From: Robert Provine
Date: 9.25.01

Decision Theory And National Security

Decision theory offers guidance for national security policy and associated civil liberty issues introduced by the tragic events of September 11. Signal Detection Theory, the state-of-the art procedure for decision making considered here, provides a powerful model for detecting the presence of a "signal," whether a sensory stimulus in the laboratory, a bomb, or a terrorist. Although the underlying mathematical model is complex, predictions from the model are straightforward and will be explored in a few of the many possible security-related applications.

In detection tasks involving yes/no decisions, there are two ways to be right and two ways to be wrong. In searching for a bomb among airline luggage, for example, a security person responding "yes, I've detected a bomb," can result either in a "hit" (the correct detection of a bomb), or a "false alarm" (saying "yes" when no bomb is present). As with "yes," the decision of "no" also carries dual consequences, one right and one wrong. An inspector deciding "no" can either correctly deny the presence of a bomb (a "correct rejection"), or produce the dreaded "miss," the failure to detect a bomb when one is present.

Enough about theory. What about security policy? The decisions are not as straight-forward as policy makers may like. We must confront a pesky problem — there is no single best decision, because each alternative has linked costs and benefits that cannot be finessed. For example, the only way to detect more bombs (increase "hits" and reduce "misses") is to lower our criterion and say "yes" more often, a result that also increases false alarms. The only means of detecting all bombs is to always say "yes, a bomb is present," and act accordingly. Obviously, we must balance this impractical standard against the expense and inconvenience to travelers and airlines, and the reduced attentiveness and credibility of security personnel who would almost always be crying wolf. But instructing inspectors to respond to suspected bombs only when they are certain of their judgement is no solution — fewer "yes" responses serves only to lower the proportion of hits and increase the proportion of misses.

One means of increasing bomb detection rates by inspectors would be to increase the number of bombs to detect. Another would be to provide a reward (cash, promotion, etc.) for successful bomb detection. Both approaches produce a bias for saying "yes" and a higher rate of bomb detection. The virtue of the reward procedure is obvious, but what of the dubious procedure of increasing bombs? The best procedure would be for roving security inspection teams to plant fake bombs in luggage. Security personnel will be more attentive if they are aware that test bombs will be present to detect, rather than the present situation in which their career will probably pass without a single "hit." Currently, variants of the fake bomb technique are used only for unsystematic exposes of flaws in the security system, not to improve inspector performance.

Security decisions become more controversial when we shift our attention from bombs to the people who may plant them. Fortunately, decision theory cuts through political hyperbole and clarifies the conflicting demands of security and civil liberty. As in the task of bomb detection, the only way of increasing "hits" (e.g., terrorist detection) is to lower the criterion for saying "yes," an act that necessarily yields more false accusations. No method of improving detection rates magically escapes the costs of more false alarms.

"Discrimination learning" is another area of behavioral research that brings insight to detection tasks but forces tough decisions. Through trial-and error, humans and other animals learn cues relevant to discovering stimuli, whether a bird developing a "search image" for a caterpillar hidden on a leaf, or the police generating a "profile" of a likely suspect. Focusing on relevant cues increases the efficiency of the search, but is the basis of "profiling," a potential threat to the civil liberty of targeted groups. However, there is no way of implementing an efficient search strategy that considers all suspects in proportion to their number in the general population, a tactic suggested by opponents of profiling. In the recent crisis, evidence suggests that airline security is more at risk from Muslims from the Middle East than from Episcopalians from the Midwest and suspect profiles should be weighted accordingly.

Approaches to stimulus detection and decision making are well understood and grounded in value-free theoretical and empirical research. The challenge is to balance the social costs and benefits of various options when decision protocols become policy.

John Brockman, Editor and Publisher
contact: [email protected]
Copyright © 2001 by
Edge Foundation, Inc
All Rights Reserved.