The Panel on Modeling Human Behavior and Command Decision Making: Representations for Military Simulations was formed by the National Research Council in response to a request from the Defense Modeling and Simulation Office (DMSO). The charge to the panel was to review the state of the art in human behavior representation as applied to military simulations, with emphasis on the challenging areas of cognitive, team, and organizational behavior.
This report represents the findings of an 18-month study in which the panel, working within the context of the requirements established for military simulations, reviewed and assessed the processes and effects of human behavior at the individual, unit, and command levels to determine what is required to move the application of these kinds of models from their current, limited state to the inclusion of realistic human and organizational behavior. Based on the results of these efforts, the panel is convinced that (1) human behavior representation is essential to successful applications in both wargaming and distributed interactive simulation; (2) current models of human behavior can be improved by transferring what is already known in the behavioral science, social science, cognitive science, and human performance modeling communities; and (3) great additional progress can be expected through the funding of new research and the application of existing research in areas the panel explored.
In addition to summarizing the current state of relevant modeling research and applications, this report recommends a research and development agenda designed to move the representation of humans in military simulations forward in a systematic and integrated manner. Both the review of the state of the art and the panel's recommendations are intended to offer guidance to researchers and practitioners
who are developing military simulations, as well as to those who are responsible for providing the research and development framework for future military simulation activities.
STUDY APPROACH AND SCOPE
In the first phase of the study, several panel members attended workshops and conferences sponsored by DMSO and the Simulation, Training and Instrumentation Command (STRICOM) at which leading military contractors described their efforts to model human behavior for a variety of military simulations. The panel heard a review of modeling requirements, the state of military modeling in general, and current initiatives from representatives of DMSO. Selected presentations were obtained from specialists in the modeling community. An interim report reflecting this first phase of the study was produced in March 1997 (Pew and Mavor, 1997). During the second phase of the study, the panel held more extensive discussions with military modelers and others involved in human and organizational modeling and, taking advantage of the expertise within its membership, explored the scientific domain of human behavior to identify those areas in the literature that are pertinent to military modeling problems. The panel conducted a thorough review and analysis of selected theoretical and applied research on human behavior modeling as it applies to the military context at the individual, unit, and command levels.
It should be noted that discussion among the experts working in the domain of human behavior representation ranges much more broadly than is represented by the charge of this panel. Our focus was on the technology and knowledge available for developing useful and usable models of human behavior, from the individual combatant to the highest levels of command and control. Because they are important to the generation and success of such models, we also addressed the front-end analysis required as a prerequisite for model development and the verification and validation needed to ensure that models meet their stated requirements. The state of the art in the management of simulation and modeling processes, including scenario generation mechanisms and human interfaces to the models themselves, was considered outside the scope of the panel's work. Moreover, because the panel was charged to emphasize cognitive, team, and organizational behavior, computer science and artificial intelligence models that are not associated with behavioral organizational theories were not pursued, nor did the panel focus on theories and research related to sensory and motor behavior.
WHAT IS HUMAN BEHAVIOR REPRESENTATION?
The term model has different meanings for different communities. For some, a model is a physical replica or mock-up; for others, a model can be a verbal/analytical description or a block diagram with verbal labels. For the panel, use of
the term implies that human or organizational behavior can be represented by computational formulas, programs, or simulations. A simulation is a method, usually involving hardware and software, for implementing a model to play out the represented behavior over time. The term human behavior representation has been coined by the Department of Defense (DoD) modeling and simulation community to refer to the modeling of human behavior or performance that needs to be represented in military simulations. In this report we use the term human behavior representation to denote a computer-based model that mimics either the behavior of a single human or the collective action of a team of humans. The term may be used in the context of a self-contained constructive computer simulation that is used to simulate a battle and is run once or many times to produce outputs that reflect the battle outcomes, either individually or statistically. Or it may be used in the context of a distributed simulation of the behavior of selected battlefield elements that can be viewed by real crews performing in other battlefield element simulators, such as squads of individual soldiers, ground vehicles, or aircraft, so that the battle can be played out in the simulated world interactively.
