Artificial intelligence has already created trillions of dollars in market value and turned a handful of techies into celebrities. The public is equally excited and frightened. Many intelligent people, from Bill Gates to Elon Musk, say that technology has just begun. However, talk to academic economists, and most of them are strangely uninterested in studying the impact of potentially world-changing technology. The center of gravity in AI economics is shifting away from universities – and a gang of “AI-pushed economists” are leading the charge (see Chart 1).
University researchers can move faster whenever they want. After Lehman Brothers, an investment bank, collapsed in 2008 and triggered the global financial crisis, economists transformed the study of bank runs and credit crises from a niche pursuit to mainstream interest. In 2020, two months after COVID-19, nearly a third of the working papers on economics published by the US National Bureau of Economic Research (NBER), a prestigious repository of economic ideas, focused in some way on the effects of the pandemic. Some of this work has gone mainstream, including from work-from-home experts Nick Bloom of Stanford University and Emily Oster of Brown University, who studied school closures.
The AI era began three and a half years after the launch of ChatGPIT, in contrast, economic analysis of the technology remains comparatively rare. The proportion of NER papers focusing on AI is increasing, but not particularly rapidly (see Chart 2). Even in 2024, when the Covid emergency ends and the AI era truly begins, the number of Covid-related papers still exceeds the number of AI-related papers. This year NBER is likely to host more conferences on health care than AI.
Some academic economists have taken advantage of the AI opportunity. Stanford’s Susan Athey is exploring what would happen if AI put people out of work. Basil Halperin of the University of Virginia has written eloquently about how financial markets value AI developments. Yet none are as recognizable as Mr. Bloom or Ms. Oster. And some economists seem to recognize the research potential. “I’m amazed that few of my colleagues have even tried talking to Anthropic or OpenAI,” says one superstar academic economist, who emphasizes that he talks to AI labs constantly.
Much of the research that exists is highly abstract. Daron Acemoglu of MIT is the highest-ranked AI winner in economics, according to Ideas/Repec, a bibliographic database dedicated to economics. A paper by Mr Acemoglu published in early 2024 presents a complex model of economic growth under AI, and indicates modest overall productivity gains. It has already received over 1,000 citations. But Tyler Cowen of George Mason University argues that the model underestimates the potentially transformative impact of new AI products coming to market. “The benefits from AI are very small because it is assumed that AI will not do new things.”
Many empirical studies related to AI also appear to be based on flawed assumptions. A paper by Stanford University’s Eric Brynjolfsson (ranked fifth by Ideas/Repeak) and colleagues shows that youth employment in AI-exposed occupations has declined sharply, meaning the technology is already changing the labor market. Yet to blame AI for this trend is to admit that companies started laying off young employees almost as soon as ChatGPT was first released – a product that was nowhere near good enough to replace humans.
Academic economists can be slow and lazy for two reasons. The first relates to the type of shock that AI represents. Covid changed the world almost overnight in 2020, and its impact became visible in the data almost immediately. On the contrary, AI is transforming the economy under its bonnet. The average unemployment rate in the OECD, the club of rich countries, is about the same as when ChatGPT was first released (see Chart 3). What’s more, the GDP numbers contain practically no AI-specific data – for example, investments in AI data centers can only be estimated. With no obvious macroeconomic effects and no microeconomic data, there is little to analyze.
The second factor is that economists are, as a rule, a fairly techno-skeptical group. Historical research shows that technology increases incomes, but gradually, due to all kinds of non-technological factors (including financial frictions and cultural resistance). It took several decades for the technological successes of the Industrial Revolution to translate into rapid growth in Britain.
A recent paper by Mr. Halperin and his colleagues, reporting on the results of a survey, captures this skepticism. Even in a scenario where AI progress by 2030 is “rapid” – meaning AI could compete with or surpass the most intelligent humans – the average academic economist expects US GDP growth to be only 3.5% in 2050 (compared to 5.3% for AI researchers). According to a survey by the University of Chicago, only 11% of leading economists agree that the use of AI will “substantially increase unemployment rates in advanced countries” over the next decade. If most academic economists do not think AI will be transformative, they may prefer to stick with other research areas that they consider more important.
