The Tech Revolution: Is This Time Different? - August 28, 2018

Transcript

Charles Davidson: Welcome to the Federal Reserve Bank of Atlanta's ECONversation—glad you could join us today. My name is Charles Davidson—I'm with the Atlanta Fed's public affairs department—and I'm joined by Dave Altig, the Atlanta Fed's director of research. Dave, thanks for being here today.

Dave Altig: Good to be with you.

Davidson: Yes. And just a reminder—we'll be answering your questions during the webcast today. So if you have questions, as they occur to you please send them to this email address, events@atl.frb.org, and we'll get to those questions toward the end of our chat today. So Dave, to start with: as I understand it, today you're going to make the case that technology in fact is not upending the economy substantially more quickly than it has in the past, basically. That sounds like it probably flies in the face of conventional wisdom, to some degree, so can you start with a snapshot of what you mean—sort of an opening argument, if you will?

Altig: Yes. Let me be clear that I'm not saying it's not upending the economy—it certainly is—but it's not happening faster than in the past. So oftentimes I run across these sorts of statistics that go something like this: it took 35 years from the time the telephone became commercially available to when a quarter of the households in the United States had a telephone. Fast forward—if you look at smartphones, the number that might get bandied around is something like: it took five years from commercial availability to 25 percent of the population using smartphones. I've got a couple of problems with that.

The first problem is: that's a very hard metric to construct. When does a product become commercially available? How do you keep track of the penetration, with respect to how many members of the public have it? So the smartphone, for example: you might think of the smartphone starting with the iPhone in 2007. You might think of the smartphone, if you've been around for a while, starting with the Blackberry about five years earlier than that.

But actually the first smartphone is really the IBM Simon, the first device that combined a personal PDA with a telephone—that was 1994. It probably took until about 2010, 2011 for 25 percent of households, or individuals, to have a smartphone, so that's 16 years. So, A, that's a long time—or certainly a lot longer than the five years that people may be thinking—when you really begin to see the acceleration.

But my more fundamental problem with that is that speed of adoption by U.S. citizens, U.S. households, U.S. consumers, is not unique in history. So let's take the vacuum cleaner. The vacuum cleaner—a product of portable power and electricity—was every bit as innovative a product in the 1920s as smartphones are to us today. So in the 1920s, from the advent of when vacuum cleaners became available to when 30 percent of households had a vacuum cleaner in their house, that took 10 years. So the smartphone doesn't look particularly fast in that context. I think the thing we want to think about very carefully is this distinction between consumer goods that embed technologies and how fast they get adopted within the population, from what I'll call "general purpose technologies," which are the much bigger picture sort of technological advances.

Davidson: Right. So in terms of these broad-platform, general purpose technologies, what are some examples from the past of technologies of that sort, Dave? And how long did it typically take those to then spawn products that really produced some pretty fundamental change?

Altig: When you think of a general purpose technology, you want to think about those big, audacious, world-changing inventions that become pervasive—and pervasive is really the key element of them. You start to see them everywhere and in every aspect of life and all types of products. So you want to think about industrial revolutions, you want to think about the steam engine, you want to think about electricity, you want to think about information technology. In the current wave of things, you probably want to think about artificial intelligence and machine learning. If we look back to those other industrial revolutions, it's a proposition where you're really talking about decades and decades and decades for the full effect of these technologies to take hold, if we've even felt them yet.

Davidson: So now, as I understand it, there are some fairly interesting historical parallels in there. Some of the patterns we've observed recently—slowing productivity growth, rising wealth and income inequality—it sounds like what's happened in the past in the wake of earlier revolutions, industrial revolutions: pretty much the same kind of stuff?

Altig: We saw exactly the same sort of stuff. As far as we can tell—the data is a little more sparse way back when, but there's some important work by—or a prominent name is—an economist named Paul David, who looked at, for example, the first industrial revolution in Great Britain, and then that coming on to U.S. shores in the pre–Civil War period. What you see are some interesting things, and you mentioned some of them. But the most fundamental one is really productivity actually seemed to get worse before it got better. That is, we saw productivity growth slow before it picked up. And along with that, rising inequality, more concentration of wealth, an elevated premium to skilled labor versus relatively unskilled labor—I think that all sounds pretty familiar right now.

Davidson: It does, yes—no question. So do we know why those patterns seem to play out like they do, Dave?

Altig: Yes, there's lots of stories as to why they do, but in general, you might want to think of these as, they're very disruptive technologies, which means you're disrupting old models of business. You're disrupting the value of capital and old machines that don't embed the new technologies, and you're disrupting the return to labor—you're really changing the skill set that's required of the labor force, so figuring out new business models—retooling both our physical capital stock and our human capital stock—are things that take time. It just takes an enormous amount of learning by doing and trial and error. What eventually happens is that these technologies get embedded in goods and services, as the case may be, that in turn spawn new invention. So this is a process that goes on for a very, very long time. It's not painless, but actually it's a very good story in the longer run.

