- "Geopolitical Competition and AI in the Military"
(with Minsoo Kang, Noam Yuchtman, and David Y. Yang).
[Abstract]
Artificial intelligence (AI) has become both a general-purpose economic technology and an increasingly salient
military capability, while geopolitical rivalry has intensified across trade, standards, and security. This paper connects these shifts by tracking
when competition-centered AI narratives arose and evaluating whether military adoption followed. We construct a newspaper-based measure of geopolitical
competition in AI from New York Times articles (1996–present) using large language model classifiers, then measure U.S. government AI adoption
through federal procurement data (2018–2024), distinguishing defense from non-defense AI awards. We show that competitive AI narratives
surge beginning in 2016 and are dominated by U.S.–China rivalry. Over the same period, Defense Department AI procurement grows
disproportionately relative to non-defense AI, reshaping which firms and places receive defense contracts and deepening supplier relationships.
The findings highlight how narratives may legitimize spending and potentially redirect innovation. These patterns also suggest that U.S. allies
face meaningful choices about their AI strategies, both as general innovation policies and also as military decisions,
as U.S. defense AI procurement continues to expand.
- "Dialing Up the Empathy: Using AI Chatbots to Conduct Qualitative Interviews in Mass Surveys"
(with Michael M. Bechtel, Aaron Cannon, and Michael Hess).
[Preregistration] [Abstract]
The advent of artificial intelligence (AI) enables researchers to qualitatively
study preference formation through unstructured question formats that were previously too costly for inclusion in large-scale
surveys. We evaluate whether AI interviewers, through their empathic capabilities, can elicit richer survey responses and
reduce social desirability bias. Our experimental design randomly assigns respondents to one of three interview modes:
(i) self-administered, open-ended questions, (ii) a non-empathic chatbot interviewer, or (iii) a highly empathic chatbot interviewer.
We then examine response behavior across sensitive and non-sensitive topics, including vote intentions, abortion, hiring discrimination,
and favorite candy. The findings reveal whether AI chatbots can qualitatively explore preference determinants while reducing social
desirability bias compared to traditional interviews.
- "Preferences for Democratic Representation in Eight Democracies"
(with Michael M. Bechtel,
Simon Lüchinger, and
Lukas Schmid).
[Preregistration] [Abstract]
What preferences do voters hold about descriptive representation?
We distinguish between self-centered, proportionality-based, and other-regarding representation preferences and explore their
prevalence using survey experiments fielded in eight democracies (Australia, Canada, Mexico, Spain, Switzerland, Portugal,
Turkey, and the United States). Our experiments test how citizens (N = 13,856) respond to varying levels of legislative misrepresentation
across ideological, geographic, and gender dimensions. We find that citizens favor over-representation of their own ideological
group and region, reflecting self-centered preferences. In contrast, both women and men support female over-representation.
These patterns appear rooted in how citizens link descriptive representation and key elements of substantive representation
such as policy responsiveness and outputs. Our findings shed light on how misrepresentation of political and social groups in
parliament affects democratic attitudes.
