ABERDEEN PROVING GROUND, Md. (March 21, 2017) - When the fight commences, the commander shifts the military decision-making process into high gear.

While the commander’s years of experience often dictate the pace of this process, even the most battle-hardened leaders must operate within the confines of the human brain’s cognitive capabilities. What if technology could automate routine, time-consuming tasks to shave minutes, or even hours, off the decision-making process? DOD thinks it can and is evaluating overarching concepts that would incorporate autonomous systems and artificial intelligence (AI) into its future combat warfare missions.

AI is a component of the Pentagon’s third offset strategy, designed to obtain strategic advantage by outmaneuvering adversaries through advanced technologies. The first offset strategy began with the development of a nuclear arsenal during the 1950s Cold War era. The U.S. implemented the second offset strategy in the 1970s and 1980s, when it introduced stealth systems, precision-guided weapon technologies and GPS. The third offset strategy, according to Deputy Secretary of Defense Bob Work, will “lead to a new era of human-machine collaboration and combat teaming.”

In a mission command construct, which integrates the warfighting functions of movement and maneuver, fire support, sustainment or logistics, reconnaissance, surveillance and intelligence, AI systems must “learn” to communicate the commander’s intent. But only by establishing parameters within which the technology can operate will it gain the commander’s trust. The goal is to allow AI to process certain tasks and functions that are data-heavy or generally repeatable and then rapidly suggest courses of action (COAs) to the commander, who ultimately will make the final decision.

As the Army’s applied research and advanced technology development organization for mission command capabilities, the U.S. Army Communications-Electronics Research, Development and Engineering Center (CERDEC) understands both the art and science of mission command and believes that AI will have considerable implications for mission command capabilities. CERDEC is closely monitoring Army strategy along with AI advancements in industry and academia as it evaluates options to produce mission command systems that communicate the commander’s intent and provide military users automated assistance with planning, monitoring and decision making.

Systems achieve different levels of autonomy based on the amount of human-to-machine interaction required to operate them. Autonomous technologies are further categorized as either “autonomy at rest,” which are software agents or the brains behind a system, or “autonomy in motion,” which represent actual platforms, such as driverless vehicles and robots.

At the lowest level of autonomy, “command by directive,” systems require a one-to-one human controller. Each autonomous machine is controlled by joystick using high bandwidth/low latency communications within a reliable communications range. In mission command systems, command by directive applies to autonomy at rest, because the user must tell software agents exactly what to do through every step of the process to achieve the objective. Most software and robotic systems operate in this manner; for example, the explosive ordnance disposal robot is joystick-controlled.

“Command by plan” provides autonomy at a mid-level, where scientists specifically instruct the software on when, where and how to carry out tasks that alleviate the burdensome direct joystick control. For example, CERDEC’s mission command experts explored autonomous route-planning technologies for intelligence, surveillance and reconnaissance missions. The operators identified an objective and instructed unmanned aerial and unmanned ground vehicles to maneuver to certain waypoints on a map. The software was able to determine how best to reach those waypoints; e.g., to avoid certain obstacles. The CERDEC technology developed under this program was a first step at moving from command by directive to the higher-level command by plan.

In another example of mid-level autonomy, CERDEC is gathering user feedback that it will use to develop an automated planning framework prototype that will allow computer scientists to use AI technologies to assist in mission planning, monitoring and prediction. This framework would allow commanders and staff to run through the military decision-making process, which includes COA development and analysis for maneuvers to identify who will go where, when and how; logistics to convey how much fuel, ammo or water will be needed, and when to refuel. Eventually the system would support all other warfighting functions, such as fires and intelligence.

To support the logistics portion of this plan, CERDEC has developed the Energy Aware Mission Planning Tool, or E-AMP, which would select supply routes, understand what vehicles are traveling along those routes, assess weather and other environmental factors and calculate how much fuel is needed to support a COA, allowing a logistics officer to design a concept of support. In the long run, AI can extend the capabilities of this software automation by performing reasoning and optimization to provide useful recommendations, such as how and where to conduct resupply and which assets to use.

In each of the above scenarios, developers still program specific tasks into the system. To achieve the highest level of autonomy, “command by intent,” programmers do not tell the computer what to do; they instruct the computer on how to figure out what to do. In this scenario, the system is provided only with the objective, such as to patrol and secure an area. It does not require further instructions on how to reach map waypoints or any other tasks required to pronounce an area secure.

It is imperative to understand that CERDEC’s efforts will use AI to augment human capabilities—not replace them. Humans and machines work best in combination when they complement each other using what some have called the centaur model, which pairs machine precision and reliability with human robustness and flexibility. In that model, commanders evaluate input collected with AI but make the final decision about what COA to take. CERDEC is teaming with industry, other government research and development entities and academia, and conducting internal working groups to understand this human-to-technology paradigm to organically grow its AI expertise.

