ISAAC is a simple multiagent-based "toy model" of land combat that was developed to illustrate how certain aspects of land combat can be viewed as emergent phenomena resulting from the collective, nonlinear, decentralized interactions among notional combatants. ISAAC takes a bottom-up, synthesist approach to the modeling of combat, vice the more traditional top-down, or reductionist approach, and represents a first step toward developing a complex systems theoretic analyst's toolbox for identifying, exploring, and possibly exploiting emergent collective patterns of behavior on the battlefield. This model was developed as part of a two-year project that assessed the general applicability of complex systems theory to land warfare. This project (completed in September 1997) was sponsored by the Marine Corps Combat Development Command, Quantico, VA.

The fundamental motivation of this study is to develop a multiagent-based software tool to address the basic question: "To what extent is land combat a self-organized emergent phenomenon?" As such, its intended use is not as a full system-level model of combat but as an interactive toolbox (or "conceptual playground") in which to explore high-level emergent behaviors arising from various low-level (i.e., individual combatant and squad-level) "interaction rules." The idea is not to model in detail a specific piece of hardware (M16 rifle, M101 105mm howitzer, etc.), but to provide an understanding of the fundamental behavioral tradeoffs involved among a large number of notional variables. In ISAAC, the final outcome of a battle -- as defined, say, by measuring the surviving force strengths -- takes second stage to exploring how two forces might co-evolve during combat.

Models based on differential equations homogenize the properties of entire populations and ignore the spatial component altogether. Partial differential equations -- by introducing a physical space to account for troop movement -- fare somewhat better, but still treat the agent population as a continuum. In contrast, ISAAC consists of a discrete heterogeneous set of spatially distributed individual agents (i.e., combatants), each of which has its own characteristic properties and rules of behavior. These properties can also change (i.e., adapt) as an individual agent evolves in time.

The basic element of ISAAC is an ISAAC Agent (or ISAACA), which represents a primitive combat unit (infantryman, tank, transport vehicle, etc.) that is equipped with the following characteristics:

• Doctrine: a default local-rule set specifying how to act in a generic environment

• Mission: goals directing behavior

• Situational Awareness: sensors generating an internal map of environment

• Adaptability: an internal mechanism to alter behavior and/or rules.

The putative combat battlefield is represented in ISAAC by a two-dimensional lattice of discrete sites. Each site of the lattice may be occupied by one of two kinds of ISAACAs: red or blue:

The initial state consists of either user-specified formations of red and blue ISAACAs positioned at diagonally opposite corners of the battlefield or of a random distribution of red and blue ISAACAs occupying the central square region (of user-specified dimension). Red and blue flags are also typically (but not always) positioned in diagonally opposite corners: a red flag in the red ISAACAs corner and a blue flag in the blue ISAACAs corner. A typical goal, for both red and blue ISAACAs, is to successfully reach the flag positioned in the diagonally opposite corner. ISAAC also has the provision of defining a notional terrain.

ISAAC is designed to allow the user to explore the evolving patterns of macroscopic behavior that result from the collective interactions of individual agents, as well as the feedback that these patterns might have on the rules governing the individual agents' behavior. While this early version of ISAAC can do no more than suggest new ways of thinking about some old issues, it is encouraging to note that, even at this early juncture, ISAAC already has an impressive repertoire of emergent behaviors:

• Forward advance
• Frontal attack
• Local clustering
• Penetration
• Retreat
• Attack posturing
• Containment
• Flanking Maneuvers
• Defensive posturing
• Guerilla-like assaults
• Encirclement of enemy forces
  many more ...

Moreover, ISAAC frequently displays behaviors that appear to involve some form of "intelligent" division of red and blue forces to deal with local "firestorms" and skirmishes, particularly those forces whose personalities have been "evolved" (via a genetic algorithm) to perform a specific mission. It must be remembered that such behaviors are not hard-wired-in but are effectively an emergent property of a decentralized and nonlinear local dynamics.

Most traditional models focus on looking for equilibrium solutions among some set of (pre-defined) aggregate variables. The Lanchester Equations, for example, are effectively mean-field equations, in which certain variables such as attrition rate are assumed to represent an entire force and the outcome of a battle is said to be understood when the equilibrium state has been explicitly solved for. In contrast, ISAAC focuses on understanding the kinds of emergent patterns that might arise while the overall system is out of (or far from) equilibrium. The payoff of this multiagent-based approach is a radically new (and decidedly non-Lanchesterian) way of looking at fundamental issues of land combat. Specifically, ISAAC is being designed to help analysts:

• Understand how all of the different elements of combat fit together in an overall "combat phase space": Are there regions that are sensitive to small perturbations, and, if so, might there be a way to exploit this in combat (as in selectively driving an opponent into more sensitive regions of phase space)?

• Assess the value of information: How can I exploit what I know the enemy does not know about me?

• Explore tradeoffs between centralized and decentralized command-and-control (C2) structures: Are some C2 topologies more conducive to information flow and attainment of mission objectives than others? What do emergent forms of a self-organized C2 topology look like?

• Provide a natural arena in which to explore consequences of various qualitative characteristics of combat (unit cohesion, morale, leadership, etc.)

• Explore emergent properties and/or other "novel" behaviors arising from low-level rules (even doctrine if it is well encoded): Are there any universal patterns of combat behavior?

• Provide near-real-time tactical decision aids by providing a "natural selection" (via a genetic algorithm) of tactics and/or strategies for a given combat scenario