Latest / The Productivity Podcast with Fexingo: Output, Efficiency, and Long-Term Economic Growth / The Solow Paradox and What It Means for AI Productivity
Transcript
- Lucas: So there's this famous line from the economist Robert Solow in 1987: 'You can see the computer age everywhere but in the productivity statistics.' It became known as the Solow Paradox, and it's still one of the most debated puzzles in economics. Luna: And he said that at a time when personal computers were entering offices, factories were automating — but official productivity numbers were actually slowing down from earlier decades. Lucas: Exactly. US productivity growth in the 1970s and 1980s was averaging around 1.5 percent a year — well below the postwar boom of nearly 3 percent. And here was this revolutionary technology, yet no macro-level payoff. Luna: So what explains it? I mean, computers clearly made individual tasks faster. Why didn't that aggregate? Lucas: Economists have proposed several answers. One is measurement: productivity statistics capture output that's easy to price — like manufacturing goods — but the benefits of computers often show up as higher-quality services, more variety, or intangible value like better customer insights, which GDP data struggle to reflect. Luna: Right, if you can now search a database in seconds instead of an hour, that time saving doesn't appear as increased output unless the saved time is used to produce more of something measurable. Lucas: And that's a huge deal. Diane Coyle, the economist who wrote about the 'GDP paradox', has argued that our measurement framework was built for an industrial economy — it systematically undervalues digital goods that have zero marginal cost. Think free search, free maps, free video calls. Luna: So maybe the Solow Paradox was partly an illusion — the benefits were real but invisible to the official numbers. Lucas: That's one camp. But another camp says the lag is real: new general-purpose technologies take decades to diffuse through the economy. They require complementary innovations — like business process redesign, new management practices, and upskilling — before the productivity gains appear. Luna: There's a famous study by Erik Brynjolfsson and his colleagues showing that firms that adopt ai powered decision tools see a productivity boost, but only after they reorganize their workflows, which can take years. Lucas: Exactly. They looked at a large manufacturing company and found that AI recommendations improved quality metrics, but the measured output per worker didn't rise until the firm redesigned jobs around those recommendations. So the paradox might be a feature of the adoption process, not a flaw in the statistics. Luna: And then there's the possibility that some technologies actually don't boost productivity much — they're just entertainment or convenience, not productivity-enhancing in the GDP sense. Lucas: Right. Gordon, the economist behind 'The Rise and Fall of American Growth', argues that the great inventions of the late 19th and early 20th centuries — electricity, internal combustion, indoor plumbing — had far larger impacts on measured productivity than digital technologies ever will. He points out that from 1970 to 2014, total factor productivity growth in the US was around 0.5 percent per year, versus over 1.5 percent from 1920 to 1970. Luna: So he's saying computers are more like microwaves than electric motors — nice to have, but not transformative enough to shift the macro trend. Lucas: That's his view, and it's worth taking seriously. But the counterargument is that we may be in a new wave. Since around 2019, US productivity growth has ticked up — it averaged about 1.8 percent in 2023 and 2024, compared to 1.1 percent in the prior decade. Some economists attribute that to AI and advanced analytics finally starting to bite. Luna: But we're still below the late 1990s boom, when the internet first commercialized and productivity hit nearly 3 percent for a few years. So maybe the payoff is real but smaller than the hype. Lucas: Or maybe it takes longer. A recent working paper from the National Bureau of Economic Research looked at US firms that invested heavily in AI from 2017 to 2023. They found that those firms saw a 3 to 5 percent boost in output per worker — but only after about two years. So for the economy as a whole, we might be in that lag phase right now. Luna: So the Solow Paradox could be resolving, but slowly. And if measurement issues are part of it, we might not even see the full boost in GDP data for years. Lucas: That's the crux. It's also worth noting that Solow himself later said the paradox might be about 'what we mean by productivity'. If you define productivity as output per hour worked, and people now work smarter, not necessarily longer, then maybe the gains are showing up in quality of life — which GDP doesn't capture. But that's a harder sell for economists focused on measurable growth. Luna: So where does that leave us? If I'm a business leader, should I expect AI to eventually show up in my bottom line, or should I be skeptical? Lucas: I think the historical evidence is cautiously optimistic. But it requires patience and organizational change. The firms that see productivity gains from AI are the ones that combine the technology with reengineered processes, better training, and a willingness to change how work gets done. Those that just bolt AI onto existing workflows tend to see much smaller gains. Luna: That sounds a lot like the lessons from the earlier IT revolution — as documented by Brynjolfsson and Hitt back in the 1990s. Lucas: Exactly. So the Solow Paradox isn't a law of nature — it's a reminder that technology alone isn't enough. The hard work of integration is what turns invention into productivity. And that's where the real opportunity lies for businesses today. Luna: If walking through the economy with us has made something click about productivity and technology, there's a small way you can help keep these conversations going. We keep this show entirely ad-free, and that's possible because of listener support. Lucas: Yeah. It's a simple setup — if you found today useful, you can buy us a coffee. The link is buy me a coffee dot com slash fexingo. No sign-ups, no newsletters, just a way to say this is worth keeping free. Honestly, it means a lot every time someone does. Luna: And for everyone listening, thanks for being part of this productivity community. Now, back to the paradox — because I think there's one more twist. Lucas: Sure. And that is the possibility that some productivity gains from AI are being captured by a small number of superstar firms, making the aggregate statistics look weaker than the leading edge. That's been a theme in recent research. Luna: Right, the productivity gap between frontier firms and the rest has widened over the past two decades. If AI accelerates that divergence, then the measured national average might not reflect what the best-in-class is achieving. Lucas: So the Solow Paradox could be a tale of two economies: one where AI is transformative and another where it hasn't yet touched core operations. And bridging that gap might be the biggest productivity challenge of the next decade. Luna: That's a good place to leave it — with a question: if the paradox is partly about diffusion, how do we speed up the spread of these innovations to the middle and bottom of the productivity distribution? Lucas: That's a topic for another episode, for sure. For now, the takeaway is: don't confuse the absence of visible productivity gains with the absence of technological impact. The Solow Paradox taught economists to look deeper — and that's a lesson that still matters today.