News

2021-06-01

Artificial Intelligence and Military Deception

By:Nikolas Gardner ,UAE National Defense College

Throughout history, deception has been an integral part of warfare. Ancient Greek accounts of the Trojan War describe the infiltration of the city of Troy by Greek soldiers hidden inside a wooden horse. Around 500 BCE, the Chinese strategist Sunzi advocated the use of deception to achieve victory in The Art of War.

Deception played an important role in the wars of the twentieth century, despite technological advances that enabled intelligence analysts to monitor closely the movements and communications of their adversaries. In 1941, the Japanese Navy concealed its impending attack on Pearl Harbor from American intelligence services by leaving the radio operators of its aircraft carriers in Japan while the ships sailed towards the Hawaiian islands.

Monitoring the ongoing communications of these operators, American analysts assumed that their ships remained in port until the attack occurred. (Wohlstetter: 1962) In 1944, the Allies executed an elaborate ruse to distract the Germans from their preparations for an invasion of Normandy, on the French coast.

The Allies erected fake military camps opposite the Pas de Calais, northeast of Normandy, and created radio traffic referring to a non-existent American army group supposedly inhabiting them. Convinced they knew the location of the impending invasion, German pilots and radio operators focused their attention on the Pas de Calais, even after the Allies began landing forces in Normandy.

In 1973, the Egyptian armed forces took simpler, but no less effective measures to conceal their preparations to cross the Suez Canal from Israeli intelligence services. To account for the movement of personnel and military equipment prior to the crossing, the Egyptians announced that routine large-scale military exercises would take place at the time of their planned offensive.

Suspecting that Israel maintained informants within their armed forces, the Egyptians did not reveal the real plan even to the units involved until days, or in some cases hours before the crossing began. (Sheffy: 2007)

In the twenty-first century, however, technological advances have seemingly made it more difficult to conceal the movement of military forces, or to create a false impression of military activity where none exists.

In addition to aircraft and unmanned aerial systems (UAS) equipped with high-resolution cameras, armed forces possess sensors capable of detecting electromagnetic and thermal emissions of equipment and personnel.

They also have access to large volumes of open-source information such as satellite imagery and social media posts. Moreover, artificial intelligence (AI) capabilities will soon give armed forces the ability to integrate and process huge volumes of data from these diverse sources in order to identify military targets.

These developments support the bold prediction of U.S. Air Force Chief of Staff General Ronald Fogleman, who claimed that in “the first quarter of the 21st century, it will become possible to find, fix or track, and target anything that moves on the surface of the earth.” (Geist and Blumenthal: 2019)

But despite its advanced data processing capabilities, AI is still vulnerable to deception.

While AI can be trained to identify specific images, it does so differently than humans. A human analyst examining an image of an aircraft would rely on their previous experiences as well as their understanding of the context in which the image appears in order to identify it.

A machine, on the other hand, would identify the image based on the configuration of the pixels that comprise it. But if this configuration is altered even slightly, AI can mistakenly identify an image in ways that a human never would. In one case, scientists at MIT fooled a machine into identifying an image of a turtle as a rifle with only minor alterations to the image. (Knight: 2019) In a real conflict it is conceivable that a machine could be fed manipulated images that cause it to overlook enemy personnel and equipment, leaving its own forces vulnerable to an unexpected attack. Alternatively, it might identify targets that do not exist. This could lead it to initiate attacks against “phantom targets”, which would reveal the location of its own forces and expose them to counterattacks. (Pikner: 2021)

Historically, many of the most famous deception operations have succeeded because they adhered to “Magruder’s principle”, which maintains that deception is more easily achieved by reinforcing an adversary’s pre-existing beliefs, rather than trying to change those beliefs.

For example, Allied deception succeeded in 1944 in large part because the Germans expected an invasion at the Pas de Calais rather than Normandy. This principle remains relevant even in a world in which AI-based systems identify targets using data from diverse sources.

Comprehending how these systems perceive the world is equivalent, or nearly so, to understanding the pre-existing beliefs of human adversaries. Equipped with this knowledge, it may be possible to manipulate and deceive even machines capable of processing data far more rapidly than humans. Thus, it is highly unlikely that deception will disappear from the conduct of war.

Sources:

Geist, Edward and Marjory Blumenthal. “Military Deception: AI’s Killer App?” War on the Rocks 23 October 2019.

Knight, Will. “The Fog of AI War”. MIT Technology Review 122/6 (2019), 44-49.

Pikner, Stephan. “Leveraging Multi-Domain Military Deception to Expose the Enemy in 2035”. Military Review (March-April 2021). 81-87.

Sheffy, Yigal. “Overcoming Strategic Weakness: The Egyptian Deception and the Yom Kippur War”. Intelligence and National Security 21/5 (2006), 809-828.

Wohlstetter, Roberta. Pearl Harbor: Warning and Deception. (Stanford: 1962).

Comments () Views (8861 ) Print

Send To Friend