The AI-curious economist is looking for a comfortable home two spots away from the academy. The first is the government, and in particular the statistical offices and central banks. Surveys from the U.S. Census Bureau and Statistics Canada track AI adoption across the economy. The Bank of England’s monthly “Decision Maker Panel” has explored businesses’ perceptions of AI, while the British government recently created an “AI Economics Institute” to improve research on the topic. At a recent conference at the OECD, government representatives puzzled over how to update productivity measures for the AI age. Much of this work won’t set the world on fire, but it serves an important public service: building the data infrastructure that future economists will rely on.
The second, more important place is at the forefront of technology. In the 2010s, AI laboratories produced many brilliant computer scientists to design their models. Ufuk Aksigit and colleagues at the University of Chicago found that by 2019, more than two-thirds of AI researchers worked in industry, up from less than half in 2001. Now something similar is happening with economists also.
Anthropic has hired Anton Korinek of the University of Virginia (who comes in second in the ideas ranking) to its economics-research team. OpenAI appointed Ronnie Chatterjee of Duke University as its chief economist. The tech giant’s in-house Frontier Lab, Google DeepMind, recently appointed Alex Imas of the University of Chicago as its “Director of AGI Economics” (a reference to “artificial general intelligence” that will match or best humans at most intellectual tasks). According to rough data from The Economist, a few dozen AI-pilled disillusioned scientists have accepted jobs in big laboratories.
The attraction of AI Lab is obvious. They have access to the best data as well as the ears of policy makers. Accept a position there and soon Silicon Valley’s favorite podcaster Dwarkesh Patel will ask you on his show. Tech companies also have more money than universities. Even a relatively junior economist position in an AI lab can pay $300,000 or more per year: perhaps not much compared to an AI programmer, but significantly more than the income of an early-career professor teaching Econ 101. (A lucky few may get stock options in the world’s most popular companies.)
The quality of extramural AI research is increasing. Working at the Peterson Institute for International Economics, a think-tank, Mr. Korinek and Patrick McKelvey of the Bank of Canada have created what they call an “AI GDP” for the United States. The paper shows that, when measured precisely, this increases by more than 2,000% in both 2024 and 2025. Mr Imas has published a useful tracker of the impact of AI on productivity. According to their judgment, there is encouraging evidence of small-scale productivity gains, but little evidence of large macro impacts.
This is all very exciting (at least for artificial intelligence economic journalists). But for every clever study by Mr. Korinek or Mr. Imas, laboratories still produce useless results. Anthropic’s “Economic Index”, released to great fanfare, is not actually an index, but a random collection of data about the use of its chatbot, Cloud. In March Anthropic published a report concluding that “people become better at using the cloud through experience”. No! Last year OpenAI published descriptive work showing that 20-25% of messages on ChatGPT involved “seeking information”. Riveting stuff.
There is no doubt that the quality will improve with time. Still, if the frontier of AI research shifts inside companies, economists may follow the path already taken by “technoeconomists” at Microsoft, Google, and elsewhere. These wunks typically spend less time on big questions of social importance (asking, say, whether social media is good for kids) and more time on narrow questions, such as how to best design an auction to sell advertising. Mr. Axigit’s study states that after making the permanent transition from academia to industry, AI researchers produce fewer papers but more patents, which is actually a “reorientation from open science to proprietary innovation.”
Then there is a conflict of interest. Lab researchers may face pressure to publish work that makes AI useful and safe. Last year Tom Cunningham, an economics researcher, left OpenAI after reportedly becoming frustrated with restrictions on what he could and could not publish. He arrived at Meter, a research institute dedicated to evaluating AI models and the threats they pose. In a world of great potential as well as great threats, society needs selfless researchers who say what they really think. Academic economists have a basis for formulating this.
