「经济学人」Sine laude

Aidan Toner-Rodgers was enjoying a meteoric rise at MIT. Then questions started to be asked about his work

When the economics department at the Massachusetts Institute of Technology issues a statement, it is often to celebrate a Nobel Prize. In the past decade, six of its professors have won the award—as many as the next two universities combined. But on May 16th it issued a different sort of press release: one disavowing research by a high-flying graduate student.

Aidan Toner-Rodgers was the author of the paper in question. It assessed the use of an AI tool by an unnamed materials-science firm. Even for techno-optimists, the results were striking: “AI-assisted researchers discover 44% more materials, resulting in a 39% increase in patent filings.” They were widely reported, including by The Economist. The work was praised lavishly by Daron Acemoglu and David Autor, two of MIT’s leading economists.

MIT now declares “no confidence in the provenance, reliability or validity of the data and...in the veracity of the research”. Mr Toner-Rodgers’s paper has been withdrawn from the pre-print repository on which it first appeared; his personal website has been taken down. The lab at the heart of his findings remains unknown.

Academic misconduct often triggers a reckoning. In 2015 political scientists grappled with the retraction of a well-publicised article that claimed door-to-door canvassers could lift support for gay marriage. More recently, behavioural science has come under scrutiny: Francesca Gino of Harvard University and Dan Ariely of Duke University have faced investigation over allegedly manipulated data (both deny wrongdoing). Economics has some protection owing to its record. The five leading journals have seen just four withdrawals in their combined 570-year history, according to Retraction Watch, a database.

But even if economics is not the worst offender, it is no stranger to social science’s replication crisis. Its biggest recent trend has been empirical research, with a focus on credible causal designs (see chart). Statistically significant results are prized, incentivising cherry-picking and selective presentation of results. Prashant Garg of Imperial College London and Thiemo Fetzer of the University of Warwick find that the share of papers reporting “null results” fell from 15% in 1980 to 9% in 2023. Use of private data doubled.

More than in other disciplines, success depends on a few high-stakes events. Job-market candidates are evaluated on a single paper, rather than a body of work. Because institutional pedigree and advisers carry lots of weight, young researchers may face pressure to overstate results.

Bad economics research has real-world consequences. Mr Toner-Rodgers’s paper was cited by the European Central Bank and in Congress. It surely led more than one research-and-development lab to consider its internal processes. Other retracted papers advised against large debt-to-GDP ratios and on how to price corporate bonds. All academic red herrings matter. But they matter more when they have implications for national budgets, financial markets and, indeed, the future of AI.

This article appeared in the Finance & economics section of the print edition under the headline “Sine laude”(May 22nd 2025)

我没有译文，仅供拓展阅读

LearnAndRecord

2015年2月8日

2025年5月30日

第3765天

每天持续行动学外语