Adversarial Risk Analysis

Sun Tzu's The Art of War exhorts the aspiring warrior to 'know your enemy,' but the mathematics of conflict (whether von Neumann-Morgenstern game theory or Bayesian decision analysis) takes very little account of the complexion of the adversary. Indeed, when Morgenstern, reading Conan Doyle, saw paradox from the infinite regress implied by the infinitely rational agents Holmes and Moriarty attempting to outthink each other, von Neumann introduced randomization precisely to excise from such analysis any need to model such outthinking. Both game theory and decision analysis offer 'optimal' strategies based on the situation alone, the two methods diverging as they optimize for different quantities. But in real-world situations in which one faces less-than-perfectly-rational adversaries, outthinking can confer an advantage. Here, Banks (Duke Univ.), Rios (IBM), and Insua (ICMAT-CSIC, Spain) identify three categories of uncertainty for the strategist: aleatory uncertainty-nondeterminism of outcomes even after players make choices; epistemic uncertainty-hidden information concerning opponents' preferences, beliefs, and capabilities; and concept uncertainty-hidden information concerning opponents' strategies. Adversarial risk analysis, a new field with roots in modern efforts to defeat terrorism, provides a framework, in principle, to cope with these uncertainties. Solving the models seems generally intractable, but the heart of the book, the first of its kind, offers exemplary case studies. Summing up: Recommended. Lower-division undergraduates and above; informed general audiences. -D. V. Feldman, University of New Hampshire, Durham, USA, for CHOICE, March 2016