For example, CERDEC, in conjunction with Carnegie Mellon University, participated in an Army technology objective that concluded in 2012 and focused its AI efforts on enhancing three mission command principles: build cohesive teams, create shared understanding and reinforce the commander’s intent. Commanders and staff naturally increase their use of automation when communicating instructions, coordinating resources and gathering information for analysis. To assist the commander with analyzing staff behavior and effectiveness, which could be two levels down, researchers conducted forensics on staff digital conversations—e.g., text and chat—to identify key actors, locations, resources, patterns of interest, hidden connections and spheres of influence from within the organization. This information helps to identify how well the staff is, or is not, synchronized, to aid the commander in propagating his intent throughout the organization.

One AI technology, machine learning, will significantly aid mission command by allowing computers to provide better answers as they are exposed to more data. Machine learning draws from big data across all of the Internet—every Facebook and Twitter post, every image, coupled with high-performance computing—which allows the computer to process billions of lines of code to find new correlations between data sets to complete complicated tasks in nanoseconds.

CERDEC is exploring options to apply machine learning to intelligence systems, which would speed up the process for finding new correlations between data sets, such as learning new patterns and recognizing images. Machine-aiding tasks and asset collection would maximize the value of reconnaissance and surveillance information while minimizing the risk to the mission and friendly forces.

To aid the Army’s cyber defense objectives, CERDEC is evaluating how AI and machine learning could improve anomaly detection and possibly incorporate a response inspired by the human immune system, which would allow a system not only to recognize when a malicious virus is attacking it, but learn to repair itself.

Another area that is primed for AI support is position, navigation and timing (PNT). CERDEC is currently supporting the Army’s priority to develop assured PNT technologies, which provide Soldiers and autonomous systems the capability to conduct operations in GPS-denied or -disrupted conditions. PNT solutions comprise an assortment of sensors such as inertials, which use accelerometers and gyroscopes to measure position orientation and velocity, and cameras to shoot and compare pictures frame by frame to determine movement. These combined sensors aid GPS, but some may be more reliable in certain situations.

Machine learning would insert intelligence into the system to autonomously determine which sensors are the most trustworthy. For example, AI could provide Soldiers better navigational information based on its understanding of how other squads crossed terrains under specific environmental conditions or times to suggest an alternative route that would be faster and safer. Soldiers who may be relying on a camera for position information may not realize that the sun is interfering with the camera, so AI would ignore that sensor (the camera) to ensure that the most accurate sensors provide the required information.

To ensure alignment with Army strategy, CERDEC plans to support the U.S. Army Tank Automotive Research, Development and Engineering Center, the Army’s autonomy lead, as it incorporates manned-unmanned teaming into its initiatives. CERDEC also plans to support the U.S. Army Maneuver Center of Excellence, the Mission Command Center of Excellence and sister science and technology organizations to develop technologies and approaches that incorporate all aspects of autonomous systems into mission command systems and doctrine.

CERDEC, as the Army’s mission command research and development arm, is working to grow its AI and machine learning expertise and rapidly integrate advances from industry and academia into effective mission command systems.

By introducing these technologies into the mission command toolset, systems will achieve the highest level of autonomy to allow decisions at the point of action—which in turn will minimize the commander’s cognitive burden to enable fast yet coordinated decisions. Communicating the commander’s intent to autonomous entities, both at rest and in motion, letting computing systems act on our behalf and trusting automation to provide recommendations and assistance where appropriate will act as force multipliers and ultimately provide the U.S. and its allies with overmatch capability.

This article originally appeared in the Army AL&T News and is scheduled to be published in the April-June 2017 issue of Army AL&T Magazine.

MR. JAMES HENNIG is the associate director for systems engineering with CERDEC’s Command Power and Integration (CP&I) Directorate. He is a Ph.D. candidate in systems engineering at Stevens Institute of Technology, and holds an M.S. in software engineering from Monmouth College and a B.S. in mechanical engineering from the University of Toledo. He is Level III certified in systems planning, research, development and engineering.

DR. PETER SCHWARTZ is a lead enterprise systems engineer for the MITRE Corporation, supporting CERDEC’s CP&I Directorate. He holds a Ph.D. and an M.S. in computer science and intelligent systems from the University of Michigan and a B.S. in computer science and a B.A. in psychology from the University of Maryland.

MS. KATHRYN BAILEY is the public communications adviser for Decision Engineering Inc., assigned to CERDEC’s CP&I Directorate. She holds a B.S. in communications from the University of Maryland University College.