Matt Chase
If you’re not sure quite how to peg the economy these days, you have plenty of company among many middle-class Americans working to get ahead. On one hand, jobs are plentiful and growth is solid. On the other, no sooner does the rent or utility bill come due than it seems like the only things bounding ahead are prices.
US politicians have various suggestions to boost the middle class. As a presidential candidate, Vice President Kamala Harris said she would cut taxes, ban price gouging, and make the purchase of a first home more affordable. But behind the lagging sense of affluence is something else entirely: a slowdown in productivity growth.
Productivity is the amount of labor and capital required to produce each unit of output, and it is what ultimately determines how much stuff a society creates out of its available resources. In its early days, the United States needed the vast majority of its workers in the fields to grow enough food. Thanks to big advances in the productivity of equipment and people, the country today needs only a fraction of its labor force to feed itself. That’s freed everyone else to devote their productive potential to other labors, such as researching a cure for cancer, flying planes, and influencing people through social media.
Since growth in labor productivity is a measure of how much more wealth workers churn out for each hour they put into their jobs, it’s the economic equivalent of free stuff. Even seemingly small gains, compounded over years and decades, translate into big differences in GDP and household incomes.
Productivity growth experienced something of a lost decade leading up to the COVID-19 pandemic. Twenty-nine of the 30 members of the Organisation for Economic Co-operation and Development suffered similarly disappointing performances, making the rout global.
Economists have been closely watching productivity figures, and many are now wondering the same thing: Are trends changing for the better? If so, that would have implications across the economy. It could mean not merely more jobs but higher-paying ones, leading to higher real incomes. Some research suggests there is cause for optimism—but questions remain about who, exactly, would benefit.
It is hard to overstate the importance of the economy in American life. Just the feelings that citizens have about the state of their pocketbooks and prospects can decide elections. Democratic strategist James Carville famously summed it up with a phrase that he used to focus staffers working to get Bill Clinton elected president: It’s “the economy, stupid.”
Productivity may be far less visible than an economic phenomenon such as inflation, but it’s just as real. Economists usually measure total factor productivity, quantifying how many more goods and services are produced from the same amount of inputs, including capital, labor, and other materials.
Since World War II, US productivity has been all over the map and can be split into four broad trends, Chicago Booth’s Chad Syverson explained in an essay for Chicago Booth Review. (Read “Why Hasn’t Technology Sped Up Productivity?”) In the quarter century following the end of hostilities, productivity grew at a brisk average annual pace of 2.7 percent. Then, in the two decades through 1994, it fell to about a tepid 1.5 percent, despite the spread of computers and other technologies. Productivity growth then nearly doubled, approaching a 3 percent annual average through the mid-2000s before once again slowing to less than half that rate. In the second quarter of 2024, the Bureau of Labor Statistics estimated, productivity grew at an annual rate of 2.3 percent.
While labor productivity grew rapidly post-World War II, it has alternately slowed and surged since then. Lately, there’s been a modest uptick.
Yet it can be hard to make sense of these figures. Why, for example, would productivity growth fall as technology spread across the world? Stanford’s Erik Brynjolfsson called that the productivity paradox, and in 1987, the late Nobel Laureate Robert M. Solow wrote in The New York Times, “You can see the computer age everywhere but in the productivity statistics.”
Some research suggests that as technology advances, the economy experiences innovation bottlenecks—and addressing them allows productivity growth to race ahead. After all, before steam locomotives could run, steelmakers had to turn out railway tracks. Automobiles couldn’t deliver their full potential until road crews replaced horse paths with paved roads. And these days, electric vehicles won’t take over the roads until chargers become as easy to find as gas stations.
Using Bureau of Economic Analysis data, MIT’s Daron Acemoglu and David Autor and Booth’s Christina Patterson analyzed spending on components and patent data for close to 500 industries between 1958 and 2011. They find that most of the US productivity slowdown can be explained by a “sizable increase” in asymmetrical innovation and productivity gains across industries. A few industries that made large contributions to GDP, including medical instruments, gasoline engines, and industrial valves, suffered some of the largest bottlenecks due to the uneven growth of their suppliers. (For more on their findings, read “Lopsided Innovation Causes Productivity Slowdowns.”)
