Using a Competitive Approach to Improve Military Simulation Artificial Intelligence Design

The research presented in this paper attempts to show how using a competitive approach to artificial intelligence (AI) design can lead to improvement of the AI solutions used in military simulations. To demonstrate the potential of the competitive approach, ORTS, a real-time strategy game engine is used. The idea is to setup a tournament of virtual battles between base case AIs and new test AIs, and by using the information from these battles to advance the test AIs. The analysis of the results from the experimental tournament shows possible advantages and applications of the competitive approach. At the end of the paper, some conclusions and recommendations for future work are made. 1.0 INTRODUCTION In recent years, the focus of military operational analysis has been switching from massive conflicts, dominant in the Cold War, to local conflicts and local fighting, shaping the post 9/11 world. In the process of exploring this new type of warfare, the modeling and simulation community is using more and more the solution suggested by Lauren (1999) by treating the complexities of warfare as a complex adaptive system (CAS). This idea is very close to the idea presented in the research of Ilachinski (2000) on Irreducible Semi-Autonomous Adaptive Combat (ISAAC)/EINStein models. His idea is to use a bottom-up approach, where the individual combatants are modeled, and their interaction in a battlefield produces desired data for combat analysis. This allows researches to address important components of asymmetrical warfare such as spatial layout of the forces, tactical movement or target acquisition and assessment, which are not addressed by traditional models. To present advanced features of combat using the principles of CAS, in many cases the designers employ solutions from the domain of artificial intelligence (AI) (Pawloski, 2001, Reece, Kraus, & Dumanoir 2000). The main usage of AI solutions in military simulations is to model different tactics and behaviors of forces. This is done by using AI as a control system for forces in the simulation, which are sometimes called synthetic forces. This allows the synthetic forces in the simulation to mimic the reactions and behavior of the real forces in the battle. The evaluation of success of these solutions is normally made by subject matter experts who subjectively compare the expected entity behaviors to those shown in the military simulations. Although this approach is proven to lead to improvements, it does not provide enough information to answer questions such as: Is there possibly a better strategy or set of tactics for solving the same problem? Do additional factors exist that can be used to produce AI solutions that can demonstrate more realistic or more natural behaviors? In the search for alternative approaches, which may answer some of these questions, this paper is focused on exploring how a competitive approach in AI design can be used to produce better AI solutions for use in military simulations.