...
...
Next Story

AI is swallowing maths. Is economics next?

The dismal science braces for disruption

Published on: Nov 22, 2026, 14:26:26 IST
The Economist
Advertisement

IN 1930 JOHN MAYNARD KEYNES predicted that capital accumulation and technological progress would allow future generations to lead an affluent life of leisure. Keynes might not have foreseen the effect of accumulated capital and progress in technology on his intellectual descendants. After artificial-intelligence models solved hundreds of open problems in mathematics, the dismal science is bracing for unleisurely disruption. The angst may be less existential than among mathematicians, says Ben Golub, an economic theorist from Northwestern University. But AI

PREMIUMAI is starting to change economics by cutting the cost of producing research. Will it also start eroding the human advantage in ideas? (Unsplash)
AI is starting to change economics by cutting the cost of producing research. Will it also start eroding the human advantage in ideas? (Unsplash)

IN 1930 JOHN MAYNARD KEYNES predicted that capital accumulation and technological progress would allow future generations to lead an affluent life of leisure. Keynes might not have foreseen the effect of accumulated capital and progress in technology on his intellectual descendants. After artificial-intelligence models solved hundreds of open problems in mathematics, the dismal science is bracing for unleisurely disruption. The angst may be less existential than among mathematicians, says Ben Golub, an economic theorist from Northwestern University. But AI is starting to change economics by cutting the cost of producing research. Will it also start eroding the human advantage in ideas?

PREMIUMAI is starting to change economics by cutting the cost of producing research. Will it also start eroding the human advantage in ideas? (Unsplash)
AI is starting to change economics by cutting the cost of producing research. Will it also start eroding the human advantage in ideas? (Unsplash)

The use of large language models is already pervasive. Stefanie Stantcheva of Harvard University, one of the editors of the prestigious Quarterly Journal of Economics (QJE ), observes a change in the papers which are being submitted: “the figure format, some of the analysis, the more polished text”. In July about one in four working papers filed with the National Bureau of Economic Research, an American repository, contained AI-assisted writing, according to an analysis by Paul Goldsmith-Pinkham of the Yale School of Management. As papers become easier to write, some journals fear being inundated. The Journal of Political Economy (JPE ), another top-tier outlet, will increase its submission fee to as much as $500 per paper in 2027 to discourage slop.

It is not just the prose. The analytical core of economics papers can increasingly be done by AI agents. David Yanagizawa-Drott of the University of Zurich has started a research programme dubbed Autonomous Policy Evaluation (APE). Like human economists before it, Dr Yanagizawa-Drott’s AI uses publicly available data to gauge the effect of policies. Among other things, it found that Switzerland’s ban on second homes substantially lowered investment in residential and commercial construction, and that Egypt’s large currency depreciation in 2016 affected consumer-good producers more than capital-goods makers.

Is agentic research any good? Dr Yanagizawa-Drott and his colleague, Olaf Willner, used an AI verifier to check 1,000 APE papers. None was error-free; 842 contained at least one severe error, including ones invalidating headline results. To see roughly how Ape stacks up against human primates, The Economist teamed up with Refine, the research-verification firm founded by Dr Golub. Refine chose 100 papers at random from those accepted for publication in the top five journals (including QJE and JPE ) in the first nine months of 2026. It then subjected them to thorough review using several frontier AI models plus a once-over for serious errors by a human reviewer. The exercise flagged 4,158 issues in the 100 papers. One in ten papers contained at least one grave error (other problems concerned interpretation or inconsistencies between tables and text).

The two exercises are not directly comparable, but human economists’ lower error rate might give them some hope. In ideas, too, people retain the edge. One experiment, by Alexis Akira Toda of Emory University, found that when considering papers on economic theory alone, AI lowers the cost of checking for a candidate flaw but “does not yet reliably substitute for the human judgment that locates the flaw in the first place”. Another, which compared APE papers with published papers found the difference is mostly in ideas, not execution.

In a further study, Michele Zampa of the Geneva Graduate Institute used an AI referee to grade 981 AI papers and 41 human ones on things like the novelty of research questions and acknowledgment of methodological limitations. AI did better in crowded fields with lots of existing research to digest. Humans performed at the same level regardless of the size of the literature.

A related worry is that AI produces plausible-looking results by brute force, slicing and dicing data in hundreds of ways until something statistically significant pops up. Such “p-hacking” can be contained by checking how sensitive results are to changes in data or methods. Nic Fishman of Harvard and Gabriel Sekeres of Cornell University calculate that the number of needed robustness checks would rise from 50 for a human author to an impractical 7,000 for agentic ones. No wonder the APE project has a warning label not to use its papers for evidence-based policymaking.

The machines are gaining ground, however. Benjamin Kohler of ETH Zürich, another Swiss university, and colleagues, let AI reproduce 48 social-science papers, using only the studies’ raw data and a description of the method. The discrepancies between the original papers and the AI versions were due in roughly equal measure to problems with the AI and human work. But the better the AI model, the more the balance tilted against humans.

Econ 10100111001111

As it gets less error-prone and more creative, AI will remove the drudgery from parsing historical data, make complex modelling as easy as vibe-coding, identify cherry-picked numbers and improve economics in countless other ways. Already it is more than a tool. “We need to think about AI as a co-author,” says Dani Rodrik of Harvard, though it will be a while before economics can dispense with human co-authors altogether. “There is still a large amount of uncertainty about the right models, how to extrapolate from the evidence and what the right research frame is.”

Keynes would agree. A master-economist, he noted, “must be mathematician, historian, statesman, philosopher…He must understand symbols and speak in words. He must contemplate the particular in terms of the general, and touch abstract and concrete in the same flight of thought.” It may be a while before AI meets Keynesian criteria. Then again, as Keynes also noted, not many humans meet them, either.

Subscribers to The Economist can sign up to our Opinion newsletter, which brings together the best of our leaders, columns, guest essays and reader correspondence.

All Access.
One Subscription.

Get 360° coverage—from daily headlines
to 100 year archives.

E-Paper
Full
Archives
Full Access to
HT App & Website
Games
 
Stay updated with the latest Technology News, gadget launches, app updates, artificial intelligence and digital trends. Find reviews, comparisons and useful insights from the world of tech.
Stay updated with the latest Technology News, gadget launches, app updates, artificial intelligence and digital trends. Find reviews, comparisons and useful insights from the world of tech.
SHARE THIS ARTICLE ON