Today's military services use human behavior representation for many different purposes. The main beneficiaries of improved behavior representations are the end-user communities for whom simulation has become an important tool in support of their activities. Training simulation users are instructors and trainees who use simulations for individual or team instruction. Mission rehearsal simulation users are members of operational forces who use simulations to prepare for specific missions. Analysis simulation users employ their simulations to evaluate alternative weapon systems, staffing requirements, doctrine, and tactics. Acquisition simulation users are those who use simulations to support acquisition decisions based on the anticipated performance of weapons systems. Joint force analysis simulation users address questions associated with improving the command, control, and communications interoperability of joint forces. In all of these domains, it has become valuable to include human behavior representation in the simulations. Of course, the scientists and engineers who will implement the models also stand to benefit from the availability of improved representations.
As the armed forces look to the future, they are attempting to identify and assess ways of effectively applying information technology, employing smart precision munitions, and integrating joint and combined operations to enhance military operations. These factors, coupled with the vision of employing military forces in an uncertain quasi-battle environment that requires information dominance to build the correct military response, add new dimensions to future battle actions. Greater importance will be placed on the ability of commanders to exercise command and control and make more precise battlefield decisions. In addition, there is increased ambiguity surrounding decisions about what military weapon systems should be developed, what joint scenarios and battle contingencies
should be trained, and what doctrine or rules of engagement should be employed.
In the face of an increasing number of military contingency missions, military planners must develop a better understanding of a broader range of force employment and potential battle outcomes for which the military services have no solid basis in experience. All of these factors lead to the conclusion that in the future, models and simulations used to train military forces, develop force structures, and design and develop weapon systems must be able to create more realistic representations of the command and control process and the impact of command decisions on battle outcomes. The representations needed are ones that more accurately reflect the impact of human behavior and the decision process of friendly and enemy leaders at multiple levels of command within real-time constraints.
In constructive simulation, it is no longer sufficient simply to use the relative strength of opposing forces, together with their fire power, to represent battle outcomes. As suggested in the Annual Report of Army-After-Next (U.S. Army, 1997)—a forward look at the implications of the Army of 2025—future battles, fought with the benefit of all the information technology now under development, will not necessarily be won by the side with the greatest fire power. To model and predict the outcomes of future wars, it will be necessary to consider information warfare as well. This implies a need for much greater emphasis on realistic modeling of the human element in battle because the human battle participants are the focus of information utilization.
The armed services are also increasingly using distributed simulation in support of technology design and evaluation, military planning, and training goals. As suggested above, in such simulations individuals participate in war games involving multiple players, each at a simulated workstation, each acting as if he or she were taking part in a real battle with views of the other participants not unlike those that would exist on a real battlefield. In this domain, human behavior representation is used to simulate the behavior of enemy forces or collateral friendly forces when there are not enough individuals available to represent all the needed players. There is also an interest in simulating the behavior of higher echelons in the command structure regarding their orders and reactions to the progress of the battlefield operations.
The rapidly changing state of the technology poses an additional challenge. Improvements in military technology—new kinds of decision aids and automation—will change the nature of the tasks to be modeled. Not only is the state of modeling technology changing, but the behavior that is to be modeled and reflected on the battlefield will change as well.
Two primary critics will view the outputs of human behavior representation and judge how successful they are. First, players in non-real-time constructive battlefield war games will observe only the resulting movements of troops and units, attrition results, and battle outcomes. Second, participants in real-time
distributed interactive battlefield simulations will see the performance of individual soldiers and higher-level units in terms of the individual and unit behavior they exhibit, the execution of plans they formulate, and the battle outcomes that result. Although explanations of how the behavior comes about may be useful for after-action reviews, they are not needed during simulation execution. Only the outcomes need to meet the expectations of the audiences that will observe them. Similarly, detailed rationales for how groups accomplish tasks are generally irrelevant. What is important is that the group behavior mirror that which is expected in the real world.
When viewed from the perspective of the simulation user (exclusive of developers), the characteristics of behavior that the visible and interpretable to the users of a simulation depend on the level of aggregation at which the behavior is presented. We consider first the individual players, either dismounted or associated with a vehicle. These individuals may be the individual combatants, ground vehicle or air system commanders, squad or platoon leaders, or commanders at a higher level. They may observe units at different levels of aggregation as well.
The most obvious behavior to be observed is the physical movement in the battlespace. It must be at an appropriate speed, and the path followed must make sense in light of the current situation and mission.
The detection and identification of enemy or friendly individual units in the human behavior representation must appear reasonable to the observer (see also Chapter 7). The visual search should depend on situation awareness; prior knowledge of the participant; current task demands; and external environmental factors, such as field of view, distance, weather, visibility, time of day, and display mode (unaided vision versus night vision goggles).
Decision making outcomes should reflect situation awareness and real environmental conditions (see also Chapter 6). The decisions concern such observations as which way to move given the plan and the situation presented by the opposing forces; they also concern whether to shoot, seek cover (evade in the case of aircraft or ship), or retreat. Movement decisions should be consistent and coordinated with the behavior of others in the same unit. Decisions should be consistent with the currently active goals. Ideally, individuals will exhibit behavior that reflects rational analysis and evaluation of alternative courses of action, including evaluation of alternative enemy actions, given the context. In practice, in time-critical, high-stakes situations, individual decisions are more likely to be ''recognition-primed," that is, made on the basis of previously successful actions in similar situations. For example, Klein et al. (1986) show how experienced fire team commanders used their expertise to characterize a situation and generate a "workable" course of action without explicitly generating multiple options for comparative evaluation and selection. In more recent work, Kaempf et al. (1996) describe how naval air defense officers spent most of their time deciding on the nature of the situation; when decisions had to be made about course-of-action plans, fewer than 1 in 20 decisions focused on option evaluation.
Representation of communication processes also depends on the specific purposes of the simulation, but should follow doctrine associated with the particular element. Communication needs to be represented only when it is providing relevant objective status, situation assessment, or unit status information that will affect action at the level of the unit being represented. Communication may take several forms and employ several modes, including direct verbal communication, hand gestures, radio communication, and data link. High-resolution models of small teams may require explicit representation of message content, form, and mode.
THE ROLE OF PSYCHOLOGICAL AND ORGANIZATIONAL SCIENCE
The panel believes that movement from the current state of human behavior representation to the achievement of higher levels of realism with respect to observable outcomes requires significant understanding and application of psychological and organizational science. Scientific psychology has more than a century's accumulation of data, theory, and experience in research concerning basic human abilities. Many of these results are so useful in practical domains that they have disappeared from psychology and become integrated into technology. For example, the design of high-fidelity audio equipment is based on precise measurements of human auditory abilities collected many years ago (Lindsey and Norman, 1977). Similarly, a number of other practical domains have been utilizing various aspects of psychological research. The development of practical human behavior representations for military simulations is especially intriguing because it presents an opportunity to construct and apply comprehensive models of human abilities that span the various subareas of psychology. The resulting synthesis of results and theory will not only be practically useful, but also serve as a stimulus for a broader and deeper theoretical integration that is long overdue.
In addition to a century of research, psychology also has about three decades of experience with computational theories of human ability. Prior to this time, most psychological theory was expressed as verbal descriptions of mental processes whose implications were difficult to define because of a lack of precision. The rise of information processing theory in psychology after World War II helped considerably by applying a metaphor: humans process information in a manner analogous to that of computer systems. Information is acquired, manipulated, stored, retrieved, and acted on in the furtherance of a given task by distinct mechanisms. The metaphor was taken further in the 1970s with theories and models of mental processes being expressed in terms of computer programs. By writing and running the programs, researchers could explore the actual implications of a theoretical idea and generate quantitative predictions from the theory.
Sociology and organizational science also have accumulated almost a century of data, theory, and experience in research concerning the behavior of groups of humans. As with psychology, many of these results are useful in practical domains and so have disappeared from these fields, in this case becoming integrated into operations research techniques and best management practices. For example, shop floor allocation procedures were derived from early work on scientific management. The development of practical applications of human behavior representations is exciting in this context because it presents an opportunity to construct and apply comprehensive models of units that span distributed artificial intelligence, organizational science, sociology, small-group psychology, and political science studies of power.
