Realtime learning of doctrine and tactics using neural networks and combat simulations

Limitations of the traditional Artificial Intelligence paradigm restrict its capacity to support manageable and verifiable knowledge base development for expert system simulations. This report argues that because expertise acquired in dynamic military domains is associated with unique aspects of memory and action-response sequences that are resistant to word-based cues and expression, an alternative model is required for acquiring and representing knowledge in these competitive environments. Motivated by an emerging research into adaptive and neural models, this report documents a USA TRADOC supported research program that proposed and evaluated an adaptive model within the Army's high-resolution combat simulation CASTFOREM. The prototype was designed to support a synthetic model of intelligence that represents complex goal functions, rule-based (deductive) reasoning in the presence of environmental activity that is consistent with expectation, as well as goal-based (inductive) reasoning in the presence of uncertainty – unfamiliar patterns of activity. The experiment demonstrated that the prototype is not only capable of generating effective tactics, but the prototype converges to stable, rule-based behavior quickly and efficiently. These results motivate further research into the application of intelligent simulations to broader, long-term goals such as developing and optimizing tactics for developmental hardware and software systems.