By Greg Ip

May 16, 2017
My column last week took on the popular prediction that robots would throw millions of workers onto the dole or consign them to low-wage hell. This prediction, I wrote, is baffling and misguided, because the evidence points in the opposite direction. In fact, lousy productivity growth shows that robots aren’t destroying enough jobs.

The piece precipitated a friendly argument with Ryan Avent, with whom I previously worked at the Economist and who now writes the Free Exchange column there. Ryan pointed me to a piece he wrote for Medium on the productivity paradox, drawing on his book, “The Wealth of Humans.” He argues what we see isn’t evidence of disappointing technological progress, but instead rapid change alongside thin safety nets for workers. Our debate follows.

•••

Ryan writes: Low labour costs discourage investments in labour-saving technology, potentially reducing productivity growth….In real terms, wage growth has been close to imperceptible for most of the workforce since 2000…the scarce factor isn’t capital equipment. What is expensive is the intangible capital thats needed to overhaul production in ways that use cheap computing power to eliminate lots of jobs….While labour is cheap, firms face little pressure to make those massive investments in intangible capital in order to automate key processes.

The digital revolution has created…an abundance of labor….Technology is increasingly capable of substituting for human workers across a broad range of tasks….People ask: if robots are stealing all the jobs then why is employment at record highs?….Some of the displaced workers might give up on work and drop out of the labour force, most…would seek out other work, glutting HR offices and employment centres and placing downward pressure on the wage companies need to offer to fill a job….Given the structure of our social safety net, automation tends to increase poverty and inequality rather than unemployment.

•••

Greg responds: There’s probably something to the argument that cheap labor was holding back productivity growth by discouraging firms from investing in labor-saving equipment. But it just can’t explain enough. Firms can meet demand through some combination of capital and labor, and while cheap labor might bias them away from capital, the real problem was that demand was so low, and thus didn’t require much capital or labor.

The question then becomes, Why is demand so low? We can blame demand and the fact interest rates can’t fall below zero. Or we could blame supply: There aren’t compelling things to invest in, which results in poor productivity growth and low wages. It gets very circular. Now that we are more or less at full employment, we’ll know whether the problem is lack of demand or lack of supply. I am pretty confident both capital investment and productivity are going to improve over the next few years, because they’ve been unusually and unsustainably weak for so long. That’s just regression to the mean, not proof a digital revolution is under way.

You argue that technology creates an abundance of labor, which then pushes down its cost, discouraging capital investment, which hurts productivity. If so, we should see productivity climbing somewhere, in the sectors that are shedding labor. But we don’t. There is little evidence that the technology we are raving about is helping any industry, except maybe retail. Productivity growth is terrible everywhere. Nor do we see strong business investment in high-tech equipment, which presumably embodies all these breakthroughs; quite the opposite. Investment in IT equipment is below its prerecession share of GDP, whereas non-IT investment has fully recovered. IT replacement cycles are getting longer, not shorter.

At the micro level, I have looked in vain for the transformation that AI is supposed to create and I just don’t see it. Remember how travel agents were going to be put out of business? That was true for a while, but since 2010 their numbers have been growing, as the Information Technology and Innovation Foundation notes in a new report. Will driverless cars put taxi drivers and truckers out of work? Maybe someday, but for now the number of taxi rides and drivers is growing, not shrinking, and freight companies constantly complain about the shortage of truckers. I would love for AI and robots to disrupt the industries that cost me the most, but it ain’t happening. I recently started washing my shirts at home because the dry cleaner doubled his prices.

True we have 10 new ways to communicate by smartphone—email, texting, instagramming, slacking, tweeting, redditting, facebooking, snapchatting, etc.—but none of these took the place of something I actually have to pay for, and they all consume a finite, precious input, i.e., time. The fact we devote so much precious time on something free means social media must have raised human welfare, but by the same logic so did “I Love Lucy.”

•••

Ryan: I don’t think you’ve got this right. You don’t see productivity growth if you look across sectors, but you do see it if you look across firms and cities. See this OECD report on the growing productivity gap between firms at the frontier and laggards, for example. The weird thing isn’t that the technology isn’t turning into productivity growth. It’s that the technology is turning into productivity growth in high-flying firms and cities but isn’t being adopted elsewhere.

One reason is that the bottleneck is intangible capital. Financial capital is abundant. But what you need to make this stuff work is the right combination of know-how and data, which are scarce. The big tech companies can print money selling their expertise and algorithms (and the data they run on) to make firms more efficient. They’re not selling it cheap, and they’re mostly selling it to a handful of huge, multinationals with operations in the same few cities as the big tech companies. Maybe the technology isn’t ripe enough to be sold as services or embodied in technology for smaller firms and individuals. Or maybe the market’s not there because firms aren’t feeling any great pressure to economize on their labour bills. Or maybe the latter reduces the pressure to tackle the former.

