Revisiting Pearl Harbor with Probabilistic Early Warning

Abstract
I will begin this talk with a short discussion of the function of warning in the US national security community, and the analytic methodology used by US intelligence agencies (in 1941 and since) to address the problem of warning. I will then present a formal model for crisis warning consisting of a Partially Observable Markov Decision Process (POMDP) intended to assist an intelligence analyst in deciding when to issue an alert to a foreign policy principal decision maker such as the President. The lead time demanded by the principal is a key element in the model. I will spend the remainder of the talk illustrating this warning model in the context of the brewing crisis in the Pacific from July to December 1941, and present results from test runs of the model using historical raw intelligence data from that period. While a probabilistic approach to warning is not a new idea, this research addresses three outstanding issues left unresolved from past efforts to develop such an approach:

1. The need to process multiple dependent signals in a manner that is combinatorially feasible;
2. Incorporation of the time dimension in which intelligence data is received into the inference, and the effect of dynamics on a warning decision where a finite horizon is imposed;
3. Consideration of the fact that the analyst serves as an advisor to the principal decision maker but is not completely aware of the principal’s preference set.

Together with my thesis advisor, Prof Elisabeth Pate-Cornell, I am currently writing a paper that covers the presented material, and I hope to incorporate feedback from this presentation into the paper. Because the paper is currently a work in progress, I am not distributing it at this time.

David Blum attends Stanford University, where he is a 3rd year Ph.D. student in the Department of Management Science & Engineering as well as a U.S. Department of Defense SMART Scholar. He is currently developing a probabilistic model of national security crises, with the goal of improving crisis early warning. His interests also include targeting in counter-terrorism, signatures of WMD proliferation, and models of decisions made by adversarial actors as games with incomplete information. He is a graduate intern in the Counter-Proliferation Operations-Intelligence Support program at Lawrence Livermore National Laboratory.

Between 2004 and 2008 David worked at the U.S. Department of Defense as an operations research analyst. He deployed twice to Iraq, in 2007 and 2008, where,  as member of Multi-National Corps Iraq, he provided direct analytic support to conventional and special operations units. He received his Master's degree from MIT in political science, concentrating in security studies, and his Bachelor's degree from Columbia University in history and physics.