Diane Coyle says you can have either robots, or secular stagnation, but not both. In a somewhat confused tweetstorm, Marc Andreessen says secular stagnation is BS. Larry Summers, who is one of the guys behind the secular stagnation hypothesis, responds. But then, confusingly, is reported to agree with Coyle.
While this is a statement one makes at one’s peril, I will say it anyway: Marc Andreessen is wrong, and it ties into his wrongness about Piketty.
Technology can be a very good complement to labor, or a very good substitute for labor.
The more a technology is human-like, the greater the elasticity of substitution between capital and labor.
In the extreme, consider a toy model economy where capital = human-like robots, and you can rent a human-like robot by the hour. Perfect substitution between capital and labor.
The wage rate is going to equalize with the hourly capital cost of the robot. If the cost of robots goes down, the robot rent and the wage rate both go down, all else equal.
Suppose the labor supply is fixed/perfectly inelastic. No departing the labor force when wages go down, no aging population, no population growth.
If you have a technology breakthrough and more/better robots for same price, then overall real labor income goes down.
So, as first order effects, when robots get better/cheaper, two things happen: there is more investment in capital, ie building more robots because they got cheaper. And labor income and consumption go down.
The question then becomes how much of each do you get, and what happens at the macro level?
Maybe after all the second- and nth-order effects work through, you have full employment. GDP increases due to increased supply of capital, but output shifts to investment, ie robots building more robots. In an extreme, you enter a singularity of faster GDP growth, wages going down, more and more robots get built to the point mostly you have robots building more robots, while consumption steadily declines, even as GDP rises.
Maybe you don’t have full employment. If animal spirits are not present and people don’t demand more robots because they don’t see sufficient end-user consumption demand, maybe there is an output gap, i.e. secular stagnation.
This is essentially the Piketty argument. If elasticity of substitution between capital and labor is greater than 1, labor gets relatively worse off over time as capital accumulates and technology improves, in a Solow growth model where capital and labor get paid their marginal product1.
It’s apparently hard to refute Piketty and show conclusively that the elasticity is < 1, either as a theoretical matter, or empirically. The best one can say is, historically it's been close to 1 in the very long run. Labor and capital shares haven't shown a consistent long-term trend either way. Historically, faced with technology that was a close substitute for labor, labor has ultimately done OK in the long run by specializing in what machines couldn't do (elasticity close to 1, very recent history notwithstanding). And historically, threats of technology making labor obsolete and specifically, how quickly artificial intelligence would improve, have proven to be over-hyped. Of course, until such time as we have fully autonomous android robots than can do everything humans can do, technology and capital are partly a substitute to labor, partly highly complementary, a force multiplier for labor. It would seem likely that over time the elasticity of substitution increases, as technology can more closely resemble human labor, perception, decision-making. You start with capital complementing and amplifying human labor, but as technology improves, it becomes more of a potential replacement. It seems impossible to conclusively refute that in the future elasticity is > 1, in the case of radically new technology that is a closer substitute for labor.
In the short run, surely even Andreessen would agree, more disruption means more structural unemployment. It’s the price we pay for productivity growth. Sure, a telegraph operator can retrain as a switchboard operator, and a good SABRE travel agent can retrain for other computer research, but it’s not good news for the travel agent/telegraph operator in the short run.
And in the long run, I think we’ll have to wait and see. Maybe we will find that capital is still a highly imperfect substitute for labor. Or maybe we will find that you can have hoverboards, self-driving cars2, and secular stagnation, and will have to figure out how to create jobs and distribute benefits of technological progress and growth.
P.S. As an aside, I find Summers’s faith in productivity statistics disturbing. In a time of disruption, productivity is hard to measure. A new BMW 3-series comes out. It’s the size of the old 5-series, has better mileage, side airbags, voice-controlled phone and navigation, traction/stability control, rear-facing video cam, heated seats, it lasts longer with less maintenance, I could go on. It costs more than the old 3-series. A ‘hedonic adjustment’ has to be applied. It’s not a conspiracy, someone has to make a judgment call, how much of the price change is inflation, how much is more car for the money. The BLS does the best job they know how, to say how much output went up, and how much price went up.
And that is a relatively easy industrial product to measure constant-dollar output. What if a new startup produces an electric, self-driving car? Deflate that.
And don’t get me started on measuring productivity in services. You are in a strange town, you have a hankering for Thai food, you fire up your phone, check menu and reviews, order Seamless. In the old days, you would have to find a dead-tree phone book, phone, talk to the restaurant, and take your chances. Suppose people switch from radio cars to Uber. You hit a button on your phone, car shows up minutes later, you pay less money for a better experience. Nothing is going to capture that productivity bump. Just fewer dispatchers and restaurant phone order-takers, which is not the real value-add.
Now, a line worker or secretary works for the car manufacturer. Her/his job is automated, robot assembly, no more phones to answer, copies to file, the engineers and lawyers do their own email and stuff gets filed in the cloud. So the manufacturer’s output of cars, however estimated, is divided by fewer people, productivity goes up.
Line worker gets new job managing a Cinnabon which is low output per hour. The economy’s overall productivity goes down, because the composition of output shifted to low-productivity services. The pressure on wages brings back a lot of services that didn’t even exist, when I was growing up upper-middle-class people didn’t have cleaning ladies, now they all do, and you can order all kinds of services on your phone. (Don’t get me started on Roomba output.)
Productivity sort of eats itself. Some are made more productive, others lose their jobs and get pushed into lower-productivity activities, erasing some of the benefit.
Or increased output doesn’t get measured at all. Auto-company paralegal does a project that involves 2 weeks of discovery in a warehouse. Technology turns it into a one-hour search. Maybe the company gets rid of paralegals and produces more cars per hour of labor. Just as likely, people do a lot more discovery. Does it make the cars any better or cheaper? No. Did the productivity evaporate into thin air? I don’t know. Is the economy better off? Depends on the value you place on that research. Maybe more better cases get made, more worse cases get defeated. Or maybe it’s a total waste. But the work and output is there, if not easily quantifiable.
Data only tells you so much.
I suspect there is some fundamental truth to the robots/globalization/inequality/secular stagnation nexus, but it will take decades to sort out and we’ll never really know for sure. You have to build the type of society you want and try to figure it out as you go along. There are always surprises and unintended consequences, and theory or ideology doesn’t reliably tell you what’s going to happen.
1 It’s interesting that Summers is arguing against Andreessen on the secular stagnation hypothesis, and against Piketty on r>g. To me, they seem to be two sides of the same coin. For good discussion of the whole Piketty debate, see: