WASHINGTON — The Army’s future capabilities lab is developing an artificial intelligence tool to help mission planners choose the best moves in multidomain battles.
The C5ISR Center tool, called the Artificial Intelligence Course of Action Recommender, allows soldiers to set their mission objectives and priorities, but it then suggests the best action to take.
The mission planning assistant will help as the military moves toward operations that span electronic domains, space and more dispersed geographic locations.
Battlefields are already enormously complex for commanders as they make quick decisions in the heat of battle. That complexity will only grow under the military’s unified war-fighting concept of Joint All-Domain Command and Control, in which any sensor connects to the best shooter across domains.
“That’s going to explode as we move towards MDO [multidomain operations] and JADC2. We’ve got multiple domains now that the mission planners need to consider, so it’s not just land, but also air, maritime, cyber and space that are all interconnected, much tighter than they have been in the past,” said Pete Schwartz, an AI expert who works on the action recommender tool with research and development partner MITRE. “It’s only going to become a more complex challenge to overcome, more combinations of decisions to consider, and shorter timelines in which to consider them.”
The development team is primarily working with the Mission Command Battle Lab on the tool in the proof-of-concept stage, but it also has reached out to the Army AI Task Force, the Network Cross-Functional Team and other Army components that could have an interest in the capability.
So far, the mission assistant is largely focused on land operations, but the C5ISR Center plans to expand it. The tool uses a ground combat system simulation, called the division exercise training readiness system. Units at the division level or below could use it.
“It is not a multidomain simulation, [but] that is what we will need in order to move beyond just the land domain and to make this a true multidomain capability, which we hope to do in the future,” Schwartz said.
As the team works to further the tool, Dan O’Neill, chief engineer of the C5ISR mission command capabilities division, said adding the cyber domain into the simulation poses one of the biggest challenges.
“Coming up with accurate models for the effects of cyber and electronic warfare vectors is really, really difficult,” O’Neill said. “You can think of a cyberattack taking down an entire operation instantly, but the chances that happens is very rare. So we have thought about developing proxies, even inside of our simulation to do some of that.”
For example, the team could mimic the effects of cyber and electronic warfare by manipulating combat power in the simulation or reducing systems’ ranges.
Another challenge for all artificial intelligence programs across the Defense Department is ensuring that tools explain how a decision was reached so users trust the systems. That challenge is amplified for a tool meant to help commanders plan missions.
“The user can select any one of those courses of action,” Schwartz said. “They can actually view a playback of the mission, just like you would watch a video, and they can speed it up, or they can pause it. So they can see exactly ... where the units are moving across the map and how they’re engaging with each other. And then they can decide whether they trust the output from the simulation in that instance.”
To accomplish this for the Army’s tool, the system tests different actions, showing the user a graph comparing the performance of those options with the original plan input by the commander.
Over the last year, the artificial intelligence system demonstrated that it could find improvements in commanders’ action plans. This year, the team is looking to make the tool more interactive, allowing commanders to work on more focused parts of a mission, Schwartz said. The system will highlight critical points in the mission with risk of failure, such as when military vehicles need to arrive on time at a specific location or when a unit would begin to run out of fuel.
“In the future, we are hoping that we can develop the artificial intelligence to a point where it understands the cause and effect enough that it can start to automatically extract some of those critical points and highlight them for the user,” Schwartz said.