Algorithmic Warfare or Algorithmic Warfare and Focal Point Analysis
By Daniel Phillips
Introduction to Algorithmic Warfare
The Military utilizes multiple algorithms to define and predict adversaries, develop staff estimates, and develop courses of action. These algorithms include, but are certainly not limited to: intelligence preparation of the battlefield and the Eikmeier method for center of gravity analysis. They have been developed in large part incidentally, from trial and error efforts by tacticians attempting to apply tenants of military theorists such as: Jomini, Clausewitz, and Sun Tzu. The evolution of said algorithms, grow and evolve when prodded by national security crisis’s, such as war. This ebb and flow timeline for growth allows adversaries to methodically maintain parity with American tacticians, whose focus transitions from research and development, to sustaining the force, to fiscal austerity depending on the political administration in power. One industry whose focus is on developing data analysis and decision making algorithms is that of machine learning and artificial intelligence experts. Who we will refer to as AI experts. Modern military tacticians and theorists would be well served to explore existing AI theories for data analysis and learning, find the overlaps where they exist and identify new applicable algorithms and implement them. In this essay I will briefly define three acknowledged AI algorithms
AI Algorithm One: Inverse Deduction
Inverse deduction can explained fairly simply to a mass group who have even the most basic understanding of Algebra. In a snap shot, the concept is that if you define the product and one/many input/s, you can identify what else was required to get to the product.
Example: A+B=C
If A=1 and C=3
B must =2
This simple algorithm utilized for pointing out what is missing seems so basic that it couldn’t possibly have a mass application to warfare, right? I’d argue that it is an algorithm that knowingly or not, is utilized by military intelligence specialists daily when they utilize adversary templates to develop situation templates. An adversary template is defined by MCRP 2-10B.1 Intelligence Preparation of the Battlespace (IPB);
“…models based on postulated adversary doctrine. They illustrate the disposition and activity of adversary forces conducting a particular operation arrayed on ideal terrain. Adversary templates depict the adversary’s nominal organization, frontages, depths, boundaries, and control measures for combat…”
Adversary templates are the product. The known inputs, such as terrain and perhaps a unit of an adversary, facilitate the intelligence specialist in building a situation template, which is defined by MCRP 2-10B.1 as;
“…to depict adversary dispositions based on the effects of the battlespace and the pursuit of a particular COA. This template accounts for the adversary’s current situation with respect to terrain, training and experience levels, logistics status, losses, and dispositions.”
Intelligence specialists utilize inverse deduction to place adversaries in physical locations, with realistic estimated missions. Upon inverse deduction being completed at the tactical level, it in itself can serve as an estimated input to factor into an operational or strategic product/or end state of an adversary. This leads into the next algorithm, back propagation.
AI Algorithm Two: Back Propagation / Focal Point Analysis
Back Propagation utilizes overlapping situation templates and an estimated adversary tactical, operational and/or strategic end state to measure adversary systems and measures of effectiveness. Depending on how the adversary campaign is going (good, bad, in the middle) American military tacticians can go back through the different warfighting functions for the adversary and identify which units and capabilities of the adversary are simply present, which are white noise, and which enable the network. This technique is best demonstrated through a situational graphic (see graphic 1. below).
For this specific situation we will focus on a tactical level problem. The enemy wants to defend an area Americans have identified as their Amphibious Objective Area.
Expected Adversary End State: Defeat American Amphibious Assault
Utilize the location as a known input, due to us knowing where we want to go. Place an enemy situation template over that area. Rather than identifying simple enemy organizations, overlay enemy warfighting functions over the area to develop the regional network for the defense of the amphibious objective area. It may look like this.
Once critical capabilities and critical requirements are identified for each of the warfighting functions, you can easily identify which elements seem to overlap across warfighting functions. These may be known as focal points. Or points that connect different complex elements of an enemy system and allow it to function in an interconnected manner. I’ll point out two focal points in this system.
