5

Focusing Warfare Research and Improving M&S

BACKGROUND

Given the need for research to improve the knowledge base on which M&S is based, how best might the Department of the Navy (and DOD) go about it? Although the issue is often posed as building better M&S, that is arguably an instance of the tail wagging the dog. Do we ask aeronautical engineers to build better models or to build better aircraft? Do we ask economists to build better models or to clarify important issues such as how to define a cost-of-living inflator? Why, then, do models and simulations have such a prominent place in current DOD thinking?

In fact, there are several reasons. First, M&S products (as distinct from constantly changing personal tools) are needed to achieve the objectives of distributed training, exercising, and planning. Second, M&S products are needed to achieve the improvements in effectiveness and efficiency associated with making reusable objects in a generally available repository. The third reason, however, is one mentioned early in the report: that many people, particularly managers and software technologists, think of models as commodities and do not worry particularly about where the knowledge comes from to support the models.¹

The point here is not to criticize that view, because it has its place. Indeed, the insights from software engineering and management have great value for military science in which models and simulations are important, which includes the study of most complex military operations and phenomena. However, there is a need to rebalance the situation by emphasizing that where the objective is to understand the nature of war and other military operations—especially the nature of future war in the information era—the focus should be on research rather than model building per se.

Research Versus Simulation Building

To appreciate the significance of research versus simulation building, consider what often happens when people define the objective as building an M&S. A committee goes about constructing wish lists, which are later translated into an expression of requirements. A request for proposals (RFP) is then issued or the tasking assigned to a government laboratory or federally funded research and development center (FFRDC). A contract is let and work proceeds. But the work is typically construed to be building software. The team might have, for example, a chief modeler/designer, a software designer, several modeler/programmers, and some specialists in graphics, databases, and operating systems. And, because the model will need data, there may be one or several individuals, as well as representatives from sponsors, actively involved in building databases (e.g., for orders of battle, temperature profiles in different portions of the ocean, weapon effectiveness, sortie rates, and so on).

Now, all of this may sound reasonable and industrious, but it is quite different from what would happen if the objective were seen as understanding the subject area, with a model as a possible by-product. In this case, the team might include scientists, engineers, operations researchers at least as interested in phenomenology and conceptual models as programming, historians, and psychologists (e.g., for interviewing experts)—as well as operations-experienced military officers. Results might include learned papers on various aspects of the phenomenology and other papers discussing future doctrinal options. Unfortunately, there might not be any products directly usable by builders of M&S. There might be no rigorous models at all, or they might not "fit" well into the larger scheme of things.²

Ultimately, what seems to be needed is a synthesis (Figure 5.1). There is a need for research, but that research could be accomplished with the recognition that it will be used to feed the building of M&S, and it could be accomplished with the same common model of the mission space (CMMS) as used by the M&S builders. Further, the M&S designers could base their designs on concepts emerging from the research rather than imposing their own concepts. The result would be an M&S better able to accommodate future research results as well, rather than M&S with data structures that do not relate well to experimental data or to changes of perspective.³

FIGURE 5.1 Synthesis of desires for research base, conceptual models, and effective M&S software.

Modularity of Knowledge

Another reason for emphasizing research rather than model building per se is that if one attempts to build a comprehensive model of complex systems, there is a good chance of failure: the computer model will eventually collapse under its own weight. By contrast, modular knowledge can endure. Often, specialists are needed to understand different aspects of system behavior (e.g., probabilistic issues versus the effects of saltwater on instrumentation versus the location errors associated with a less-than-complete GPS constellation).

PRIORITIZING WARFARE SUBJECTS FOR RESEARCH

Calling for an across-the-board program of research would be of little value. Further, the panel is recommending a significant change in the way business is done, which raises the barriers. With this in mind, the panel identifies a first set of warfare subjects for priority attention by the Navy and Marines. Success in these domains might lead to more general changes later.

In developing this priority set, the panel established several criteria:

• Military importance to the Navy or Marine Corps,
• Importance to higher-level joint operations,
• High potential payoff for empirical and theoretical research, and
• Need for interest in and oversight by "operators" and "warriors," rather than scientists alone.

With these criteria as background, Table 5.1 provides a possible first list of subjects. Each has major knowledge gaps that could be narrowed by empirical and theoretical research closely tied to the "warrior communities."

TABLE 5.1 Warfare Areas Needing Empirical and Theoretical Research

Warfare Area	Shortfalls in Knowledge of Phenomena	Importance	Potential Value of Focused Research	Comments
Joint task force operations with dispersed forces	•••	•••	•••	Issues of survivability and effectiveness (may need probabilistic depictions)
Effectiveness of long-range precision strike against armies taking countermeasures	••	•••	•••	Likely large differences among weapon enthusiasts, planners, and on-the-ground reality
Short-notice early-entry operations against opposition	•	••	•••	Short-notice planning and mission rehearsal
Theater-missile defense, including counterforce, and including speed-of-light weapon options	••	•••	•••	M&S will be only mechanism for evaluating effectiveness in large-scale battles
Expeditionary warfare and littoral operations	••	•••	•••	Problems with smart mines, opposition, missiles, and WMD

DESIRED ATTRIBUTES OF RESEARCH PROGRAMS

Although the research needed would obviously vary from subject area to subject area, the following features would seem to be strongly desirable in most cases. An overarching theme is the need to take a holistic approach rather than one based on either top-down or bottom-up theology. For each warfare area the panel recommends developing hierarchically integrated families of models with different characters and resolutions—not to predict detailed behaviors, but rather to explore and understand military phenomena. Such simulation-based exploration is a form of experimentation that can yield profound insights.

1. Top-down thinking from the perspective of a JTF commander. A common failure of warfare research, and of model building, is not representing from the outset some of the principal factors that would affect higher-level operational decisions. As a consequence, even detailed models often have an overly narrow scope. Further, there are forest-and-trees problems. This can be mitigated by defining the problem initially so as to include at least primitive representations of higher-level strategy and command-and-control.
2. Tactics development. Whenever one considers something new (e.g., a new operational concept or weapon system), it is important to think about the range of possible tactics—for both oneself and the adversary, perhaps through some cycles of measure and countermeasure. Historians repeatedly remind us that the principal changes wrought by previous revolutions in military affairs (RMAs) have been at least as much organizational and doctrinal as technical.
3. Realistic decision behaviors. In each domain, there is likely to be a need to build realistic decision models, for both friendly and adversary sides (and third parties as well), as well as models that represent important limiting cases such as optimization (feasible only with perfect information, but an important bound) and doctrinal behaviors. In addition, the panel believes it important—in both research and the subsequent M&S—to allow for human play. Even so-called "constructive models" for analysis should have human play options. Failure to allow for this will often guarantee slipping into a pattern of computer-comfortable but unimaginative and unrealistic behaviors.
4. Theory development for multi-resolution model families. This issue is discussed elsewhere in the report (Chapter 6 and Appendix E), but it should be a key element of research in most warfare areas.
5. Well-designed empirical work exploiting high-resolution simulation, DIS-modulated military exercises, and training experiences. It is generally agreed that higher-level M&S lacks an adequate empirical foundation, but it is often suggested that little can be done about the matter. This is no longer true. At least two developments have changed what is possible: (a) the advent of high-quality, high-resolution simulation with extraordinary computer power, and (b) the advent of distributed simulation, including distributed interactive simulation (DIS) as a core activity of training and exercising. With respect to the latter, it is now possible to collect operational data unobtrusively. Figure 5.2 sketches this notion (Davis, 1995b).
   
   
   FIGURE 5.2 Exploiting DIS experiments for empirical information. SOURCE: Reprinted, by permission, from Davis (1995b). Copyright 1995 by IEEE.
   
6. Stochastic and human-in-loop options at all levels of play. In many of the important warfare domains, decision makers—including commanders about to send their people into battle—need to understand the stochastic nature of events. Expected value models can be exceedingly misleading. Or, when their shortcomings are obvious, they can inappropriately discredit otherwise valuable simulation (e.g., as when a fractional carrier is still critical in an operation's success). The panel therefore believes it essential that research attend seriously to such matters. One aspect of this will, again, involve including humans in the loop for at least some of the activities.
7. Comprehensive models, including "soft" effects. If simulations are to be realistic, whether or not precise, it is essential that they reflect a vast range of soft factors that include, for example, random human errors, virtual attrition (as when pilots achieve poorer air-to-ground performance when flying in an intense air-defense environment, even if they take no attrition), "frictional" effects, and suppression of enemy effectiveness by information warfare or barrage bombing. While these are sometimes notoriously difficult to incorporate precisely, the principle to be kept in mind is that to omit them is to assume implicitly that they have no effect (i.e., to assume that model correction factors are all unity).
8. Complex MOEs. Although operations researchers and mathematicians often like to identify a single measure of effectiveness (MOE) on which to focus, there are relatively few circumstances in which that is appropriate for higher-level decision support. JTF commanders, for example, must worry not only about damage caused to opponent forces, but also about impacts on opponent effectiveness. They must worry even more about effects on the success of their own strategy. And they must typically worry about casualties—to their own forces, to allies, and even to enemy forces. Thus, research and analysis should in the panel's view increasingly provide a rich set of MOEs.
9. Empirical data development. Empirical work using DIS exercises is mentioned above, but much broader activities are possible. These include historical research, dispatching operations researchers to observe and report on operations and doctrinal experiments, structured interviewing of military experts—foreign as well as American—and field tests.
10. Models as products. In the past, military research has often not connected well with the needs of the M&S community. While it is evident that the panel does not recommend focusing scientists on the building of large-scale M&S, it does believe that more can be demanded of them in terms of expressing their conclusions in the form of discrete models (mathematical, logical, or computerized) that connect conceptually to the larger realm of M&S. Here the panel believes that it is important that military scientists inform and be informed by the emerging work on common models of the mission space (CMMS). As an example here, someone conducting research on command and control might need to see how his ideas could be modularized so as to be useful in a DOD-wide model repository constructed with the imagery of Figure 5.3 in mind.⁴

FIGURE 5.3 An illustrative slice of a CMMS for JTF strike operations. SOURCE:  Jefferson, DMSO (1996).

Footnotes

1. Also, many "modelers" are more focused on constructing interesting programs than on applications. This violates principles of what operations researchers are taught, but it is a sociological fact.

2. As an example of this, there have been a number of interesting historical studies on when and how battles are won and lost, but they have seldom related easily to the simulation models on which DOD depends. Incorporating their insights, much less their data, has been difficult.

3. This issue of databases not being easy to change, is connected to the difference between declarative and procedural knowledge (a distinction much discussed in the computer science and artificial intelligence literatures). Declarative knowledge often takes the form of relationships (conservation laws, Newton's laws, and the like), whereas procedural knowledge usually takes the form of a recipe-like method for solving a problem or operating some system (e.g., "After 6 p.m., turn on the lights"). In many respects, declarative knowledge is more powerful, because it can be used to address a wide range of situations. In contrast, procedural knowledge is often "brittle" (at some times of year, sunset may not be until 9 p.m. or so). When we speak of computers lacking common sense and humans being more adaptive, one of the underlying considerations is that computers are typically programmed to be extremely literal, while humans are able to draw on more general considerations to tailor actions to the task at hand.
    A quintessential example is the difference between mission orders and detailed instructions. A commander can specify objectives, describe issues and constraints, and then let his subordinates achieve those as proves feasible and appropriate. Alternatively, he can lay out a plan that "scripts" their activities. The former expresses the problem with declarative knowledge; the latter with procedural knowledge.

4. Adapted from a MORS briefing on CMMS issues sponsored by the Defense Modeling and Simulation Office (1996).