Drawing causal inferences about the effect of one variable on another is something people do all the time. Machine learning (ML) and artificial intelligence (AI), in contrast, are only able to perform statistical correlations and pattern discovery. Umbrellas are associated with rain, for example, but carrying an umbrella does not make water droplets fall from the clouds. This is obvious to humans, but neither machine learning nor even classical statistics can even pose, let alone investigate, such basic assertions. One of AIХs most distinguished research groups is embarking on a mission to bring causal considerations like these to their work on ML. Founded almost 15 years ago by Geoff Hinton, the Learning in Machines and Brains program (LMB), organized and partially funded by the Canadian Institute for Advanced Research, has made causal inference a top priority for its next phase. Funds from this grant support efforts by LMB to hold a multidisciplinary conference on causal inference in AI, bringing together top experts in economics, econometrics, statistics, neuroeconomics, and logic, to discuss the challenges and opportunities in developing machine learning protocols and platforms that can detect causal relationships in data. Additional grant funds will support a series of six $50,000 ТcatalystУ grants that will spur innovation by funding targeted research projects on these and related issues.

Funds from this grant support work by a team, led by David Cousins at a company called Duality, to develop mathematical tools and techniques that will enable social scientists and other researchers to link and analyze privacy-sensitive financial data without the risk of exposing protected information in the data. Partnering with the Institute for Research on Innovation and Science (IRIS), which collects administrative records from universities with the expectation that no sensitive information about one institutionХs employees, students, or finances will be revealed to any another, Cousins and the Duality team will work on the development of methods that use a type of advanced cryptography known as fully homomorphic encryption (FHE) to develop analysis tools that are powerful and scalable enough to be used in a wide variety of research contexts, yet satisfy very stringent data privacy standards. All developed software and code will be open source and deposited in freely accessible software libraries.

The Summer Institutes in Computational Social Science (SICSS) are annual summer courses that aim to accelerate the growth and strength of computational social science; to seed interdisciplinary research that builds on this field; to create open source teaching materials that support training in the field; to ensure that the new field develops appropriate norms and standards that are in the long-term interests of science and society; and to create a community of scholarsСsupported by partner organizations such as companies and universitiesСthat will help advance computational social science into the mainstream of economics, data science, sociology, political science, and other social science fields.К Founded by Christopher Bail of Duke and Matt Salganik of Princeton, the courses are held each summer on the Duke and Princeton campuses, while simultaneously being broadcast to 11 different educational institutions around the world. Almost 300 graduate students, postdoctoral fellows, and junior faculty participated in 2019 alone, with applications far exceeding the spots available. Funds from this grant support the continued operation, expansion, and development of the SICSS for a period of three years.

This grant provides three years of continued operational support to the Household Finance Working Group (HFWG) at the National Bureau of Economic Research. The group brings together leading economists, econometricians, regulators, and policymakers working on issues broadly related to understanding how households make economic decisions. Specific topics addressed by the working group include household saving, portfolio behavior, borrowing decisions, investment choices, risk management, and bankruptcy.К Over the next three years, the HFWG will continue its work with a focus on how changes in technology are affecting both household decision-making and the opportunities to study it. Topics to be examined include the rise of fintech, big data, and machine learning in the provision of consumer financial services; the impact of electronic payment systems, such as cryptocurrencies, on traditional financial institutions; and both the challenges as well as the opportunities posed by the use of administrative data for research on consumers and their financial decisions.

The Тexternal validityУ of a scientific finding is its robustness in the face of additional observations or alternative model specifications. Statistically significant findings can often be weakened or reversed if either the same analysis had been done with a different sample or if a different model specification had been applied to the same data. Bayesian statistical techniques are particularly well suited to address such issues, but their uptake has been impeded by their awkward, difficult implementation in the standard statistical programs most commonly used by researchers. К This grant provides funds for the continued development and adoption of Stan, an open source, probabilistic programming language developed by Columbia University statistician Andrew Gelman. Stan elegantly implements advanced Bayesian methods for analyzing external validity and for many other issues and has gained increasing popularity in recent years. Grant funds will allow the continued growth of Stan with a specific focus on developments aimed at making the program more useful and useable by economists and other social scientists. Planned grant activities include the development of new modules specifically addressing the complex and multilevel systems that social scientists study, as well as the production of open-access tutorials, research papers, and reproducible case studies.

The Brookings Papers on Economic Activity (BPEA) synthesize and popularize the policy implications of cutting-edge research in economics. Founded almost 50 years ago, BPEA remains a highly respected and influential outlet for economic ideas, as BPEA articles are consistently reliable, readable, and nonpartisan. They are often cited both in the popular media and in official policy documents produced by Congress, the executive branch, and other institutions like the Federal Reserve, the World Bank, and the International Monetary Fund. Funds from this grant provide operational and administrative support for the continued production of the Brookings Papers on Economic Activity for a period of three years. Some of the topics that BPEA will focus on during this period include productivity and the economics of technological innovation, household decision-making, and data improvement.К