The fact that there are tons of people everywhere doing low-wage, low-productivity work is evidence for my argument, not against it. The number of taxi drivers is growing because they’ve gotten cheaper! And they’ve gotten cheaper because a ton of drivers are willing to work for Uber for not so much money. The surprisingly and persistently slow rate of wage growth in the face of really low unemployment and steady hiring is also evidence for my case. And, come on, you shouldn’t be falling for this Bob Gordon argument that modern technology doesn’t amount to much more than social networking. To take just one example: Instant translation has gotten incredibly good. It has, on a number of occasions, made my work as a journalist vastly easier.

Your comment that there is little historical precedent for this is striking. The argument crystallised for me as I looked at economic history. Look at Paul Romer’s explanation for the slowdown in productivity in the 1970s. Or just look at poor countries like India. Economic historians are very comfortable with this kind of argument—much more so than economists. But economists are getting there. Some of Daron Acemoglu’s recent papers get very close to building models of the dynamic I’m describing.

•••

Greg: The OECD data on how the productivity gap between leading and lagging firms is persuasive. But it doesn’t solve the puzzle for two reasons.

First, for technology to transform the economy, it has to diffuse broadly. The fact that it hasn’t may explain why productivity is poor, but it cannot be turned into an argument that therefore productivity is actually great. The diffusion problem may, as you say, reflect insufficient intangible capital or some other obstacle, but this simply makes my point: Automation hasn’t gone far enough. (At some point, the transformation you’re talking about may occur. My argument is that we haven’t seen it yet, or any sign that it’s about to happen.)

Second, technology’s biggest impact on our standard of living should occur among the customers of firms that use technology, not the producers of the technology. The big beneficiaries of the dot-com boom weren’t Cisco but the companies that used Cisco gear to put more of their business online. Yes, firms like Google and Facebook and Amazon have phenomenal productivity, but the test should be, what has it done for the productivity of the rest of us? And we ought to see some sign of it in the data. But we don’t (except for retail).

I am, indeed, going to use Bob Gordon’s argument, because I happen to think it’s true. Google Translate is great, as are tons of other digital innovations of recent years. But too many people cite the ever-rising stock of innovation as evidence of progress when the problem is the stock of innovation seems to be growing more slowly. In other words, it’s not enough to say Google Translate is great. The question is, How many times per year does something as useful come out? Less often than during the heyday of electricity, the internal combustion engine, personal computers or the internet.

Regarding history, I’m referring to the many times people, including John Maynard Keynes, have worried about mass unemployment from technology. It never came true. Mass unemployment has almost always been due to lack of demand, not an excess of supply. History does show that technology revolutions take a long time to raise our standard of living, but this simply reinforces my point that fears of widespread job destruction are premature.

•••

Ryan: I’m not saying productivity growth is actually great. You said that for my argument to be true, productivity would have to be growing strongly in at least some corner of the economy. As we seem to agree, it is. We also agree on what’s not happening, which is diffusion of that productivity and, through that, the achievement of broad gains in consumer welfare. But that’s not an argument against my story. It is the set of facts I’m explaining. You are answering me back with the premise of my book.

Further, my argument is exactly that we are looking in the wrong place for evidence of rapid and disruptive technological progress if we are waiting for mass unemployment. There are many historical episodes in which technological change led to hardship for workers and soaring inequality, for a time at least.

I think Bob Gordon is very clearly mistaken. Google Translate isn’t just another product. It is an example of a general-purpose technology—machine intelligence—that has improved much faster over the last decade than many in the industry anticipated. Gordon thinks driverless cars are no big deal, because to get somewhere people still have to sit in the car whether they’re handling the wheel or not. But that’s nuts. On the one hand, it neglects all the stuff driverless vehicles can do with no humans in them. More importantly, it neglects the fact that driving is an incredibly difficult cognitive problem, and machines which can solve it can do all sorts of other things humans currently do. But investing in the intangible capital needed to figure out where these things can be profitably deployed is costly, and firms aren’t going to do so much of it if bosses aren’t desperate to solve their labour problems.

And when such opportunities are found and jobs are destroyed, that won’t lead to higher unemployment. It will lead to more people competing to work in low-pay, low-productivity jobs, further dulling the incentive to use the technology in other contexts.

•••

Greg:  This is a valid, internally consistent prediction, and we will have to wait to know the answer. I am more optimistic. The evidence I’ve seen is that higher productivity driven by technological change leads to higher wages, even in the industries that are being disrupted.

I agree that AI is a general-purpose technology with great disruptive potential. I’m just skeptical that it is more disruptive than previous general-purpose technologies, none of which created mass unemployment or mass low-wage unemployment. So I think we are in the same place in agreeing that disruption is coming, but we disagree on the outcome for work.

•••

Ryan: Fair enough. And yes, if we were to see sustained strong wage growth that generated more inflation than productivity growth, that would be a big problem for my story. We shall see!