Artificial Intelligence / Large Language Model Course of Action Analysis

In today’s high tech, interconnected world, international competition is very different than in the past. Not only is it DIMEFIL > PMESIIPT - wide, but strategic campaigns unfold much faster than before. And this is important. Unless one can operate inside the adversary’s OODA (Observe, Orient, Decide, Act) loop, one risks being caught off unprepared, or worse, blindsided.     This new reality presents significant challenges to the creation of effective Courses of Action (COAs). First, it suggests that the existing COA development approaches – e.g., conventional military-centric JIPOE analyses -- are no longer sufficient. To ensure competitiveness, JIPOE must be extended to include diplomatic ops, info/influence warfare, economic and financial warfare, and lawfare. Second, they suggest that speed in planning is no longer a luxury but a necessity. What once took days to accomplish must now be completed in minutes or hours — without sacrificing precision or effectiveness. And third, they demand that planning processes must be adaptable, and capable of responding appropriately to clever, flexible adversaries.A mere five years ago mitigation of these challenges would not have been possible. ‘One-offs’ could have been accomplished, but timely, general purpose DIMEFIL-wide COA development was beyond the state-of-the-art.  However, as we demonstrate, this is no longer the case.  Through the innovative application of AI, we can now meet these challenges quite handily. Moreover, we can do so with minimal impact to current protocols and tools.Our objective in the proposed activity is to take a significant step forward in meeting the COA development challenges. In particular, we propose that we leverage state-of-the-art AI to develop tools and processes to enable us to rapidly develop COAs capable of  i) effectively and efficiently disrupting adversary campaigns while ii) facilitating our own.  Recognizing that the adversary has a say in how things evolve, our process is designed to be flexible and adaptable. At each step in the workflow, the process considers options or sets. For instance, for an objective, it will generally offer optional sub-objective sets capable of accomplishing the objective; in the case of a sub-objective, it may offer optional Red targets; and, in the case of DIME sectors, it may suggest optional sectors. These possibilities will be considered, and the most effective options selected by means of a scenario-specific selection process. In addition, our process is designed to be extensible. Not only will it handle today’s problems, and leverage today’s technologies, but we are designing it to deal with envisioned problems and to exploit anticipated technologies.