Air Mobility Command tries AI to speed up airlift planning

WASHINGTON — While activists and officials angst about “killer robots,” software developers and staff officers are quietly working on artificial intelligence tools for something far more likely to help win the next war: logistics.

Flying troops and supplies around the world takes complex planning that juggles dozens of variables. How many transport planes, of what types, will it take to move a given unit the desired distance? How much gas will those planes burn? Where should they refuel along the way? Do they need tanker aircraft to refuel them in flight? Should planners allow time for the flight crews to rest (which is slower) or have backup crews in place (more expensive) to take the plane on its next leg without a break? How many spare planes are needed if one breaks down? What backup bases should the planes re-route through in case of bad weather, cyber attacks, or enemy missiles shut an airfield down?

RELATED: ‘Largest conundrum of them all’: Air Force still unsure how to keep forces supplied in Indo-Pacific

That’s hard enough to figure out when planners have plenty of time. When imminent war or humanitarian disaster requires airlift ASAP, they have to literally wing it.

“That process is typically short-circuited, because whatever the crisis is, it’s in motion before the operational planners are able to complete their work,” said retired Air Force Gen. Paul Selva, who served as head of Air Mobility Command, then US Transportation Command and finally vice-chairman of the Joint Chiefs of Staff. To run the numbers properly, he told Breaking Defense in an exclusive interview, “the planners would tell you that takes a minimum of days. If they’re really fast, maybe hours. But they’re doing a lot of things with estimations and Kentucky windage.”

Two years ago, four-star chief of Air Mobility Command Gen. Mike Minihan approached his fellow transport pilot, Selva, and asked for help: Could the retired vice-chairman’s contacts in the AI world make planning easier and faster? After a deep dive with AMC planners into their processes and problems, Selva said, “my answer was a qualified yes” — and he co-founded a company, DEFCON AI, to build the software.

Last month, after two years and $7.5 million in Small Business Innovation Research contracts, DEFCON AI delivered version 1.0 of that software, the company announced today. Called ARTIV — which doesn’t actually stand for anything — the AI toolkit is currently available to up to 100 authorized users on GovCloud for use with highly sensitive but unclassified data (aka Impact Level IL-5). ARTIV should be deployed on systems using secret data (IL-6) within “weeks,” the company said.

Using ARTIV to help build an airlift plan, Selva said, “the run to get all of that data takes between five and 10 minutes.”

Keep It Simple, Stupid

The secret to this speed — both how fast the software was developed and how fast it can run — lies partly in the old military maxim, KISS: “Keep It Simple, Stupid.” ARTIV doesn’t calculate numbers to “20 decimal places,” Selva said, because that takes more processing time and military operations are never so precise in real life anyway. The program also doesn’t include any of the generative AI/LLM/chatbot features that have become so popular in the last two years, because they take vast amounts of data to train and a lot computing power to run. (Another military initiative, GIDE, has tested that emerging tech for logistics planning.)

In fact, the foundation of ARTIV isn’t artificial intelligence in the sense of machine learning at all, but deterministic algorithms doing what’s called mathematical optimization to calculate the best combination of planes, crews, routes, bases and so on. The AI element comes in after ARTIV creates that optimized “baseline air network.” At that point, another part of the program uses game theory and military intelligence data to guess what an intelligent adversary could do to disrupt the operation, then recommends backup plans to minimize those disruptions.

Users can choose whether to turn on the simulated-enemy feature or not. They can also tell the program whether they want it to maximize efficiency, using as few planes and as little fuel as possible; maximize speed, getting the given unit or cargo to the destination on time no matter what; or to maximize resiliency in case of weather, accident or enemy attack, which makes the software recommend robust but potentially costly backups, like extra planes and crews.

“There are real material savings in the numbers of airplanes and crews that are required to make moves happen, unless you turn the resiliency knob really high,” Selva said. “It’s literally a click of a mouse and you drag a slider.”

DEFCON AI is already working on upgrades to ARTIV, such as adding trucks, trains, and cargo ships to the current airlift-only planning. “That is much more computationally intensive,” Selva said. “The challenge to our engineers is to keep that below the 10-minute [threshold]. They’re still scratching their heads.”

Other upgrades might be easier. As currently configured, Selva explained, ARTIV is a tool for Air Mobility Command’s planning staff, not its Air Operations Center (AOC), which actually gives detailed directions to individual airplanes. But, he said, the software is already able to output its recommendations in an aircraft-by-aircraft format “that could be — but isn’t today — exportable into the air ops center’s tasking tools.”

That operations center, formally known as the 618th AOC, has already started its own digital initiative, “Project Nightmare” an AMC officer explained at a recent Potomac Officers Club conference. “Nightmare” replaced phone calls from staffer to staffer with chat messages that the 618th and its partners at MIT could then automate and use to train AI. The 618th is also using commercial open-source intelligence tools like Dataminr to track emerging crises around the world, so staff can start preparing options before the orders come to act.

Indeed, ARTIV is just one piece of a much larger push to move Air Mobility Command harried staffers yelling into phones and laboriously retyping data to a streamlined, digital and AI-enabled world.

“ARTIV is one part of an entire ecosystem of integrated operational planning tools that AMC is developing to sense globally, predict where we will be needed, and seize opportunities on behalf of the combatant commands,” said Maj. T.J. “Fever” Britt, of AMC’s Commander’s Initiative Group, in an email to Breaking Defense. “This will be a tool we integrate for crisis planning that is central to the operational wargaming. We are optimistic about the potential for the technology that underpins ARTIV.”

AMC’s effort, in turn, is part of an all-service overhaul of command-and-control systems worldwide, known as CJADC2. While AI hype has focused on generative AI replacing humans and robotic weapons killing them, the Pentagon’s focus has been on streamlining unglamorous staff processes, so commanders can make better-informed decisions in much less time. Having a faster “decision cycle” across the force can be a more decisive advantage in real war, even when the other side has technically better weapons, as shown by examples from Korean War dogfights to the Nazi blitzkrieg of France.

“If the process of planning is exponentially faster,” Selva said, “the process of communicating the recommendations and the course of action to the field force carry that same level of speed … which makes it very useful in a crisis.”