Davidson: Yes. So I wanted to ask you in particular about slowing productivity growth. I think in your last answer you basically answered this question, but I want you to maybe expand on it just a little, so: Why is it that productivity growth tends to slow before it then accelerates, in the wake of these sorts of—

Altig: Well, that's a good question, and there are lots of hypotheses about that or guesses about why that might be, and I'm not sure any one is really the full story. One is that there are just a lot of false starts. People don't understand how to use the technology. They don't understand how to harness it. I think an example that's often given is, when we moved from steam engine to electric power, people tend to think of just replacing the steam engine with a big ol' electric motor—which of course is not where the real power of that technology was. The real power was you can miniaturize the power source, and you could create something like a vacuum cleaner, which you couldn't with a steam engine.

That takes a lot of time for people to figure out, and even after people have figured it out there's a lot of old infrastructure in place that depends upon the old technology. And it's not immediately that it will seem profitable to shift to the new technologies from the old technologies. Those old technologies will be around. They'll become increasingly obsolete, so a period of time where you're learning about the technology, where you have this mixture of old and new, is—maybe not surprisingly—associated with some productivity effect, which doesn't look all great in the shorter run.

Davidson: Yes. Well, Dave, I want to, before we leave the vacuum cleaner behind [laughter]—of course, we take that for granted. A vacuum cleaner is this utilitarian thing everybody has, it's just part of how you keep your house clean. But did that have some fairly profound effects? I'm guessing that—maybe not immediately, obviously—but at some point—inventions like that helped lead to the big surge in women's labor force participation? Is that a stretch?

Altig: Yes, that's exactly right. And that's what I sort of meant when I said "new products get invented that embed these technologies, which themselves spawn all sorts of enormous changes." And it's in these mundane places we might not think of looking that we find them, and the vacuum cleaner is a beautiful example. Mechanized refrigeration is in that class—those things come online at about the same time as we figure out portable power that harnesses electricity to the household, which frees women up, basically, at the time to spend less time in the home doing work, and it gives them an opportunity to enter the labor force.

Interestingly, in 1912 or so, Thomas Edison foresees this, and he has this beautiful quote, and I won't be able to give it to you right off the top of my head in its specificity, but it's basically: "The future of women, and what they have the ability to do, is going to be fundamentally changed by electricity. It's going to free them from the work that they have to do in the house to other more productive means."

I would say the analogy that might sing to people today is the App Store. If you think about the App Store—and this is what I think is really phenomenal about smartphones—it's not that the smartphone is such an interesting device in itself. As I said, it's been around since about 1994, and it's really just a mash-up of handheld computing, which is just a continuation of computing in general—

Davidson: ...and telecoms.

Altig: Yes, and wireless communication—both of those things go back to the 1940s. But what the smartphone delivered—and this is where the iPhone really becomes kind of a more unique device, is with the App Store. We couldn't think of Uber without the App Store, and the really startling thing about what Uber delivers is it delivered the capacity to think about capital—think about a car that you have in your garage—as a piece of capital that releases productive capacity, much like the vacuum cleaner did for women's labor supply back in the 1930s.

Davidson: Right. So the typical automobile owner, your car just sits most of the time, right? It doesn't do—

Altig: It's a consumption good, and is not an investment good. And maybe that's changing.

Davidson: Right, yes—very interesting. I love the sort of concrete, real-world example here. So, Dave, clearly when we talk about this stuff writ large, there's still a lot we don't know, right? About why this stuff happens, where it happens, what do we gain from figuring it out? Where are we better off if we more deeply understand this kind of...these—

Altig: Well, one thing it does is it might help us sleep better at night. I mean, we are in a situation in the U.S. economy—in fact, in the global economy, particularly among advanced economies—where kind of a dominant feature of the environment is this so-called productivity puzzle. Productivity growth has slowed way down off of its pace, even of the decade prior to the crisis, and certainly a lot compared to the decades before. To know that we've seen this before—and that it is symptomatic of gigantic leaps forward in technology that ultimately improve everyone's wellbeing—is I think kind of a nice thing to keep in our pocket as a reminder that patience and perseverance actually pay off.

Davidson: Yes, I like that. Some optimism from the dismal science, right? [laughter]

Altig: Don't ask for that again! [laughs]

Davidson: Won't do it. Okay, so you mentioned artificial intelligence [AI] as an example of a general purpose technology, and that's one we hear a lot about these days. So is it in fact—you would term that a "general-purpose technology?"

Altig: Very definitely.

Davidson: Okay, so in historical terms—as far as how long we've seen it take these sorts of technologies to mature to the point where they're producing products—how far along is AI?