If eliminating such kinks could lead to a productivity explosion, doing so should be a top priority. Imports are one way, but this approach has limits at a time when overreliance on foreign suppliers has become a national security concern, the researchers write. Another avenue: Government and industry could invest with an eye to balancing the uneven research and development that leads to such bottlenecks. As an example, public and private money is being put toward improving electric-energy transmission and storage—and that could jolt growth in electric cars and trucks, as well as other facets of a greener economy.
The upshot is that breaking up the bottlenecks should give productivity a boost. “If and when lagging industries ultimately increase their innovation and productivity growth rates, a rapid takeoff in aggregate productivity should ensue,” the researchers write.
Total factor productivity growth in the US has been sluggish since the mid-2000s, which research attributes to uneven productivity gains across suppliers of large manufacturing industries. Some suppliers have advanced rapidly while others have lagged behind, restraining the manufacturers’ overall TFP growth.
Digging further into the statistics uncovers another cause for optimism: Many companies may be making intangible investments whose payoffs are undercounted in current productivity statistics.
General-purpose technologies such as computers and artificial intelligence require significant investments in complementary areas. For example, after stores installed self-pay kiosks, their managers had to teach employees and customers to use them before self-checkout lines were widely adopted.
Brynjolfsson, University of Pennsylvania’s Daniel Rock, and Syverson devised a method to measure differences between a company’s observed investments and its market value. They find that the large investments companies make early on in adopting general-purpose technologies often involve intangibles, such as worker training and the retooling of business processes.
Conventional statistics account for these investments as expenses rather than capital creation, which initially underestimates the productivity-enhancing effects, the researchers argue. “We count a new factory as output but we don’t count intangible investments as output,” Syverson says. “You’re not just burning resources. You’re creating something useful in the future.”
When the benefits of the intangible investments are later harvested, the mismeasurement swings in the opposite direction, the researchers find. The result is a productivity J-curve that helps explain why the advent of general-purpose technologies is often accompanied by an initial productivity slowdown that’s later followed by a burst in output.
Given that society is in the early stages of adopting AI, there may be reason to believe we are currently in the productivity underestimation phase, Syverson argues. As was the case with earlier technologies that had broad applications, AI can only reach its full potential after companies invest in necessary intangibles, such as training workers, reorganizing workflows, and educating customers in how business will be conducted under the new paradigm.
There are yet more reasons to anticipate a productivity pickup. As postwar history demonstrates, there’s nothing unusual about economic output advancing in fits and starts, and a number of precursors that preceded periods of robust growth in the past are now starting to emerge. Among them: increases in business formation.
Entrepreneurship is an important driver of economic growth, technological advances, and, eventually, higher productivity—and it was surprisingly hot during the COVID-19 pandemic. Many people were stuck at home, temporarily laid off, and had the time and motivation to get creative. In the meantime, Americans started shopping online. Hence, the time was ripe for networking remotely and launching new businesses. This is reflected in the data: There were surges in business applications in the US in both 2020 and 2021, write the Federal Reserve Board’s Ryan Decker and University of Maryland’s John Haltiwanger, pointing to US Census Bureau statistics. Applications were still high in late 2023 and even into 2024. They’ve since fallen but remain above pre-pandemic levels.
Some business plans were designed to capitalize on COVID-era changes in how people live and work. Another notable feature of the surge in applications that the researchers find was the prominence of filings by likely employers, those who are particularly inclined to hire workers and generate growth. Typically, it takes a year or two to begin learning if applications result in actual job openings, or if the businesses peter out. The pandemic-era surge in new business applications stands in sharp contrast to the economic weakness on display during the Great Recession, and research suggests that it has translated into genuine entrepreneurial activity resulting in jobs that people want.
Business applications spiked during the pandemic, including for businesses considered likely to hire workers. Although applications have since decreased, they remain above historical levels.
Decker and Haltiwanger used numerous data sources to track business formation activity, including the Census Bureau’s Business Formation Statistics, which draw on requests for new employee identification numbers submitted to the Internal Revenue Service and include the bureau’s modeling of business characteristics. Looking at features such as corporate structure and hiring plans helped them identify trends among the sorts of startups most likely to transition from new businesses applicants into actual employers. They also used several data sources on actual hiring by new businesses, looking at business creation patterns across industries and geography.
The pandemic’s increased entrepreneurship left its mark on the economy in a number of ways, they write. Companies became younger and smaller, on average. In major metro areas, a “donut pattern” emerged, with less growth in city centers than in surrounding areas, closely tracking trends in work-from-home activity.