Following from early work in cybernetics, an information processing tradition emerged within sociology and organizational science. This movement arose more or less in parallel with that in psychology. However, unlike the movement in psychology, which focused on how the individual human acquires, manipulates, stores, retrieves, and acts on information, the movement in sociology and organizational science concentrated on how cognitive, temporal, physical, and social constraints limit the acquisition of information and the consequent actions taken by individuals and groups. The part of this tradition that focused on temporal, physical, and social constraints became known as structural theory. Sociologists and organizational theorists found further that people's opinions, attitudes, and actions are affected by whom they know and interact with and by what they believe others think of them. Social information processing theory and the various mathematical models of exchange and influence grew out of this research.
Many social and organizational theories are expressed as verbal descriptions of institutional, social, and political processes. As with such descriptions of psychological theories, the implications of these processes are difficult to determine, particularly for dynamic behavior. The primary reason it is difficult to derive a consistent set of predictions for dynamic behavior from these verbal models is that the behavior of units is extremely nonlinear, involves multiple types of feedback, and requires the concurrent interaction of many adaptive agents. Humans, unassisted by a computer, are simply not good at thinking through the implications of such complexity.
In addition to almost a century of research, sociology and organizational science have about four decades of experience with computational modeling of unit-level behavior. Most of these computational models grew out of work in information processing, social information processing, and structural theory. By writing and running these computational programs, researchers can explore the actual implications of theoretical ideas and generate quantitative predictions for unit-level behavior. Also, such models can be used to examine the impact of alterations in group size and composition on the resultant outcomes.
To review the state of the art in human performance modeling with specific focus on potential military applications under the purview of DMSO is especially challenging because the way the models will be used differs substantially from the goals and purposes of typical academic researchers studying and modeling human performance. Most academic researchers concerned with human performance are interested in the interplay between empirical data (experimental, field, or archival) and theory. They implement their theories through executable models so the associated detailed assumptions will be revealed, and so they can validate and evaluate the implications of those theories. Their theories are typically about specific human performance capacities and limitations, such as attention, decision making, and perceptual-motor performance. Rarely do these researchers articulate a comprehensive model of human performance that will in the aggregate reflect the behavior of real humans. Nevertheless, this is the challenge presented by the requirements of military simulations.
At the unit level, theories are typically about group performance and how it is affected by the communication and interaction among group members, procedures, command and control structures, norms, and rules. Many of these theories can be articulated as computational models. These models often illustrate the potential impact of an isolated change in procedures or structures, but they are not typically simulation models in the sense that they generate observable outputs.
The panel has been challenged by the need to focus on behavioral outcomes and to connect knowledge and theory of human behavior with realistic behavioral outcomes, rather than becoming bogged down in details of theory. However, it is our underlying belief that achieving the desired outcomes with realism and with generality requires models that are based on the best psychological and sociological theory available. In fact, the lack of such a theoretical foundation is a limitation of the current modeling efforts the panel reviewed. In the absence of theory the models are "brittle" in the sense that mild deviations from the conditions under which they were created produce unrealistic behavior and simplistic responses that do not correspond to the behavior of real individual soldiers or units. To avoid this brittleness and lack of correspondence between the model and real behavior, it is necessary to approximate the underlying structure correctly.
An example will illustrate this point. In a simulation of the behavior of a flight of attacking helicopters, the helicopters were moving out to attack. One was designated the scout and moved ahead out of sight while the others hovered, waiting for a report. The scout was shot down. Having no further instructions, the others continued hovering until they ran out of fuel. The model could obviously be fixed to eliminate this specific bug in the program by concatenating further if-then rules. However, what is really needed is a more general decision making process for the lead pilot that can select among alternative courses of action when expected information does not become available as needed. Existing
theory is the best means of defining this structure. Furthermore, as one attempts to aggregate forces and model larger units, the unique processes and theory become even more important.
The panel also examined models of learning as frameworks within which to build specific behavioral models. These models of learning are often not based on an explicit theory and use a representational framework that is broader than the specific behavior to be simulated. However, operating within such a framework may be helpful and important in minimizing brittleness.