Altig: Very early on. There are obviously great strides that have been taken in a very short period of time. For example, over the span of probably less than eight years, certain machine learning algorithms can recognize images with error rates that are as low as human beings can accomplish—and in fact, error rates that are below what human beings can accomplish. So there is very rapid advance, but there is a long, long way to go. And I would just note to anyone who has tried hard to get Siri or Alexa to understand what it is they really wanted, they will completely understand that we're at the very beginning stages of all of this.

Davidson: So I'm going to ask you to kind of "blue sky" here. With AI, what do you think that portends for human workers and for labor markets more broadly—let's say, in the reasonably immediate future?

Altig: Well, it's going to mean I think precisely what was meant by the shift from an agrarian economy, or an economy dominated by agriculture, to an economy dominated by manufacturing and electric power and motorization. It's going to mean the same things those meant, which is we are going to have to rethink what our labor force looks like, what our business practices look like, and so on and so forth.

There are definitely—I mean very definitely—things that we think are solely in the province of human capacity these days that artificial intelligence is going to take over, and we will have to rethink what our contribution will be and what skill sets are necessary to make these contributions as we go forward.

Davidson: Right. So what do you think the biggest impacts we've seen already from AI are?

Altig: Well, AI—I'm not sure that we have seen the very large impacts yet, particularly with respect to the replacement of labor and those sorts of things. We're beginning to see it, I think more importantly, has been sort of robotic mechanization. That would have some elements of AI built into it. But that very definitely has had an impact, and that's an offshoot in many ways of the continuing evolution of computing power and information technology technologies.

There's a well-known fact among people who follow labor markets that middle-skill jobs have diminished, both in terms of employment growth and in terms of wage growth. And we often point to the fact that jobs that have been routine—meaning the same task repeated over and over again, in kind of a mechanical way—have been replaced, and those were kind of predominantly middle-skill jobs. If you look at what we would classify as lower-skill jobs and higher-skill jobs, those have grown in wages and employment, and those tend to be jobs that are not so routine. So on the lower end, think about housekeepers, and on the upper end maybe think about lawyers or something like that—they have been the source of growth in labor markets because those are the things that have not yet been easily replaced by robots or artificial intelligence. Although, particularly if you think about lawyers and some doctor services—and economists and engineers—AI is probably a pretty decent threat [laughs].

Davidson: Well, I was going to ask you—what do you tell someone who says, "Well, jeez, Dave: at some point are we going to be superfluous—the humans? Will the robots let us hang around?" I mean, obviously I'm exaggerating, but you know—is the future necessarily grim?

Altig: I think a lot of people think you're not exaggerating, actually. So there's another way to cut the labor market data of the past 25 years or so. There is this important distinction between routine and nonroutine jobs—jobs that require lots of thinking, and jobs that don't. But another interesting way to think about the dimensions of jobs is those that, for example, require high math skills versus not high math skills, those that require high social skills versus that those that don't require high social skills.

So it's also a fact that, again—if you think about jobs that are growing in numbers and jobs that are associated with wage growth—the common feature is high social skills, independent of the level of math that's required in the jobs. So it certainly is true that it's always a good thing to have high technical skills and high math skills, but you want to be a STEM student with good communication skills. Because that's the thing: those social skills and human interaction of various sorts are still the things that AI is nowhere close to being able to accomplish. I can give you an example, if you'd like.

Davidson: Yes, I wanted to ask you—so, have we seen this start to lead to new types of occupations?

Altig: Well, it's not so much that. I mean, I think in some sense we may have already seen it, with this fact that it's the social-skill dimension that really seems to be being rewarded in the marketplace.

Davidson: I don't know a damn thing about math, and they pay me.

Altig: [laughs] And you've got great social skills, so...here's an example of something artificial intelligence doesn't do very well. Consider the following sentence: "The councilmen denied the demonstrators a permit because they feared violence." Who is "they"? Is it the councilmen, or is it the demonstrators? Well, you and I—without even thinking about it—understand it's the councilmen. Machines have a very difficult time with that. So there are lots of things that artificial intelligence either can figure out already or will eventually be able to, but there still seems to be some fundamental roadblocks. And that, if I'm going to kind of casually characterize it, is robots aren't good at being human beings.

Davidson: Right—[laughs] makes sense. Now, have we seen new occupations, maybe—have we seen—and maybe it's a premature sort of question here. But have we seen the wage scales shift much? If there's indeed a premium on social skills, does that mean good social skills are already rewarded more so—

Altig: Yes. So if you just look back over the past 25 years' worth of experience that I was referring to, jobs that have the characteristics of low social skills—independent of whether they require high math skills or low math skills—that wages have actually shrunk. So growth has been negative. Those categories with high social skills have actually experienced wage growth, so there's a different kind of skill premium in the background here and that's communication skills and all the sorts of things actually employers say they are increasingly looking for when they look for a worker.

