possibly explain why technology is slower, that it takes a while for this stuff to to get going? It's a long time to lag, isn't it? Over a half century.
Jerry Jacobs: Well, I think, you know, again, I think Moore's Law is amazing. You know, the fact that computing power gets smaller and cheaper and doubles every couple of years is a remarkable thing. And the fact that it's sustained for so long is really amazing. And I think it has resulted in a lot of things that we— it's embedded in a lot of things that have changed our world. And so the Moore's Law obviously leads to the iPhone. Yeah. And the iPhone is responsible for changing so many things in people's lives.
John Markoff: But it would be— it's probably taken away some jobs, but then the iPhone, you probably, if you looked at the net job creation and destruction around the iPhone, I bet that it's created more work in the world than it's destroyed.
Jerry Jacobs: Well, I've just started to read a book about India called India Connected. And it makes a couple of really interesting points. One is that a billion people are going to have phones. It's just a staggering fact. Mostly not iPhones, mostly simpler. Second point was that a lot of those people didn't have a landline previously. And so the author makes the point that the introduction of mobile technology is gonna have a bigger impact in India because people are gonna go from no communication to this incredibly sophisticated mobile communication system. Um, but one of the points also explored is the idea that communications— the introduction of rapid communication automation is going to have a lot of impact in India in terms of creating new economic opportunities, that it's going to reduce inefficiencies and create tremendous opportunities. But back to your question, why does it take so long to impact jobs? I think there are, there are a number of reasons, but I think in each particular case you have to do a an analysis of the particular case. One reason is, I think, related to the Gartner hype cycle. So we start talking about new technologies because they're exciting and they're news, and we talk about them before they're ready for prime time. But beyond the simple idea that technology, you know, often requires a lot of refinement alignment before it works fully. It has to be incorporated in an economic framework, and it has to be incorporated in company, and it has to be connected to other ways of doing business. And I think that effort, that incorporation, often involves kind of rethinking the business model and rethinking how work is done and how it fits mixed in with other kinds of work.
John Markoff: What about the capital costs? You've written about labor-saving devices, and I was thinking about both cars and trucks. Let's say we could make trucks safely drive on freeways and replace drivers. The reality of replacing the fleet of trucks with autonomous vehicles at scale, it wouldn't happen in any, in any particularly short term, right? The capital investment would probably be immense to do that. And nobody, I think, has discussed the cost of replacing the nation's trucking fleet with autonomous vehicles.
Jerry Jacobs: There are some interesting reports on how long it would take to introduce self-driving trucks and self-driving cars and what the— so it's a matter of decades.
John Markoff: If it was a matter of decades, wouldn't that tend to normalize or minimize the rate of change in terms of the labor? Component?
Jerry Jacobs: Exactly. So, so there are a couple of things. So one is that trucking, for example, is a high turnover occupation. It turns out that long-haul trucking is a challenging lifestyle, and most people find it very hard to do that kind of work for a long time. So again, this doesn't provide jobs for everybody, but it suggests that is that over time a lot of the replacement of truckers would happen through attrition because people— there's turnover in this area anyway. It has a different impact than if, you know, 10,000 people walk into work one day and they all get pink slips indicating that they've been fired. The other thing I think is interesting about self-driving trucks is that they're not as geographically concentrated as the factory jobs that we've lost in the past. Okay, so, so there are trucking jobs in every corner of the country. It turns out it's one of the leading occupations in many different states. Yeah, so I certainly don't want to minimize the, the consequences on people if they work in lose this job. It's certainly valuable and important to them. But I think it's different if there's a few thousand truckers in every city as opposed to 10 or 20 or 50 or 100,000 people in the steel industry or in the automobile industry who are all concentrated in the same place and who all get a layoff notice on the same day that the factory's closed down. So I think in some ways the The impact of job loss in the service economy may be quite different than the impact of job loss in the manufacturing economy and certainly in the farming economy.
John Markoff: In which direction would you say different?
Jerry Jacobs: It's because it's not as geographically concentrated. You know, when the dairy industry is in distress, you know, it really hits communities in Wisconsin it really hits them hard. The loss of trucking jobs, I think, would be a little bit like the loss of telephone operator jobs. It's spread out across the country. It turns out that there are other jobs that require many of the same skills that telephone operators had, and so women were able to move from telephone operating work to other kinds of work without it creating a a national crisis.
John Markoff: And is that model, is that orthogonal or is it related in any way? You know, David Autor at MIT has done this work where he sort of posits a U-shaped model where the economy has continued to grow at the bottom and the top, but the middle has declined. And is that related to this kind of a horizontal impact or is it different?
