Szymon Sacher

This paper presents a framework for assessing the economic consequences of AI between 2026 and 2030. In the model, AI automates a growing share of cognitive work, raising productivity and displacing workers who must search for jobs in other occupations. The model maps future paths of AI capabilities into implied paths for GDP, the labor share, wages, labor reallocation, and unemployment. We illustrate the framework by considering three scenarios: modest, substantial, and extreme. Under modest change, AI adds less than half a point to GDP growth by 2030 and raises unemployment by a tenth of a point. In the extreme change scenario, AI has transformative effects, with AI performing almost half of today’s cognitive work by 2030. GDP growth then rises to 15 percent per year, the labor share of income falls from 60 to 45 percent, and nearly one in five cognitive workers is unemployed. We also surveyed US adults about their expectations for AI. Views vary widely, but the median respondent’s answers are consistent with our substantial change scenario in which, by 2030, GDP rises by 8 percent and cognitive employment declines by 4 percent. The model offers a structured way to compare possibilities for our economic future under different expectations about AI.

Researchers now routinely use AI or other machine learning methods to estimate latent variables of economic interest, then plug-in the estimates as covariates in a regression. We show both theoretically and empirically that naively treating AI/ML-generated variables as “data” leads to biased estimates and invalid inference. To restore valid inference, we propose two methods: (1) an explicit bias correction with bias-corrected confidence intervals, and (2) joint estimation of the regression parameters and latent variables. We illustrate these ideas through applications involving label imputation, dimensionality reduction, and index construction via classification and aggregation.

This paper extends the AKM framework to account for endogenous worker selection into firms and locations using a structural discrete-choice model estimated via MCMC. The model is applied to matched employer-employee data from Brazil (RAIS) to study wage determination and worker mobility across labor markets.

This project designed and delivered a digital media literacy intervention to 50,000+ participants. We measured the impact on knowledge using surveys and on behavior using observational data from Meta’s platforms.

We use variational inference (VI), a technique from the machine learning literature, to estimate a mortality-based Bayesian model of nursing home quality accounting for selection. We demonstrate how one can use VI to quickly and flexibly estimate a high-dimensional economic model with large datasets. Using our facility quality estimates, we examine the correlates of quality and find that public report cards have near-zero correlation. We then show that in contrast to prior literature, higher quality nursing homes fared better during the pandemic: a one standard deviation increase in quality corresponds to 2.5% fewer Covid-19 cases

To what extent do mass media outlets influence viewers’ trust in scientific evidence and compliance with behavior recommended by scientific experts? Exploiting the U.S. lockdown period of the COVID-19 pandemic in early 2020, we analyze a large longitudinal database that combines daily stay-at-home behavior from approximately 8 million mobile phones and local viewership of cable news networks. Early in the pandemic, several of Fox News’ hosts downplayed the severity of the pandemic and the risks associated with the transmission of the virus. A combination of regression analysis and a natural experiment finds that a 10% increase in viewership of Fox News in a zip code causes a 0.76-percentage-point reduction in compliance with stay-at-home behavior. The results imply a media persuasion rate that is larger than typical advertising persuasion rates on consumer behavior. Similar analyses using viewership of MSNBC and CNN, which supported lockdown measures, were inconclusive but suggested a smaller, positive effect on compliance with social distancing regulations.