The researchers also relate the rise in business formation to stories that developed among economists and in the news media to describe the job market, noting that reported opportunities to create a new company, or to work for one, appear to have played a significant role in fueling the Great Resignation that began to emerge in early 2021.
The surge in new-venture formation occurred after decades of declining business dynamism, represented by a shift of activity toward large, mature companies in the US. The pace of job reallocation (the rate at which jobs flow from shrinking businesses to growing ones) fell by about a quarter between the 1990s and the months just before the pandemic, driven in part by a drop in the rate of formation of new businesses that hire paid employees.
In detailed labor-market statistics, Decker and Haltiwanger find more hope for productivity growth in what they peg as early signs of a revival in business dynamism. And in a separately published note, they write that the jump in business applications included a large number of tech companies.
It could still be too early to tell if business entry will be persistently higher in a manner that would suggest an end or reversal of pre-pandemic trends, the researchers caution. Before it can be deemed long-lasting, the rate of new-venture entry would have to remain elevated for an extended period, and at least some of the pandemic’s entrepreneurial entrants would have to grow substantially. However, in a September 2024 update, the researchers write that “while there is clearly some cooling evident in the most recent data, potential entrepreneurs continue to submit business applications at an elevated pace, particularly in sectors that are intensive in high tech activity.”
If we are indeed entering a better age for productivity, that would make the world a more prosperous place—but everyone wouldn’t necessarily benefit equally. If the sectors where output rises the most differ from the past winners, wage gains could follow suit. This reshuffling of incomes could, in turn, alter the economic pecking order of society at large.
Over the long haul, technological advances tend to create relative winners and losers. When the first wave of computing power transformed the economy several decades ago, the main beneficiaries were workers already at the high-skilled, high-pay end of the market, according to research by MIT’s Autor. Thus, the affluent got more affluent, and income inequality increased.
As returns trickle in from early case studies involving AI, there’s reason to believe the pendulum is poised to swing in the opposite direction and that low-skilled workers will reap a relatively large share of the benefits. This appears to be the case with customer service agents, a job class with one of the highest AI adoption rates to date. In a study released last year, Stanford’s Brynjolfsson, MIT’s Danielle Li, and MIT PhD student Lindsey R. Raymond analyzed the staggered deployment of a chat assistant tool built on a generative pre-trained transformer, or GPT, developed by the research organization OpenAI.
Increasing output requires promoting employment mobility.
Job-Hoppers Could Boost Productivity
The decisions of companies, governments, and educators will help to shape the ultimate outcomes of the A.I. revolution.
A.I. Is Going to Disrupt the Labor Market. It Doesn’t Have to Destroy It.Generative AI is a class of machine learning that analyzes patterns in existing data to create new text, music, video, and still images. The researchers looked at 3 million chats by about 5,000 customer service agents who worked for a Fortune 500 software company. The AI model was used to monitor customer chats and provide the agents with real-time suggestions about how to respond. It was ultimately left to the agents to either follow the model’s suggestions or ignore them, and customers weren’t informed of any AI involvement.
The AI assistance markedly improved worker productivity, the researchers find. Agents using it shortened the time it took to handle individual chats and increased the number they handled per hour. Overall, the agents were able to boost the volume of issues they successfully resolved by 14 percent per hour.
Access to AI assistance also improved how customers treated agents, as measured by the sentiment of chat messages. When the tool was in use, customers were less likely to question the competence of agents by requesting to speak to a supervisor. Retention of new workers jumped while overall attrition fell, the researchers find.
Notably, productivity improved the most among less-skilled and less-experienced workers, who increased the number of issues they resolved per hour by 34 percent. In other words, the AI tool helped newer agents move more quickly up the experience curve. In contrast, the tool’s introduction had a minimal impact on the productivity of more-experienced and better-skilled workers. Among the most-skilled agents, it may even have decreased the quality of conversations, the researchers find.
Increased productivity among less-skilled workers should, in theory, make them more valuable to their employers. That said, the research did not seek to shed light on how employment and wages could be affected among workers aided by AI tools. Nor did the methodology allow the researchers to observe changes in wages, overall labor demand, or the skill composition of workers hired for customer service jobs.
Thus, a large question remains about what AI tools will mean for these workers—and others. Better productivity is good news overall, but what’s true in the aggregate isn’t necessarily true for individuals. Will employers respond to novice workers’ increasing productivity by hiring more of them—or could they seek to develop more powerful AI systems that replace lower-skilled workers entirely? “I think the evidence so far is that AI is a task replacer, not a job replacer,” says Syverson.
