THE LANDPOWER ESSAY SERIES
LPE 19-1 / January 2019 • PubLishEd by thE institutE of Land WarfarE
Training the Machines:
Incorporating AI into Land Combat Systems
by Lieutenant Colonel Stephan Pikner, U.S. Army
Introduction
Recent developments in articial intelligence (AI), or machine learning, have the potential to revolution-
ize how humans interact with technology. Rather than merely responding to direct inputs in predened ways,
systems that can sort through vast amounts of data and rene their network structures are rapidly improv-
ing their ability to predict and categorize. These advances are driven primarily by the massive quantity of
information that can be captured and correlated. The more data that is integrated, the more precise the model
becomes—whether it is driving patterns on city roads, overhead imagery of farm elds or medical scans of
disease-prone organs. AI also has broad military applications, ranging from cybersecurity to aviation main-
tenance diagnostics to higher echelon military intelligence analysis.
1
Its potential in military applications,
however, is sharply limited by the rarity of war. Lacking real-world data to train on and unable to draw from
historical cases, AI-enabled land combat systems* may be severely limited in their eectiveness, especially
in the critical opening phases of war.
This is not a call to abandon research into military applications of AI. Instead, Army leaders must recog-
nize the limitations of machine learning in particular contexts, especially in situations in which adaptation and
the recognition of causal links are critically important. Unquestioning acceptance of AI and the wholesale dis-
missal of it are both awed approaches. Eectively integrating these systems into Army combat formations
requires a general understanding of how they predict and classify, how they are trained and how they can best
complement the innate strengths and adaptability of our nation’s warghters.
What Drives AI Predictions and Classifications?
The power of AI lies in its ability to pull from increasingly massive amounts of data collected from a
range of sources and discern correlations between variables to predict behavior. As the amount of data grows,
so does the power of these predictions. One example, if controversial: By integrating closed-circuit television
image data, mobile phone eavesdropping and retinal scans at police checkpoints, the Chinese regime has built
the powerful and deeply intrusive Integrated Joint Operations Platform system to monitor the ethnic minor-
ity Uighur population in Xinjiang Province. The data predicts people who demonstrate patterns of behaviors
deemed threatening to the regime, which then arrests those people and places them into re-education camps.
2
More widely, technology rms are routinely mating their increasingly ne-grained inventory of human char-
acteristics—captured through the integration of data generated by mobile phone applications, publicly ac-
cessible nancial and demographic characteristics, and consumption patterns—with deliberate variation of
product characteristics. This integration allows internet retailers and social media platforms to use machine
learning techniques to sharpen and personalize their business models in ways that were previously impossible.
The Landpower Essay series is published by the Association of the United States Army’s Institute of Land Warfare. The series is designed
to provide an outlet for original essays on topics that will stimulate professional discussion and further public understanding of the land
-
power aspects of national security. This paper represents the opinions of the author and should not be taken to represent the views of
the Department of the Army, the Department of Defense, the United States government, the Institute of Land Warfare, the Association of
the United States Army or its members. For more information about AUSA and the Institute of Land Warfare, visit www.ausa.org.
* Land combat systems refers to tactical-level weapons platforms, such as armored vehicles, helicopters, air defense or indirect re systems, whose
eectiveness may be improved with articially intelligent navigation, targeting, self-defense and maintenance diagnostic capabilities.
