Search

The 2019 AI Expert Survey seeks to survey some of the world’s top AI researchers to forecast AI progress and study expert opinion on governance issues related to this technology. Working papers from this survey are in the “Working Paper Drafts” folder.

This project attempts to estimate the causal effect of data-sharing on citations. There is a fair amount of evidence that published academic papers that make their data publicly available have, on average, a higher number of citations, but ours is the first evidence that attempts to address the causal nature of this relationship.

A survey of AI experts on the progress of AI technology

Improving the design of survey experiments

This report presents a broad look at the American public’s attitudes toward AI and AI governance, based on findings from a nationally representative survey of 2,000 American adults.

Supplemental Material, sj-pdf-2-jcr-10.1177_0022002721995549 for Reputations for Resolve and Higher-Order Beliefs in Crisis Bargaining by Allan Dafoe, Remco Zwetsloot and Matthew Cebul in Journal of Conflict Resolution

Supplemental Material, sj-pdf-1-jcr-10.1177_0022002721995549 for Reputations for Resolve and Higher-Order Beliefs in Crisis Bargaining by Allan Dafoe, Remco Zwetsloot and Matthew Cebul in Journal of Conflict Resolution

Leaders throughout history have been concerned--often obsessed--with how other leaders perceive them. Historians have argued that many wars have been fought for purposes of reputation, honor, status, and prestige. However, there is little systematic study of these phenomena, and specifically of the effects of concern for reputation for resolve on interstate conflict behavior. After precisely defining resolve, and reputation for resolve, this dissertation examines this question by developing a family of formal models of escalation and reputation-engagement. From these models I explicitly deduce four testable implications of variation in concern for reputation for resolve that take selection effects into account and are robust to a variety of assumptions.To test these implications I search for research designs where concern for reputation is manipulated in a manner that is reasonably well-understood, and in which large unknown biases are unlikely. I find two such designs. The first, analyzed in collaboration with Devin Caughey, compares U.S. conflict behavior depending on whether the president is from the U.S. South. The U.S. South has a "culture of honor" that places greater importance on an individual's reputation for resolve. Using permutation inference and NPC, a technique new to political science, to provide a joint statistical test of the predictions, we find that conflict behavior under Southern presidents is substantially and significantly different from non-Southern presidents in a manner predicted by my theory. Furthermore, this difference remains significant under a large number of matched comparisons. This result is unlikely to be spurious because of its robustness to conditioning and because of theoretical and empirical inconsistencies in the alternative accounts.The second research design compares the conflict behavior of leaders early in their time- in-office with those same leaders later in their time-in-office. Leaders at the start of their tenure, as compared with themselves later in their career, should care more about their reputations because they have less developed reputations and longer time horizons. Each prediction of my theory is examined and finds statistically significant support with relatively large effect sizes. Alternative potential explanations are systematically ruled out through critical tests. In summary, the results from a set of clear tests of the effect of concern for reputation are consistent with the strongest claims made in the literature: concern for reputation seems to be an extremely important cause of war.

The Transparency and Openness Promotion (TOP) Committee met in November 2014 to address one important element of the incentive systems - journals’ procedures and policies for publication. The outcome of the effort is the TOP Guidelines. There are eight standards in the TOP guidelines; each move scientific communication toward greater openness. These standards are modular, facilitating adoption in whole or in part. However, they also complement each other, in that commitment to one standard may facilitate adoption of others. Moreover, the guidelines are sensitive to barriers to openness by articulating, for example, a process for exceptions to sharing because of ethical issues, intellectual property concerns, or availability of necessary resources.