Focal point one: Ability to maneuver along interior lines for: classes of supply, casualties, fighting forces… etc.
Focal point two: Accurate/timely information is disseminated to the forces for execution of follow on actions.
Disruption of either of these focal points would have ripple effects across all of the warfighting functions. They’d either break a portion of the network, or force an overreliance on other critical aspects of the network, or even instigate an evolution (we’ll discuss in the final algorithm).
Let’s discuss focal point one and distinguish the difference between aspects of that focal point that are merely at play in the system, which ones are white noise, and which ones enable the entire the warfighting function.
Based on inverse deduction we know that the enemy unit is populated by light infantry who are largely foot mobile. They receive their logistics in large part from three truck companies. Due to the large amount of area that we are analyzing, it is not realistic to assume that the light infantry elements can transport themselves to where they need to be. Therefore, the truck companies must be required to maneuver the infantry to meet us in combat. We also know that the truck company is conducting daily logistical operations to simply keep the unit running. Therefore, there must be a prioritization or a command given to redirect their efforts to transition from operational sustainment to movement of fighting forces.
(In our mock scenario) The amphibious objective area covers a developed region with more improved roads than we can either target or accurately observe. That having been said, we know the few locations that we want to land. Rather than target the roads, we can identify any choke points to that area, to both contain and isolate.
The three aspects of this specific focal point are: Trucks and roads exist and are required in this scenario, but would be too costly to target. Because of this, the daily movements of logistical goods may become white noise, distracting from where we should focus our efforts. Targetable items that can fracture the network must be refined and economized to strike only the required choke points that serve to both isolate and contain specifically where we want to go. Thus the interior lines focal point is leveraged to not negatively impact the amphibious assault.
Focal Point Two focuses on the dissemination of information to the enemy formations, to enable them to make decisions that would defeat our amphibious assault. There is not one warfighting function that doesn’t require accurate information flow. Because of this, there is an incredible amount of white noise in the scenario. Meaning there are many aspects of the enemy network that are ever present but will not significantly degrade the focal point. An example is this: an enemy infantry unit does require communications with outside agencies to execute its mission. That having been said, if that enemy infantry unit losses its specific ability to communicate along a set frequency spectrum, it does not take the infantry unit’s ability to maneuver based on other information outlets, or it does not degrade adjacent infantry units’ abilities to maneuver. Therefore, the actual radio, or gadget is not more than white noise. The targetable crux of the information would be the information itself.
Information is a difficult focal point to kinetically strike. Especially if it is known that when one command and control node goes down, another will inevitable arise due to the proliferation of highly technical communications devices that are readily available. To target information, the military tactician must understand and utilize capabilities from domains outside of the physical realm (such as: cyber and information).
Rather than waste millions of dollars of finite stocks of ordnance; maneuver, and media campaigns could work to deceive the enemy as to the amphibious assault force’s true intent. The idea would be to utilize the most cost effective methods per each focal point, to maintain the highest level of preparation for when inevitably the fight does degenerate to killing via combined arms.
To summarize a militaresque interpretation of back propagation, you must first conduct inverse deduction to develop adversary templates for who you expect to fight. Once you’ve identified the battlefield, build situational templates on the physical terrain across all warfighting functions. Find the overlapping focal points and see which domains those focal points pass through. Separate the components of those focal points between those that are present, white noise, and critical. Then target the critical components of the focal points through the domain that will have the most wide ranging effect across the adversary network, while protecting your own capabilities. (It must be understood that the specific example discussed was simplified significantly to fit the parameters of this essay; but the concepts remain.)
Macro Back Propagation
I just discussed back propagation, through a scenario at the tactical level. It can be elevated to an operational or strategic level as long as there are subject matter experts in the process that understand the functions of the other elements of national power: Diplomatic, economic, military (Our warfighting function), and information.
The process is similar to that of the tactical level, with the addition of one step.
