HomeAU PressBook ReviewsDisplay Review
Air University Press Banner

Warbot 1.0: AI Goes to War

Warbot 1.0: AI Goes to War by Brian Michelson. War Planet Press, 2020, 421 pp.

Warbot 1.0: AI Goes to War is a science fiction novel of near-future conflict between the United States and China. In 2033, China has toppled the government of the Philippines. The US Army is on the ground to support the Filipino government and military in its efforts to retake Manila. This is the setting for a vivid, detailed tactical depiction of the impact of artificial intelligence (AI) and robotics on warfare.

In its focus on the transformational impact of tech, the book is in the same genre as the novel Ghost Fleet. Indeed, the author thanks Ghost Fleet coauthor August Cole for encouraging him to write the short story that eventually became the novel.

The author, retired US Army colonel Brian M. Michelson, served in assignments around the world in the XVIII Airborne Corps, the 101st Airborne Division (Air Assault), the 97th Civil Affairs Battalion (Airborne), United States Special Operations Command (USSOCOM), and Joint Special Operations Command (JSOC). As a senior fellow at the Atlantic Council, he focused on AI, robotics, and warfare.

There is no consensus definition of “artificial intelligence,” which refers to software programs enabling machines to undertake tasks previously thought to require human intelligence. One of the principal approaches within AI is machine learning (ML). In ML, data is used to develop (“train”) models that, given new data, can predict or classify rapidly. The models can be continuously updated with new data, with or without human supervision, allowing adaptation. These programs have a wide range of possible military applications, including navigation for autonomous vehicles, targeting, logistics, or the identification of unusual patterns of activity.

Warbot 1.0 is a novel used as a delivery mechanism for a payload of ideas about future war. AI-enabled robots fight alongside humans or in groups or swarms on land, in the air, and in the sea. The book opens with an engagement between US and Chinese autonomous armed vehicles. The Chinese rely on robots disguised as shipping containers to guard their installations and to identify and target Filipino citizens. Drone swarms seek and kill in urban combat, while groups of autonomous armed submarines target ships at sea and overflying planes. AI-driven robots identify, target, and attack at a speed that no human could match.

AI is also used to rapidly integrate and process multiple streams of information, from which it identifies patterns and makes predictions. Soldiers have improved, real-time situational awareness of the battlefield via receiving information and giving orders using virtual reality helmets and special gloves or exchanging with AI intelligence programs in natural language. Chinese psychological operations, directed against the families of deployed US soldiers, are individually tailored for maximum effect. Ultimately, the book concludes that US victory is delivered by means of human-AI teaming—compared to the Chinese reliance on AI alone—and a superior AI intelligence system that was easier to train and learn, enabling faster adaptation.

Warbot 1.0 also delves into some of the second-order effects of these technological advances. The use of robots in combat reduces human casualties, inviting military adventurism and allowing the use of tactics that involve sacrificing units without qualm. Conflict escalates rapidly when AI systems react to other AI systems. Real-time information streams tempt senior officers to micromanage, undermining mission command. Humans wrestle with ethically fraught decisions about settings that control the robots’ level of autonomy and tolerance for civilian casualties. Humans anthropomorphize and develop emotional attachment to the robots with which they work. Humans must decide how much they trust their AI systems.

The book’s portrayal of the likely ways in which AI will transform war are broadly in line with current and emerging capabilities, even if the idea that all of these will be in place by 2033 is overly optimistic. Where the book is more fiction than science is in its portrayal of AI as working well and largely without error, even when predicting human behavior.

ML models are probabilistic, statistical models of phenomena developed from data sets. When the models are left to learn on their own, taking in data without human supervision, there is no guarantee of the outcome. If the data used to train the model is not representative of the phenomenon, the models may be biased or simply wrong. This is a recurring problem in facial recognition, for example, because most of the available images for training models are of men of European descent. Consequently, facial recognition is much less accurate for others. Also, if the underlying phenomenon changes in some important way, then previous data and models will no longer work well. This is why some stock market models were thrown for a loop by the COVID pandemic. Finally, humans can err in building models by failing to include important features; choosing the wrong algorithm; or making mistakes in data collection, structuring, cleaning, or coding. ML models work best when applied to concrete, clearly defined, stable phenomena for which there is representative data, and even then, they are sometimes wrong.

Warbot 1.0 does provide an example of an AI error due to novel input. It also acknowledges the probabilistic, statistical nature of machine learning models by imagining AI systems that communicate confidence levels for their claims and predictions, allowing humans to decide whether to trust that information. However, many of the things that would cause a ML model to be wrong—such as biased data, inclusion of the wrong features, or changes to the underlying phenomenon—would also make it impossible to calculate meaningful confidence levels. This, in turn, should have implications for human adoption of and trust in AI. The Warbot 1.0 vision gives AI too much credit—but maybe that is what is needed for a good story.

M. A. Thomas
Air Force Cyber College

Note: This book is also available on Kindle Unlimited.

 

"The views expressed are those of the author(s) and do not reflect the official policy or position of the US government or the Department of Defense."

 
Strategic Studies Quarterly (SSQ) and the Air & Space Power Journal (ASPJ)publish book reviews to inform readers and enhance the content of articles in the journals.