The Art and Science of Battlefield Decision Making

Angel Mesa | CalypsoAI Sr Solutions Architect | US Army Col. (ret) 

In the winter of 369-370 BCE, the obscure Theban general Epaminondas marched his army of democratic yeomen and their allies one hundred and eighty miles south into Laconia, the legendary home of the Spartan army and a region inviolate from invasion for more than six hundred years. In a little over four months, the Theban-led invasion dismantled the system of Spartan apartheid – the enslavement of two hundred thousand Messenian serfs – and permanently weakened the military culture, which had rested upon the principle that elite warriors should not work. The culture of classical Greece was never again the same.

Since antiquity, the battlefield has evolved from open fields with massive formations conducting frontal attacks, to the introduction of armored vehicles and air warfare in the 20th century. Enter the 21st century: Multi-domain operations entailing numerous platforms collecting and transmitting data globally are creating security challenges along with the ability to efficiently and thoroughly analyze the enormous amount of data generated daily. To further compound the friction is the numerous decisions U.S. Commanders make to rapidly project power in the time of conflict while simultaneously sustaining a competitive advantage in the infinite game known as the Great Power Competition. How do commanders and their staff today and in the future sort, arrange, visualize, and trust the data to make decisions? The challenge facing the Army and arguably the National Security Community is the ability to adapt to the “Four Vs:” volume, variety, velocity, and veracity of data encountered in daily operations.

Former Secretary of Defense, Dr. Mark Esper stated during the DoD Artificial Intelligence Symposium and Exposition, “Artificial Intelligence has the potential to change the battlefield, and the country that is first to field it will have enormous advantages over competitors.” This creates another challenge of understanding the vulnerabilities associated with employing AI/ML at scale. Additionally, comprehending what the algorithmic models are doing and why. Most importantly, are the algorithms accomplishing what they were developed to do. Science is outpacing the art of decision making; specifically, the power of the AI Triad – algorithms, big data, and computer power is exponentially outpacing the art of battlefield decision making.

Systems are being developed without rigorous testing and with limited (if any) security, while organizations face a critical talent gap. This causes challenges across National Security with regards to trusting AI’s security, robustness, and performance resulting in degraded capabilities and technological utility for warfighters and responders. Current challenges around AI include the following:

– Validation – Does the model do what is intended?

– Model Insight (akin to explainable)- How does the ML model work?

– Robustness – How is the ML model impacted by attack or error?

AI’s potential to positively impact Army’s readiness and operational decision making during a contingency is improved by a deep understanding of the data that drives our autonomous decision-making tools. As the commander of Army Futures Command, Gen Murray states, whoever can see the data, understand the data and act first will win. No longer is the ability to conduct battlefield calculus enough. A comprehension of human-machine interaction is pivotal. Commanders must be able to understand, and in most cases visualize how the algorithms (AI/ML model) arrive at the recommended solution. A shift-left movement, or in warfighter terms, “get left of the bang” by emphasizing robustness and transparency from data ingress to model deployment. Utilizing a building block approach to establish a foundation in AI/ML employment centered on security and AI trust will radically change the scale to which AI/ML is deployed in the Army.

A way to increase the trust in AI is by defeating AI’s “Black Box” perception. Through incredible technological advances, the U.S. force projection capability can occur in hours and not four months as it did for Epaminondas. To arrive at the decision to project power there are numerous decision points. Implementing transparency at the forefront of AI development will allow leaders to quickly confirm or deny facts and assumptions through the military decision-making process. Though the example is simple in nature, the key takeaway is an autonomous tool with built-in transparency can defeat the “Black Box” perception while simultaneously addressing the challenge of the four “Vs:” volume, variety, velocity, and veracity of data encountered in daily operations. Commanders will continue to operate within various planes of decision-making and assess the quality of their decision within the spectrum of conflict and the great power competition. Decision dominance can be attained by leveraging AI/ML to sort, arrange, and visualize data in an easily consumable way. However, the suite of AI/ML tools must be secure, trustworthy, and transparent to achieve decision dominance.

CalypsoAI’s Secure Machine Learning Life Cycle product, VESPR is an end-to-end workflow for the development and deployment of robust, secure, and explainable machine learning models. VESPR analyzes AI systems for a range of vulnerabilities while providing real-time information regarding algorithm performance, robustness, and risk levels. CalypsoAI’s mission it to enable Artificial Intelligence (AI) creators and consumers to build, train, and test their Machine Learning (ML) models. CalypsoAI delivers ML models that are ready to be deployed safely and securely across the National Security community. VESPR entails a suite of tools to assist commanders with the art and science of battlefield decision-making in the 21st century and beyond.