SETTING EXPECTATIONS IN THE USER COMMUNITY
In the panel's discussions with various representatives of the user community, it became clear that there is wide variation in users' expectations of what is possible with regard to generating human behavior that is doctrinal, realistic, creative, and/or adaptive. We suspect that what can be achieved in the near term is much more limited than some of these users expect. One purpose of this study was to elaborate those aspects of model theory and implementation the panel believes are achievable now and those aspects that require significant translation of scientific theory and principles before being developed as components of computer-based behavioral models, as well as those aspects of behavior for which the behavioral/social science community has inadequate knowledge for use in developing realistic models in the near future. In addition to presenting an approach to modeling methodology, including both model development and model validation, the panel's goal was to set forth in general terms the theoretical and operating principles of models that are applicable to human behavior representation, to describe specific applications of these theories and principles, and to identify the most promising paths to pursue in each modeling area. Much work remains to be done. There is an enormous gap between the current state of the art in human and organizational modeling technology on the one hand and the military needs on the other.
Subsequent chapters examine the potential psychological and sociological underpinnings of extensions to both the approaches and content of such models. These chapters represent the collective judgment of the panel concerning representative promising areas and approaches for expanding the models' behavioral content. Given the scope of psychological and sociological inquiry, it is likely that another panel at another time would put forth an equally appropriate, overlapping but different set of areas and approaches. What is presented here reflects this panel's expertise and collective judgment.
A fundamental problem that faces the human behavior representation community is how to determine which of the many modeling requirements will make a difference in the resultant quality of the models, based on the intended use of the simulation. As the panel deliberated, it became clear that a consultative behavioral science panel cannot set these priorities without much more experience
in dealing with the specific concerns of the military community. The panel may be able to say which requirements will produce models that behave more like real humans, but this is a different set of priorities from the requirements that will produce models likely to be perceived by simulation users as being more like real individual combatants and military units. If one asks program managers or subject matter experts, they will say they need all the fidelity they can get, but this is not a helpful response in an environment of limited resources where design decisions involve tradeoffs among a set of desirable simulation goals. It is just not known which of the many improvements in human behavior representation will really make a difference in the way a modeled combatant will be viewed as regards meeting the expectancies of the opposing force and minimizing the ability to "game" the simulation. This issue is analogous to the traditional problem of simulation fidelity. Analysts would like to have high fidelity only where it matters, but no one has shown clearly just what aspects of fidelity matter. The situation is no different with human behavioral representation.
ORGANIZATION OF THE REPORT
Chapter 2 characterizes the current and future modeling and simulation requirements of the military and reviews several existing military modeling efforts. The central portion of the report comprises chapters that focus on the science and technology of human behavior representation. Chapter 3 provides a general review of several integrative architectures for modeling the individual combatant. Each review includes a brief discussion of how the architecture has been applied in the military context. Chapters 4 through 9 are devoted to an analysis of the theory, data, and state of modeling technology of individual human behavior in six key areas: attention and multitasking (Chapter 4), memory and learning (Chapter 5), decision making (Chapter 6), situation awareness (Chapter 7), planning (Chapter 8), and behavior moderators (Chapter 9). The emphasis in these chapters is on the current state of research and development in the field under review. To the extent possible, the focus is on the behavioral theory on which understanding of cognitive mechanisms is based. However, in the chapters on situation awareness and planning, where less behavioral theory exists, we focus on the strengths and weakness of current modeling approaches. Chapters 10 and 11 address issues and modeling efforts at the organizational level: Chapter 10 covers command, control, and communications (C3), whereas Chapter 11 deals with belief formation and diffusion, topics of particular interest in the context of information warfare. Chapters 3 through 11 each conclude by presenting conclusions and goals for the short, intermediate, and long terms in their respective areas. Chapter 12 presents general methodological guidelines for the development, instantiation, and validation of models of human behavior. Finally, Chapter 13 provides the panel's conclusions and recommendations regarding a programmatic framework for research, development, and implementation and for infrastructure/information exchange.