Dr. Nina Deliu, YoungStatS editorial board member, talked with Professor Susan Athey, Stanford University.

Susan Athey is The Economics of Technology Professor at Stanford Graduate School of Business. She is an elected member of the US National Academy of Sciences and serves as the 2023 President of the American Economic Association. She is the recipient of the John Bates Clark Medal, awarded by the American Economic Association to the economist under 40 who has made the greatest contributions to economic thought and knowledge. Her research spans the areas of causal inference, econometrics, statistics, and machine learning.

Nina: The breadth and depth of your education and work is astonishing. Which are the pros and cons of having such a hybrid background, and which is the discipline you mostly feel you belong to?

Susan: I do feel that my training as an economist is most central to what I do. It is the discipline that grounds the questions I ask and how I think about what is important. I find that working across disciplines forces me to understand things more deeply. And sometimes that leads you to question assumptions or find a different way of looking at things. An exciting part of changing areas is that it always feels new, and I’m always learning; but I do think it is important to spend long enough in an area to understand it deeply. Finding fantastic coauthors who are experts in a new area is also really helpful for moving into that area, especially when each author brings important problems and expertise to the table. I’ve been very fortunate to work with people like Stefan Wager from statistics and David Blei from computer science and statistics, and I have learned a lot from students.

Nina: In your career, you have collaborated with many private and public technology firms. What do you think academics can do to get engaged with companies in a way that is mutually beneficial?

Susan: Working with companies has been extremely important for me in terms of getting exposed to technology and to new problems, and for teaching me how to build things and generally how to get more complicated projects done. Companies give you the opportunity for large-scale impact, and my recent collaborations focused on social impact applications. However, it can be challenging to balance conflicts of interest and confidentiality; I chose projects where the companies were aligned in my interest in publication. My lab at Stanford has built collaborations through which we built impactful products, introduced social impact companies to new methods, and produced high quality academic research, all while providing valuable training to students.

Nina: Jointly with your husband Guido W. Imbens you have been a forerunner in the inclusion of artificial intelligence (AI) in the economic field. Would the future mean higher and more frequent inclusion of machine learning in statistical and econometric models?

Susan: The intersection of statistics, econometrics and machine learning has come a long way since I started working on it in the late 2000s. Despite initial skepticism, now, it is fairly well understood that machine learning methods can be extremely effective in combination with established tools for causal inference and economic modeling of human and firm behavior. In terms of what is next, I am excited for the economics profession to find new uses for “foundation models” from AI, that is, models that provide a low-dimensional representation of high-dimensional data. There is also a great potential for new interventions that can be designed using AI, whose benefits could be measured in randomized experiments.

Nina: What do you think about the future of statistics and its interplay in data science, machine learning and artificial intelligence?

Susan: Recently the fields of machine learning and AI have been moving so fast that there is a lot of pressure for young people to focus only on the latest developments. However, I think something is lost if students don’t get a solid foundation in statistics, and don’t build intuition on how and why models work. My guess is that the pendulum will swing back towards statistical understanding as we attempt to analyze the performance and understand the weaknesses of new AI algorithms.

Nina: What do you think is an under-appreciated area within statistics and data science? On the other hand, what is overemphasized?

Susan: It used to be that causal inference was under-
appreciated, but it seems that recently it is much better appreciated! I think that in the future, so much of our life and economy will involve digital interactions, that there will be an even larger role for designing experiments that can be used both to gain generalizable knowledge and to guide the path for incremental innovation and personalization. I think there is a lot more to do in designing complex experiments, which may consider interactive environments or bandits for adaptive experiments, among others.

Nina: What qualities do you most value in students and young researchers? What advice would you share with them?

Susan: There are a lot of ways to make a contribution: more theoretically oriented students can share insights, while students with implementation skills find creative ways to get things done. My own lab is interdisciplinary, and I love that people have different career goals and objectives. I especially value people who can think conceptually and reason about how and why things work. I like working with people who are can-do types, who are problem solvers. In terms of advice, learning to write precisely and unambiguously is always helpful. A lot of my students are in between fields and are trying to decide which path to follow. For them, I suggest to find a community that shares your values, so that your peers will appreciate what you are contributing.