Paul K. Davis

Families of models that cross levels of resolution: issues for design, calibration and management

This review paper summarizes an ARPA-sponsored project to study variable-resolution modeling (VRM) and the connection of models across levels of resolution. We describe work introducing basic concepts, highlighting a design approach called integrated hierarchical variable-resolution modeling (IHVR), exploring mathematically some long-standing issues of aggregation in ground-combat modeling, and experimenting with cross-organizational efforts to develop and compare models created with different techniques and perspectives. We also describe highlights of a conference held in May, 1992 and touch briefly on some subsequent work. Despite the substantial progress, we conclude that the VRM problem is central, difficult, and greatly under invested. It should be approached as a matter of military science rather than technology, although we discuss new software tools that can be quite valuable.

Exploratory analysis enabled by multiresolution, multiperspective modeling

The objective of exploratory analysis is to gain a broad understanding of a problem domain before going into details for particular cases. Its focus is on understanding comprehensively the consequences of uncertainty, which requires a good deal more than normal sensitivity analysis. Such analysis is facilitated by multiresolution, multiperspective modeling (MRMPM) structures that are becoming increasingly practical. A knowledge of related design principles can help build interfaces to more normal legacy models, which can also be used for exploration.

Strategic Planning Amidst Massive Uncertainty in Complex Adaptive Systems: the Case of Defense Planning

In this paper I describe certain core problems of defense planning (Section 2), which have much in common with more general planning problems that arise in dealing with complex adaptive systems (CAS) characterized by unpredictability. I then describe concepts and methods that my colleagues and I have brought to bear with not-inconsiderable success. The concepts relate to managing risk and uncertainty (Section 3). These involve portfolio management and an emphasis on building blocks and at-the-time assembly of building blocks. The methods use measures of effectiveness focused on achieving flexibility, robustness, and adaptiveness. They involve exploratory analysis in large scenario spaces, and multi resolution modeling to facilitate such analysis. Such models often need adaptive agents. Human gaming, as part of a family-of-models approach, can also be helpful and even essential.

Defense planning when major changes are needed

ABSTRACT The principles and formalities of modern U.S. Defence planning stem from the 1960s and have largely served well. This paper, however, is about the special challenges that arise when major changes have been needed, some even transformational in character. It discusses how changing realities, independent studies and analysis, events, leaders, and political processes have led to changes not easily instigated within normal processes. Several examples are discussed for the period 1976–2016. Today, the United States and allies again face major challenges that require major military changes. Those have not yet been decided, much less accomplished. The paper draws on lessons from earlier periods to identify obstacles to and mechanisms for change. The last section focuses on defence analysis, which has sometimes been an obstacle but can be part of the solution. The paper urges a new ethic for analysis and the analysts who perform it.

Military applications of simulation

This chapter provides a selective overview of simulation activities taking place within the U.S. defense community. It touches upon work with virtual reality, entity-level simulation for analysis, highly distributed simulation in exercises and experimentation, and low-resolution exploratory analysis for higher-level force planning.

Multiresolution multiperspective modeling (MRMPM) as an enabler of exploratory analysis

Exploratory analysis examines the consequences of uncertainty--not merely by standard sensitivity methods, but more comprehensively. It is particularly useful for gaining a broad understanding of a problem domain before dipping into details. Although exploratory analysis can be accomplished with models of many types, it is facilitated by multiresolution, multiperspective modeling (MRMPM) structures. Moreover, a knowledge of related design principles facilitates the characterization of more normal models in terms that permit exploratory analysis. This paper describes the connections and notes that, with current and emerging personal computer tools, MRMPM methods are becoming practical.

Representing qualitative social science in computational models to aid reasoning under uncertainty: National security examples

Representing causal social science knowledge in models is difficult: much of the best knowledge is qualitative and ambiguously conditional, unlike the knowledge in “physics models.” This paper describes a stream of RAND research that began with qualitative models providing a structured depiction of casual factors creating effects. That has subsequently been extended to an unusual kind of uncertainty sensitive computational modeling that enables exploratory reasoning and analysis. We illustrate the approach with applications to counterterrorism, detection of terrorists, and nuclear crises. We believe that the approach will complement other approaches that can reflect social science phenomena [see other papers in this special issue of JDMS] and that the approach has broad potential within and beyond the national security domain. We also believe that it has the potential to inform empirical work—encouraging a transition from the step-by-step empirical testing of simple discrete hypotheses to the testing and refinement of more comprehensive causal models.

Using causal models in heterogeneous information fusion to detect terrorists

We describe basic research that uses a causal, uncertainty-sensitive computational model rooted in qualitative social science to fuse disparate pieces of threat information. It is a cognitive model going beyond rational-actor methods. Having such a model has proven useful when information is uncertain, fragmentary, indirect, soft, conflicting, and even deceptive. Inferences from fusion must then account for uncertainties about the model, the credibility of information, and the fusion methods - i.e. we must consider both structural and parametric uncertainties, including uncertainties about the uncertainties. We use a novel combination of (1) probabilistic and parametric methods, (2) alternative models and model structures, and (3) alternative fusion methods that include nonlinear algebraic combination, variants of Bayesian inference, and a new entropy-maximizing approach. Initial results are encouraging and suggest that such an analytically flexible and model-based approach to fusion can simultaneously enrich thinking, enhance threat detection, and reduce harmful false alarms.

Developing improved metamodels by combining phenomenological reasoning with statistical methods

A metamodel is relatively small, simple model that approximates the behavior of a large, complex model. A common and superficially attractive way to develop a metamodel is to generate data from a number of large-model runs and to then use off-the-shelf statistical methods without attempting to understand the models internal workings. This paper describes research illuminating why it is important and fruitful, in some problems, to improve the quality of such metamodels by using various types of phenomenological knowledge. The benefits are sometimes mathematically subtle, but strategically important, as when one is dealing with a system that could fail if any of several critical components fail. Naive metamodels may fail to reflect the individual criticality of such components and may therefore be quite misleading if used for policy analysis. Na*ve metamodeling may also give very misleading results on the relative importance of inputs, thereby skewing resource-allocation decisions. By inserting an appropriate dose of theory, however, such problems can be greatly mitigated. Our work is intended to be a contribution to the emerging understanding of multiresolution, multiperspective modeling (MRMPM), as well as a contribution to interdisciplinary work combining virtues of statistical methodology with virtues of more theory-based work. Although the analysis we present is based on a particular experiment with a particular large and complex model, we believe that the insights are more general.

Simple Culture-Informed Cognitive Models of the Adversary

Simple cognitive models of the adversary are useful in a variety of domains, including national security analysis. Having alternative models can temper the tendency to base strategy on the best-estimate understanding of the adversary, and can encourage building a strategy that is better hedged and more adaptive. Best estimates of adversary thinking have often been wrong historically. Good cognitive models must avoid mirror-imaging, which implies recognizing ways in which the adversary’s reasoning may be affected by history, culture, personalities, and imperfect information, as well as by objective circumstances. This paper describes a series of research efforts over three decades to build such cognitive models, some as complex computer programs and some exceptionally simple. These have been used to represent Cold-War Soviet leaders, Saddam Hussein, Kim Jong Il, and modern-day leaders of al Qaeda. Building such models has been a mixture of art and science, but has yielded useful insights, including insights about the sometimes-subtle influence of leaders’ decision-making culture.