Groundbreaking tools are around the corner, 3-star general says. The limiting factor is lack of computing power to grind through giant Pentagon datasets.
Many military applications for artificial intelligence — from grand strategy to supply-and-maintenance management to predicting incoming fire for platoon commanders — aren’t far off, says the new director of the Pentagon’s Joint Artificial Intelligence Center. But Marine Corps Lt. Gen. Michael Groen says AI still faces resource challenges and hopes that incoming defense leaders will continue to prioritize the fieldit as budgets face new headwinds.
Groen said AI will lead to a revolution in defense decision-making, particularly by combatant commanders.
“They’re the people who are going to be awake at night when something weird is happening in Belarus,” Groen said in his first one-on-one interview since becoming the JAIC’s second commander in October. “There’s a combatant commander who’s on the hook for reacting to that. Shouldn’t we enable those combatant commanders with data-driven, clear pictures of what’s happening on the ground? “What’s happening in the information space, tipping and queuing in an automated way, natural language processing and then being just as cognizant of ‘What’s blue doing? What tools do I have to react to that?’
“Today we do all of that, combatant commanders, great decision-makers all,” Groen said — but too much remains “seat-of-your pants.”
While seat-of-your-pants decision-making may sound rapid and agile, it can also mean delay and misuse of resource for the troops who are waiting for orders, Groen said.
“That’s how you wind up with a lance corporal sitting in the middle of nowhere going ‘ok what idiot told us to go here and why?’” he said. “That's the result of seat-of-the-pants decision-making based on imperfect information.”
The cure for that is for military leaders to get “much better at using data to drive the decisions that we actually have data on.” This kind of data already exists — and if it could be collected, structured, and presented more coherently, Groen said, it might answer questions such as “’What is the status of your unit? What is the status of your reserve? Is that unit really ready to go or not? What’s the status of your logistics? What’s the status of your medical [evacuation] capability; can you meet the Golden Hour?” he asked, referring to the best window of time to get an injured operator to medical care.
Groen sees military AI as little more than the wise application of well-known and common machine learning and methodologies to vast stores of Department data. This means, he believes, that AI can begin to change planning, strategies, and operations today.
AI tools aren’t just for combatant commanders. Groen thinks it’s long past time to give platoon and squad leaders battlefield options and services that they enjoy in civilian life in apps like ride-sharing app Uber or navigation app Waze. “Shouldn’t I be able to pull up…map data the way I’m riding home in my car? Shouldn’t I be able to do automated navigation or assisted navigation? Shouldn’t I be able to click on things that [are] ‘artifacts of the battlefield?” he said. “So if you’re doing a [helicopter] linkup…want to make sure you have resupply because you’ve been in a place for a long time. That kind of stuff can be data-driven so you don’t even have to ask. You report and things happen. The logistics start flowing to you.”
And even before AI changes soldiering, Groen expects it to change less glamorous Defense Department activities. “The business case stuff, it’s not as compelling as the warfighting stuff but it’s closer and it's easier. The data is more available.” Activities away from combat are also the sort of thing that a wider variety of companies, like Google, may want to help with.
Getting the military to where it needs to be to fully use the data it has as part of an AI-first enterprise won’t be cheap, he says. And the JAIC, just like every data-based enterprise, is constrained by the amount of sheer computing power it has to throw at problems.
“Training an algorithm is not that expensive,” but it’s not free, Groen said. JAIC could move faster to exploit the Pentagon’s large datasets if it had access to more computers, and particularly graphical processing units. “The cost is a limiting factor, certainly a limiting factor for the JAIC. And we don’t have that big of a budget,” he said. “We have a capacity issue, like how fast can we get to all of the people who need AI across the Department? We were going to run as fast as we can but our duty cycle is going to be… we’re going to be working out butts off.”
Groen hopes future Defense Department leaders appreciate how AI could change much of what they do. There’s some indication that will be the case. Michelle Flournoy, widely regarded as president-elect Joe Biden’s likely Defense Secretary pick, has been a vocal supporter of robust investment in AI.
“I think with leadership from the top within the Department that’s really focused on: ‘Okay, we agree that artificial intelligence and information-age decision-making can really facilitate our transformation and it’s necessary.’ With leadership that’s focused like that, we’ll get to where we want to go,” he said.