Step 1. Identify the adversary strategic policy or end state desired.
Step 2. Identify the known/expected elements of national power being utilized by the adversary government or organization. Execute inverse deduction to determine any elements that are being concealed, or may be concealed. (This is the added step)
Step 3. Identify the characteristics of the doers of actions along the different elements of national power through inverse deduction again. In a tactical scenario, this is where adversary templates would be drawn up and built into situation templates. Warfighting functions would be analyzed for critical capabilities and critical requirements.
Step 4. Identify focal points across the elements of national power that are being leveraged.
Step 5. Identify which domain/s will be maneuvered through to have a desired effect on the adversary focal points. Unless there is overt war, adversary focal points should be attacked via our elements of national power other than military, and domains that allow for non-kinetic actions.
Step 6. Monitor for adversary evolution.
AI Algorithm Three: Evolution
The evolution algorithm can easily be defined in its lifecycle, but is inherently difficult to observe for valuable results, given that once it is being conducted, known aspects of the nature of war are probably present, such as: uncertainty, fluidity, disorder, complexity, the human dimension, and danger. That aside, the process is this:
Step 1. Identify the population you will be studying
*This will have been done upon completion of inverse deduction at the early stages of planning.
Step 2. Evaluation
*Evaluation is conducted by translating adversary templates in to situational templates and developing your cross functional and focal point analysis.
Step 3. Fitness Value
*After targeting has occurred, conduct an analysis to identify which focal points ceased to exist, which ones broke the system, which ones were not impacted.
Step 4. Reproduction
*Once the fitness value of each targeted focal point has been assessed, which ones did the enemy try to simply reproduce and reintroduce into a battle of attrition. Which ones did the adversary attempt to reduce relevance in and focus in other areas. Which ones did the adversary continue to want to use, but totally modified the way in which they did.
Step 5. Crossover
*As the adversary focal points go away, change, or are reinforced, the way in which they are targeted must be updated to continue to find the most efficient means for targeting. This is done by placing a value on the different methods exerted for targeting.
Step 6. Mutation
*With successful targeting of adversary focal points, the adversary system will change. Some changes will be evident during the planning process and will be able to be forecasted. Others will not.
Summary and acknowledgements
My worry in writing this, is that it will be adopted completely as prescribed. The idea behind this paper is that there are many existing algorithms in AI research and other areas of civilian and technological sectors that have relevance in warfighting theory. Warfare should never be planned for in a haphazard way. Attempting to implement Sun Tzu idioms is in fact haphazard. Planners and tacticians need to develop a structured means to approach planning for war. Professors such as Pedro Domingos (author of The Master Algorithm, and whom I adopted several of my terms from) and many others spend their careers studying ways to analyze data, learn from it, and build algorithms to more efficiently work off of that data. It is the warfighting theorist’s job to tap into them and refine their own processes. We can and must do better. There is no excuse to fall back on doctrine published decades ago when planning to fight adversaries of the future.
Citations.
Books and Published Works.
-Domingos, Pedro, The Master Algorithm, How The Quest For The Ultimate Learning Machine Will Remake the World, Basic Books, 2015.
It is from this work that my naming convention and basic outline for algorithms was taken.
-Marine Corps Technical Publication (MCTP) 3-40B, Tactical Level Logistics, Marine Corps Logistics Operations Group, May 2016
-Marine Corps Reference Publication (MCRP) 2-10B.1, Intelligence Preparation of the Battlefield, Combat Development and Integration, Quantico, Virginia, March 2015.
-Marine Corps Doctrinal Publication (MCDP) 1, Warfighting, Combat Development and Integration, Quantico, Virginia, April 2018.
Videos.
-The Master Algorithm/Pedro Domingos/Talks at Google, Youtube, 2016.
Authors and Theorists Referenced (specific work wasn’t referenced).
-Karl Von Clausewitz
-Sun Tzu
-Antoine-Henri Jomini
-Dale C. Eikmeier