Jerry Jacobs: It's different. As I see it, that's not a theory. I mean, that's a description That's the description of what he observed. And then again, there are not too many better observers of things, so this is not to take anything away. But I don't think we're at a point where we can say that the tech economy generates high-tech, you know, PhD jobs, you know, and janitor jobs. I just, I think, again, that may be true, But I don't see any reason why that's true in principle.
John Markoff: Yeah.
Jerry Jacobs: As I study the case of elder care, I see technology having more of an impact on where the work is being done as opposed to the nature of the work.
John Markoff: Yeah.
Jerry Jacobs: So I see technology as facilitating aging in place. And so that would create more jobs for visiting nurses. Which are kind of high-skilled, but also nursing aides, home health aides, which are low-skilled. But I don't see it necessarily carving out the middle. Maybe it will in some areas, but I don't see it as a matter of principle that technology is going to have that impact across the board. The other thing that happens with our data society is that we we need a lot of data. In the Silicon Valley view of the world, data is all passively gathered from activities that people are already doing, so that the phone tracks where you are and it tracks what you've purchased and so on. But a tremendous amount of data has to be entered. We ask people questions and somebody has to enter this data. So it turns out that medical record systems that have been introduced by Obamacare create a tremendous need for data entry. Currently, a lot of this work is being done by doctors who complain bitterly about all the time they spend entering data into these medical record systems, and they, they want to be spending time with looking at their patients and talking with their patients, and they find themselves staring into screens and clicking repeatedly before a prescription can be ordered. So that's generated a new job called medical scribe, and so there are people who are following doctors around doing all their data entry. And again, is that job— is that going to be a wonderful middle-class job that's going to to have benefits and replace manufacturing jobs, I don't know, maybe not. But the idea that clerical work goes away because of information systems, sometimes it does and sometimes it actually generates tremendous demand for new data that actually has to be entered by somebody.
John Markoff: Could you say it's true that we're better at predicting better at predicting what jobs will be killed by technology than we are at predicting what kind of jobs might be created by technology?
Jerry Jacobs: You're hitting on it, I think, a really important point. It's easier for people to see the jobs that disappear. It's also easier for people to understand how technology is going to replace a certain task. And it's, it's often very hard to see the jobs that may be created. And there's, I guess I have one little caveat about that, and then, and that is that it's not all about the invention of new jobs. Lots of jobs grow, jobs that we're already familiar with. So there are lots more waiters and waitresses today than there were 30 years ago. So it's not like It's all gonna be jobs that we've never heard of before. Then there are lots of things that are hard to really track down effectively with— even with the most detailed occupational data. So, as best I remember, coding is not an occupational category. Computer coding is not. Programming is a category. Programming, and the number of computer programmers is smaller than there used to be. That's fascinating. But on the other hand, coding is often touted as a skill that everybody's— that we want to try to teach as many people as possible because it's going to be one of the fastest growing areas of work.
John Markoff: Is programming being pushed down in terms of work status? Is a coder of less status than a programmer even though they do what I don't really know.
Jerry Jacobs: And what I'm trying to say is computer jobs have been growing, but a lot of times the popular label on a job, it takes a while for that label to be incorporated in the categories that are used by the Census Bureau and the Bureau of Labor Statistics. And then when it is incorporated, then you have the problem of is today's coder really different than yesterday's programmer? And, you know, in terms of the work that's being done, in terms of the status, in terms of where the work is being done, and, you know, all that kind of stuff. What I'm trying to say is that following occupational change is kind of a slippery business. We think that we've had this issue since settled, you know, because we've been trying to track this for over 100 years, and it turns out every 10 years they come up with a slightly different system, and then there's an issue of trying to match things over time. And it's— none of it's perfect. Again, I'm not trying to criticize the skilled professionals who do this work, but it's just inherently messy. So for whatever skepticism I might have about predictions. I think their predictions are probably closer to the mark than the Frey and Osborn idea that categories X, Y, and Z are just going to disappear and it's just a matter of time. The BLS— again, I've been trying to get a better handle on what their methodology is. Some of it reflects just sort of growth of different industries. What I haven't been able to get a good handle on is how the BLS incorporates technological change in their predictions.