Even so, the effects on pay are also unclear. Developers create AI tools by feeding them relevant raw data. Media companies and others have begun to demand compensation when such data is scraped from their intellectual property. Some owners have declared their data off-limits to AI developers entirely. In the customer service industry, developers need raw data involving interactions between high-performing workers and customers. It remains unclear whether the customer service workers who produce them will be compensated for these data they generate, and if so, how.
How will AI affect economic productivity more broadly? Generative AI tools have performed well in laboratories, but doubts remain about whether real-world results will follow the same trend. So far, the adoption of these tools has been concentrated among large, young companies with relatively high productivity, according to three studies of the US and OECD countries. And AI’s broader economic effects remain unclear.
Like other once-new technologies did, AI is facing some resistance, and a host of things could interfere with its advancement and development. In 2023, The New York Times sued OpenAI and Microsoft, accusing them of copyright infringement for using its articles to train the large language models that power the chatbot ChatGPT. Other companies could respond by locking down their data.
Further, if the technology threatens US jobs, Congress could conceivably impose restrictions. And AI’s credibility could be undermined by instances of chatbots producing misleading information with real-world consequences. Screwups could create a popular backlash or make organizations resistant to adopting AI. The news site Tech.co has a page dedicated to tracking what it calls AI’s “worst failures to date.” Among them, X’s chatbot wrongly accused a player for the NBA’s Golden State Warriors of vandalizing homes after misinterpreting the basketball slang phrase “shooting bricks.” In New York, a chatbot that was released to help small companies find legal information falsely suggested that it’s permissible to fire workers for disclosing a pregnancy or refusing to cut their dreadlocks.
But MIT’s Autor takes an optimistic view of what AI could mean for workers. He writes about the technology’s broad potential to enable workers to capitalize on the relevance, reach, and value of human expertise that they don’t personally possess. Specifically, he argues that certain tasks that are currently performed by elite experts could be dramatically democratized. If used well, AI could extend certain medical care beyond doctors, document production beyond lawyers, and software coding beyond programmers. AI’s potential to extend the efficiency and scope of expertise is especially important in an era when birth rates and the labor pool are shrinking, Autor writes.
He cites a simple analogy from a Pew Research study indicating that slightly more than half of adults find the video site YouTube “very important” in figuring out how to perform new tasks. Autor’s point is that the content democratizes expertise by letting those best positioned to solve problems share their knowledge with nonexperts.
While AI is not a major player in the creation of YouTube how-to videos (not yet, at least), Autor cites a trio of studies to highlight evidence that it is already being used as a teaching tool. One controlled experiment—by Microsoft’s Sida Peng, GitHub’s Eirini Kalliamvakou and Peter Cihon, and MIT’s Mert Demirer—finds that software programmers who were given access to generative AI–based tools completed tasks 56 percent faster than a control group.
A second study, by MIT PhD students Shakked Noy and Whitney Zhang, finds that AI led to significant improvements in the speed and quality of work by writers involved in marketing, grant proposals, consulting, and other tasks. ChatGPT enabled the best writers to work faster and the less capable to improve both the speed and quality of their output. The productivity gap between the excellent and the adequate narrowed as a result. Overall, the time needed to complete the tasks studied fell by 40 percent.
If similar trends spread broadly across the economy, a US labor model that’s been hollowed out by waves of automation and globalization could begin favoring middle-class workers, writes Autor.
If AI does usher in such structural changes, the effects would be profound. It could help restore the stature that many middle-class workers once enjoyed in the postwar years while lowering costs in critical sectors including healthcare and education. Such a happy ending is neither an inevitable nor an intrinsic consequence of AI. But, in Autor’s view, it is an economically coherent and morally compelling one. And if it’s correct, the long productivity slump would soon be history.
When consumers barely trust institutions, banking fines might lead people to withdraw their money.
Are Big Bank Penalties Good or Bad for the Financial System?
While consumers are a little hazy about overall inflation, asking them about prices for individual categories yields more realistic forecasts.
People Can Forecast Price Rises—If Asked the Right Questions
Ralph Nader joins the podcast to discuss ethically profitable business, the political disillusionment of the public, and the state of capitalism today.
Capitalisn’t: Ralph Nader’s CapitalismYour Privacy
We want to demonstrate our commitment to your privacy. Please review Chicago Booth's privacy notice, which provides information explaining how and why we collect particular information when you visit our website.