Janet Bush of the McKinsey Global Institute interviews Chad Syverson in “Unpacking the Mysteries of the Global Economy” (July 2, 2024, audio and text available).

Does productivity growth just mean lost jobs?

Janet Bush: There’s a perception that productivity means efficiency and lost jobs. I remember somebody said to me, “Oh, productivity—you’re fired.” Unpack that for us.

Chad Syverson: That is an example of the fallacy of reasoning causality from an accounting identity. There are many specific ways to measure productivity, but they’re all basically ratios of output to input, how much comes out of a production process divided by how many inputs go into it.

And the notion that you’re describing with that person’s comment comes from looking at that definition and thinking, “Oh, that’s how you causally affect productivity. So OK, I want productivity to be higher. It’s outputs over inputs, so if I make inputs smaller, productivity will go up.”

Well, the problem with that is—and this is true whenever you reason from an accounting identity—it’s not just inputs that are changing when you decide to cut inputs. You know, those inputs are doing something, presumably, and you’re going to affect what they’re doing if you try to cut those inputs, like, say, workers or worker hours.

And that might be useful stuff that makes output. And it’s quite possible you could actually reduce output even more than you reduce input, so therefore, your productivity actually has gone down.

It’s kind of interesting. You know, I totally understand sort of the sentiment behind what that person said. I hear it a lot, but it’s usually in that direction, the messing up the identity for causality. Because if someone said, “Oh, I need productivity to go up, I know what I’ll do, I’ll just make more output”—if you said that to someone, they’d say, “OK, what magic wand do you have that lets you wave and get more output for nothing?”

Because everyone recognizes, well, if you want more output, you need more inputs, too, et cetera. But somehow that doesn’t quite always become as obvious when someone does the inverse, which is, “Well, I’ll just cut inputs, and of course productivity will go up.” But as I said, that’s not the only thing that’s going to change when you do that.

On the enormous potential of artificial intelligence for productivity gains:

I will say, I think it’s the best candidate for a new general-purpose technology we’ve had in decades. And it’s made me more optimistic that we will end the productivity growth slowdown than anything else that’s happened since I started looking deeply at the slowdown ten years ago. So, yeah, I’m on the optimistic side.

I think it has amazing potential. As you say, it hasn’t diffused that widely yet, but the early returns, so to speak, I think are quite optimistic. … But one lesson of general-purpose technologies is, their full effect comes when they’re put together with complementary investments, often intangible.

It’s not usually the direct replacement effect that drives the productivity gains. It’s these complementary things. What that means is, you wouldn’t look at AI and say, “Oh, what does AI do?” “Well, it predicts text, so I’m going to go look and see where text prediction would be the biggest thing.” OK, if you’re a lawyer and you’re writing briefs, you’re not going to do that anymore. That all may happen. I’m not denying that’s going to be part of what AI does. But I think the biggest things and the broadest things that AI could do, we haven’t really fully grasped yet, because it needs to be put together with other stuff that’s currently being invented and created.

My guess is, the biggest, the most affected sectors, we don’t really know yet. And we might be surprised actually by quite a few of them.  … Something I’ve written about lately is that there can also be a period when new technologies arise where you actually get slow measured productivity growth. And so, the technology can be present. It can be being placed into service by businesses, but you don’t actually see it in the productivity numbers. …

This is work I did on what’s called the productivity J-curve. You get this period of initial undermeasurement of the true productivity effects of new technology, and then, later, overmeasurement.

This is work I did with Erik Brynjolfsson and Daniel Rock. We did some calculations treating existing technologies, kind of going back in time and supposing, “Oh, here comes this new technology. We’re going to pretend like it’s got this AI sort of effect going, where you’re not going to fully see it early and to compute exactly what you ask.” Like, how long does this stuff take for it to work through this measurement issue? And the answer is, we looked at computer hardware, computer software, and R&D spending in general. In each of those cases, the answer is decades. You can have a period of undermeasurement that’s ten to 20 years long. And then, of course, the overmeasurement period on the back end can last just as long. The size of this undermeasurement varies over that period. So it might be five to ten years before you hit bottom of the undermeasurement, and then you start coming back in the other direction. But the point is, you can go pretty long periods of time where the technology’s out there, it’s being installed, it’s starting to be used, but you’re still not seeing its full effect reflected in the productivity statistics.

On a possible resurgence of dynamism in the US economy

[T]here’s been a long-running decline in measures of dynamism in the economy, especially in the US, but throughout much of the OECD. What do I mean? Total labor turnover, people leaving jobs and getting new jobs, for example. Business formation. How many new businesses are being created every year? Those things have been on a long-run decline, and when I say long run, I mean, back to the ʼ80s at least. OK, so there have been 30, 40 years of slowdowns in measures of dynamism. And some people were saying, “Well, maybe this is tied to the productivity growth slowdown.” Like, these chickens are coming home to roost.

Well, what’s happened since COVID, as we’ve emerged from COVID, those things have turned around. After decades of decline, labor market dynamism accelerated again. Business formation in the US is up one-third. One-third more businesses are being formed per month now than were in 2019. This isn’t just folks who are tired of the office life starting a consulting company in their spare bedroom. If you look specifically at what are called high-propensity businesses, or businesses that at foundation have features that we know predict hiring and growth for those businesses in the future, those are also up a third.

So it really does look like there’s a reinjection of whatever the secret sauce is that creates a dynamic economy in the last several years. And I think that that’s what makes me encouraged that the last three quarters of pretty fast productivity growth might continue. We might actually accelerate through that return to trend, rather than just stop at the trend.