Steven W. Popper

A General, Analytic Method for Generating Robust Strategies and Narrative Scenarios

Robustness is a key criterion for evaluating alternative decisions under conditions of deep uncertainty. However, no systematic, general approach exists for finding robust strategies using the broad range of models and data often available to decision makers. This study demonstrates robust decision making (RDM), an analytic method that helps design robust strategies through an iterative process that first suggests candidate robust strategies, identifies clusters of future states of the world to which they are vulnerable, and then evaluates the trade-offs in hedging against these vulnerabilities. This approach can help decision makers design robust strategies while also systematically generating clusters of key futures interpretable as narrative scenarios. Our study demonstrates the approach by identifying robust, adaptive, near-term pollution-control strategies to help ensure economic growth and environmental quality throughout the 21st century.

Confronting surprise

Surprise takes many forms, all tending to disrupt plans and planning systems. Reliance by decision makers on formal analytic methodologies can increase susceptibility to surprise as such methods commonly use available information to develop single-point forecasts or probability distributions of future events. In doing so, traditional analyses divert attention from information potentially important to understanding and planning for effects of surprise. The authors propose employing computer-assisted reasoning methods in conjunction with simulation models to create large ensembles of plausible future scenarios. This framework supports a robust adaptive planning (RAP) approach to reasoning under the conditions of complexity and deep uncertainty that normally defeat analytic approaches. The authors demonstrate, using the example of planning for long-term global sustainability, how RAP methods may offer greater insight into the vulnerabilities inherent in several types of surprises and enhance decision makers’ ability to construct strategies that will mitigate or minimize the effects of surprise.

Making computational social science effective: epistemology, methodology, and technology

There has been significant recent interest in Agent Based Modeling in many social sciences including economics, sociology, anthropology, political science, and game theory. This article describes three problems that need to be addressed in order for such models to become effective tools for formulating new social theory and informing policy debates and suggests approaches to meeting them. These issues are computational epistemology, research methodology, and software technology. These innovations augment Agent Based Modeling to create an effective new tool base to help better understand complex social systems.

Computer-assisted reasoning

Over the past few years, a novel approach to understanding complex and uncertain problems has emerged. The central insight is to conceive of any model run on a computer as a computational experiment. Instead of constructing and running only the single model that we believe best represents the system in question (after making various assumptions and a priori decisions), we can examine large numbers of models that depict alternative plausible future states of the system. This ensemble of plausible models can provide information not captured by any single best-estimate model. Furthermore, working with such an ensemble enables methodological approaches leading to more powerful and appropriate means than have heretofore been available for reasoning about these problems. By looking in many mirrors, each necessarily flawed (albeit in different ways), we can see truths that no single mirror can reveal. The authors show how they have implemented this approach, giving examples of problems where it has been fruitful.

Supporting decisions with (less than perfect) social science models

It is increasingly appreciated that models of combat and social-political behavior can be informative, helping us to anticipate possible future developments and the possible implications of contemplated actions. However, these models often cannot be relied upon to make predictions as accurate as those possible with models that have been developed in engineering and the physical sciences. Consequently, different standards for quality and methods for developing and exploiting these models are needed. Failure to understand this has resulted in general misuse of anticipatory modeling, either in treating anticipatory models as predictive, or by relegating such models to weak uses such as hypothesis generation. For many decision support applications, much of our knowledge is best expressed as computer models, and so it is important to develop valid means for aggressively exploiting anticipatory models. This paper will describe methods and technology developed at Evolving Logic for exploiting models that contain class knowledge, but do not accurately predict future events. Particular emphasis will be given to recent work which demonstrated the feasibility of developing robust decision options based on anticipatory social science models