Faster, Please!
March 20, 2025
When will artificial intelligence begin to reshape the American and global economies? It’s a multi-trillion-dollar question. Economists, true to form, will likely be the last folks to declare a revolutionary Age of AI finally at hand.
Economic history reveals a familiar pattern that explains this scholarly hesitation: Productivity gains from transformative technologies — from electrification to computers — typically manifest in statistics (and those stats are volatile) long after their practical, real-world impacts becomes evident elsewhere. The profession, thankfully wedded to empirical rigor, maintains a stubborn reluctance to pronounce seismic economic shifts without robust quantitative evidence. This caution stands in stark contrast to technologists and CEOs who are far more willing to forecast a technological tsunami of change fast approaching.
Silicon Valley predicts supersmart AI will allow us to, say, cure cancer and Alzheimer’s. Economists predict supersmart AI will allow doctors to spend less time taking and organizing notes during patient exams. You get the picture.
And among economists, those working for the Federal Reserve might be the most cautious of all. That’s why I took note of an intriguing speech recently given by economist Austan Goolsbee, president of the Federal Reserve Bank of Chicago, at the Stanford Institute for Economic Policy Research. Pointing out that US productivity growth since late 2022 has notably outpaced pre-pandemic trends, Goolsbee gently suggested that generative AI and other technologies might be the driving force behind this welcome acceleration.
Four theories have emerged for the outperformance, as cited by Goolsbee. Three suggest the surge is temporary: the rise of remote work, better worker-job matching following the Great Resignation, and increased entrepreneurial activity. Each would provide a one-off boost to productivity levels rather than a sustained increase in the growth rate.
But that fourth theory? Well, here’s Goolsbee:
The fourth and final explanation is that this boom in productivity has been tech and AI driven. I realize that might have been where many of you first started, but note that economists are still skeptical—mainly because there hasn’t been enough adoption yet to explain why the economy-wide productivity growth rate would’ve increased this much. But here’s a key point, a key difference from the other three explanations: If this surge in productivity growth is the result of a new technology—whether that’s AI or something else—then history shows it is possible that this surge is not just a one-time bump. It could keep moving through the economy, industry by industry.
An intriguing bit of good news in the data
Goolsbee went on to cite Chicago Fed analysis that shows many of the industries experiencing the most significant productivity improvements are tech-intensive sectors such as internet publishing, e-commerce, and computer system design. This pattern resembles previous general-purpose technology diffusions where productivity gains initially concentrated in the tech-producing sectors before gradually spreading throughout the economy, according to Goolsbee.
Goolsbee: “Now, of course, being associated with technology doesn’t prove this is a general-purpose technology. But it could be a reason to expect continued productivity growth, rather than just a one-time jump.”
In other words, an enticing bit of early evidence — evidence, unfortunately, less persuasive to the economics team at Goldman Sachs. From the bank:
As we first argued two years ago, generative AI could eventually raise US labor productivity by 15% following its full adoption. Several commentators—including Chicago Fed President Austan Goolsbee in a recent speech—have speculated that generative AI has already boosted productivity growth and is one reason why US labor productivity has recently outperformed (as exhibited by its outsized 2.7% increase in 2024). We disagree with this view, and in this Global Economics Comment we show that the overall labor market and productivity impacts from AI have so far been very small despite our still positive longer-run outlook.
And some bummer news in the data
The data paint a nuanced picture. AI adoption has now reached 7.3 percent of American firms, according to GS, and there are some early signs of impact in the form of “productivity benefits in a few specific areas, most notably computer programming, customer service call centers, management consulting, and legal services.” The flipside: Employment in computer programming, management consulting, and call centers has weakened relative to pre-AI trends. Job postings in AI-exposed industries have fallen more sharply than in less-exposed sectors.
But the broader economic impact remains slight. Goldman’s economists found no meaningful correlation between AI exposure and key labor-market indicators in 2024. In the aggregate, job growth, unemployment, working hours, and wages showed no statistically significant relationship with AI adoption. Aargh!
Most telling, perhaps, “neither AI exposure nor adoption rates were correlated with industry-level productivity growth in 2024” — undermining claims (and hope) that AI underpinned America’s robust 2.7 percent productivity gain last year.
This disconnect between expectation (even hype) and reality reflects the still-nascent state of AI deployment. The technology appears concentrated in specific industries rather than broadly transforming the economy. For now, then, AI’s economic impact still appears to be a story of future potential rather than present reality. But let’s not dismiss that potential! Goldman sure doesn’t:
While recent improvements in AI model capabilities, declining costs, and increased competition suggest that this dynamic could change in the coming years as adoption becomes more accessible following the build out of platforms and applications, our analysis suggests that the labor market and productivity impacts of generative AI have so far been negligible outside of a few specific industries. Our forecasts continue to assume that US adoption reaches levels necessary to impact aggregate productivity statistics in 2027 with a peak impact in the 2030s, with other DMs (and China) lagging this timeline by a few years
I’m taking a holistic approach when looking for signs of transformative AI moving out of the lab. I look at everything: business anecdotes and surveys, economic data, expert forecasts, prediction markets, and financial markets. Yet it’s my hope and expectation that before too long, those signs will be too obvious to miss or to ignore.
Learn more:Humanity’s Capacity To Grow and Thrive: A Quick Q&A With … Research Analyst Ben Landau-Taylor | The Economics of AI Apocalypse | Have We Really Passed Peak Brain Power? | The Right vs. The Robots