John Markoff: And they're probably trying to get a handle on it too, you know. Here we are, you've got pretty clear evidence that things are actually not on an exponential curve in terms of, at least in terms of new jobs, or in terms of jobs killing things on the other hand. But I wonder if, you know, there's also this perception in Silicon Valley is that there's this technological breakthrough that's right around the corner, it's just not here. And the best example I have is a couple of professors I know at UC Berkeley have just left to set up companies to do grasp. And my sense is that if you could accurately do grasping of individual products, for example, in the warehouse or wherever, that that might will lead to a new wave in manufacturing and other things. It leads me to that famous Rocky and Bullwinkle sort of passage, you know, "But Bullwinkle, that trick never works. This time for sure." You know, that's sort of a Silicon Valley truism in a way.
Jerry Jacobs: So as you know, when people talk about all the amazing things that computers can do, the objection is often raised, but they also can't do very simple things. They also can't open a door and recognize things in a way that a 3-year-old can. I was at a robot conference in the fall where the pick-and-place arms were picking up cupcakes. Okay, so the idea was that they were getting much better at being able to pick up things. So of course my reaction was to say, "How long do you think it's going to take before you can pick up strawberries?" strawberries. And I got a very thoughtful response. The thoughtful response was, first of all, not anytime soon. Second reaction was, the problem with picking up strawberries is that you don't see them very well. The strawberries are hidden behind leaves. So you need to know where to look for them. In addition, you have to not only gently pick up the strawberry, but you have you have to snip the stem. So you, you have to have subtle hand but also a scissor of some sort. Putting that all together at current price scale too, yeah, is a challenge. I think the biggest example of the change that's just around the corner that's going to change everything is AI, uh, and, and it might But again, it turns out, you know, it often takes a while to figure out exactly what tasks can be automated and how they fit in with everything else. So, in 2008, among the many fields that people were worried about were lawyers. Okay? So there was, for the first time in anybody's memory, law school graduates graduates had difficulty getting jobs, and this was partly due to the economy and partly due to the idea that legal work was just getting way too expensive. But also there was this idea that so much legal work could now be searched through documents, and you could search through precedents, and you could search for the legal issues that were relevant to the this particular case, so who knows how much of the work of junior assistant lawyers would not be necessary. There are more lawyers today than there were in 2008, for better or for worse. And so, and that's not to say that, you know, those predictions were wrong, it's just that it may take a while to figure out just what part of legal work can be done. And then the other thing is, you know, it can be done by machines. The other thing is that, so it turns out 50% of people don't have wills. So technology can reduce the cost of various tasks, but they have the potential to expand the market. So if we could get wills to all those people who don't have them now, that would be an expansion of the legal market. Then it turns out that once you start talking with people about their wills, there are lots of other— upsell— possibly things that you might, they might be in the market for in terms of legal advice or other kinds of advice. So again, that the idea that we have a fixed set of tasks, that we have a fixed market its size, and that the only thing that technology does is subtract tasks that, uh, away, and each task that's subtracted reduces the amount of work, and it's just a matter of time before there's very little left for us to do. That idea is kind of core to the Frey and Osborn logic. And it just— there's just case after case in which that's just not how these things have been playing out.
John Markoff: So describe to me briefly either your current or your sort of next project, which is, I think, focused in extending that, or extending the examination of that question. Is that right?
Jerry Jacobs: So one of the things I'm hoping to do fairly soon, and I'm working on a grant proposal that is looking at professional education. So one of the areas that, that technology is— it's a little bit more on the horizon— is the idea that many of the most established professions can have artificial intelligence systems that figure out better than what the professionals can do how to diagnose an illness or how to recommend the a strategy to a client in terms of legal work. So I'm hoping to study law schools, journalism schools, and schools of— graduate schools of education to see how the next generation of professionals in this area is— to what extent are they being taught what tomorrow's technology technologies are going to be. To what extent are the leaders in the field recommending that everybody, you know, is technology going to be required for all teachers? Or is tech education going to be an educational specialty? Or do people feel that education is really a humanistic activity where technologies can have a relatively minor role? I've begun to have some very interesting conversations with professors and deans of schools of education that I'm hoping to learn more about how that's all developing. And the idea is to take a few professions, and the ones I have in mind at the moment are journalism, education, and law, because I think there are going to be some interesting similarities and some interesting differences across those fields. So for example, there are a number of law schools where you can get a specialization in legal technology. So what's going on with that? And how did that emerge? And again, this is very recent. And does every law school need to have one because Stanford has it and Harvard has it and University of Chicago has it? So, yeah, that's where I'm headed.
John Markoff: Thank you for taking the time. It's always great fun to to find that the world is a little more complicated than it seems from looking at it from Silicon Valley's point of view. So this has been, this has been great.
Jerry Jacobs: Thanks.
John Markoff: Thanks for chatting with us today. And for our listeners, take a look at the show notes for this episode for links to some of the topics and organizations we've discussed.