Davidson: Right, okay. Well, Dave, let's try to put this into a little bit of a policy perspective, if we can do that. So are there implications for monetary policy making from all this? We've been talking about a lot of stuff here, but...

Altig: Well, part of the Fed's mandate is to support maximum sustainable employment growth, so it is absolutely imperative that we understand the structural developments that are underlying phenomena in labor markets. It is true that monetary policy is a pretty blunt instrument, and it's not so clear that monetary policy can be of much value in, for example, retraining the workforce to deal with the changing environment. Policy is sort of aimed at workforce development—education, vocational training, entrepreneurship—all those sorts of things are probably the key elements of dealing with this disrupted future that we are dealing with.

But monetary policy authorities absolutely need to understand these dynamics, or life will get a lot more difficult in meeting our mandate—not just on the employment front, but actually maybe even on the inflation front. If we guess wrong, what's happening in the economy—I mean, that's where mistakes can happen.

Davidson: Sure. Well, we have an audience question here I'll throw in—and this is related: When it comes to advances in AI, what are the implications for the Fed's influence over the money supply and availability of credit?

Altig: If there is an implication, I haven't seen it yet. I think that there's a well-known argument that one of the things that happened in the 1970s, when we experienced the Great Inflation, that we were in in the midst of a big productivity development—in particular, a big productivity slowdown—that we didn't see. And as a consequence the Federal Reserve interpreted the slowdown in growth as something that had to do with labor markets not functioning appropriately, and were overly accommodative in response to that. That's the channel through which I think it can affect our influence, not because we're becoming less influential or we're losing control of the tools, but because there are things that we need to react to in the environment that we haven't quite caught up to. And that's why my answer to your previous question was, we really do need to understand these structural developments, and we need to look to historical precedents, and we need to understand the big picture if we're going to calibrate our policies correctly.

I will say that, obviously, and another example that's pretty apparent, is fintech, and all the AI and information technology and innovation that that embeds, is one of those things that we're going to have to think carefully about how it's impacting the supply of credit, for example, and figure out then what are the appropriate reactions on our end.

Davidson: Right. So talking about the '70s: just to put this in perspective, we saw inflation well over 10 percent, right? Now it's weird to think if we get it to 2 [percent].

Altig: We got there [laughter].

Davidson: So, let's see—another audience question here. What do you see on the horizon as the most imminent—meaning, immediately coming—tech disruptor, or what is the thing we don't know we don't know?

Altig: Yes, well, if I knew the things [laughter] we don't know, then I would be in a much better position to answer that question. I mean, I think it is true that we literally don't know what advances in biomedicine, for example, and stem cell research—we don't know what impact that's all going to have, at a time where we're already trying to figure out the demographic trends in our economy and how we adjust to an aging population—which, you can imagine, is going to age even more as we tap these technologies that embed artificial intelligence, machine learning, with very basic biomedicine. And I've always been convinced that we're going to wake up one morning and realize we're going to be around for a lot longer than what we thought, and that'll be an interesting policy challenge. So that's one of the things I kind of think about a lot.

Davidson: Yes, the life expectancy. I mean, we've seen some moves in the other direction, in some segments of the population, but in general we're going to be living longer it looks like, right?

Altig: I think so.

Davidson: Yes. Well Dave, so these waves of innovation—if indeed that's the right way to think of this—is it possible to clearly distinguish, say, "Okay, here's when one started, and here's where it stopped?"

Altig: No. In fact, it's not possible at all, and I'll just return to my smartphone example, which is, you could think of the smartphone as being this mash-up of wireless communications and computing power that was started in the 1940s, as discernable things. So it can last a really long time.

Davidson: Yes, all right. Well Dave, just real quickly—we've got about a minute and a half, but—what would you like to leave folks with that's just maybe one or two really big points from our discussion today?

Altig: I think the big point is this—to refer to the title of this talk—this time is not different. We've seen this before. I think we clearly can look to historical precedents to feel our way through what is in fact a very disruptive time.

Davidson: Yes, all right. Well, Dave, thanks so much for your time today. We do have about a minute to go, so we might have time for one more really quick question. Let's see, another audience question. This is a tricky one: Who are today's Luddites, and does the Fed pay attention to...well, this says "to these disrupters," and if so, how and why? That's a little confusing, but...

Altig: So today, I don't know if I want to call out anyone as Luddites today [laughter]. Look, there will always be people who, quite justifiably, are threatened because they are the ones getting disrupted. And I don't think those voices are voices that ought to be ignored. The big challenge is, we need to figure out how we make this work for everyone at a time where that takes work.

Davidson: Yes, all right. Well, Dave—thanks so much for your time. Fascinating discussion today, I really appreciate it. And thank you for joining us at another ECONversation. Please go to our website for more information: frbatlanta.org.