Enemy Location Prediction in Naval Combat Using Deep Learning; New Direction for an Intelligent Military: Dueling Perspectives

McLaughlin, Kent, School of Engineering and Applied Science, University of Virginia
Beling, Peter, EN-Eng Sys and Environment, University of Virginia
Baritaud, Catherine, EN-Eng and Society, University of Virginia

The integration of artificial intelligence (AI) into the military is a momentous and accelerating process. Infusing military systems with AI has the potential to revolutionize the way wars are fought. The technical research applies machine learning to naval combat in order to provide officers with real-time insights from the mountains of data available on the naval battlefield. The science, technology, and society (STS) research has a broader focus, addressing the ecosystem of military AI research and development (R&D) as a whole. In addition to assessing the current state of affairs in the development of intelligent military systems, the STS research aims to identify the shortcomings of the current system and propose potential improvements. The STS and technical research are closely linked to one another, with the STS research holistically examining the intelligent military R&D system of which the technical research is a part.
The technical research applies machine learning to naval combat to predict the location of an unseen enemy vessel. Naval combat is an immensely complex domain in which quality information can be a decisive factor, and as such is an ideal candidate for the application of machine learning. Machine learning, at its essence, is a means of gaining actionable insights from a large and unwieldly data set, precisely like what is produced in naval combat. In order to build a prediction engine for locating an unseen enemy ship, the technical project uses data from a sufficiently realistic naval warfare video game, World of Warships. To ensure the battle data closely resembled the strategy and coordination of real-world naval combat, it was extracted from a tournament in which retired naval officers served as fleet commanders. The data from this tournament served as the testing and training set for the prediction algorithm, with the location and types of all known ships serving as inputs and the predicted location of a single unseen enemy ship as the output.
The technical research culminates with a predictive model that is reliably able to provide actionable information concerning the location of an unseen enemy vessel. The machine-learning-based algorithm accurately predicts the location of the enemy ship approximately 40% of the time. While this level of accuracy may seem pedestrian, the model is immensely more valuable than the alternative, in which an officer can only blindly guess the location of the enemy ship. With further training on more sophisticated naval simulators, the predictive model developed in this technical research can be refined and improved to eventually be adopted for use in real-world naval combat.
The STS research takes a step back from the specific application of the technical work in naval combat, and instead examines the development process of intelligent military systems as a whole. Given the highly structured and segmented nature of military R&D, the STS research seeks to answer the question of whether the limitations of the current R&D system produce societally suboptimal results in the realm of AI technology. The research finds that the imbalance of the current military research ecosystem results in a development pattern that largely ignores the potential for societal benefits from military AI research. This thesis is examined and proven through the lens of Pinch and Bijker’s Social Construction of Technology framework, focusing on the varied interpretations of AI and its applications by different social groups.
The STS research details how the combination of an arms-race mentality and a lack of civilian input prevent the military development of AI technology from reaching its full potential for societal benefit. Intensifying global competition for intelligent military systems blinds decision-makers to the potential civilian benefits of AI technology. The STS research posits that the solution to this predicament is to bring the R&D system into alignment by involving all relevant social groups, namely incorporating civilian input into the development process.
Military research into AI-based technology will continue to accelerate and expand in the coming years. In anticipation of this continued technological pursuit, careful consideration must be given to the societal implications of such an endeavor. The STS research provides a framework wherein military research projects, such as the technical research discussed here, can be directed responsibly and with an awareness of potential civilian benefit.

BS (Bachelor of Science)
machine learning, military, artificial neural network, Social Construction of Technology (SCOT), artificial intelligence

School of Engineering and Applied Science
Bachelor of Science in Systems Engineering
Technical Advisor: Peter Beling
STS Advisor: Catherine Baritaud
Technical Team Members: Morgan Freiberg, Adinda Ningtyas, Oliver Taylor

All rights reserved (no additional license for public reuse)
Issued Date: