Can AI predict if a Marine will quit? Corps wants to know

The U.S. Marine Corps isn’t for everyone, but there’s no real process for weeding out folks who sign the enlistment paperwork but are unlikely to make it through their four-year contract. A new experiential initiative uses artificial intelligence to predict whether a Marine recruit will complete their full term and, more importantly, pinpoint the factors that might get in the way of that goal, said the AI lead for the U.S. Marine Corps.

“We are using artificial intelligence to look at how we can recruit Marines, and we're looking at how we can retain them specifically,” Capt. Chris Clark said during a Defense One Genius Machines event that aired Sept. 25. “We can assess what attributes might…first predict whether that Marine will complete that four-year term or not, and then go back and look at what attributes might be impacting that prediction. So then we can inform the recruiters. And they can use that to better prepare these Marines in different ways so that they can go out into the Marine Corps, do…great and amazing things, complete their four-year term, and then decide if they want to stay in or if they want to go on and do something else.”  

The Marine Corps doesn’t regularly publish figures for Marines who leave the service before their contract is up. But in 2017, the number was 5 percent of the entire active component, comprising 27 percent of all departures. 

The Corps has started offering new incentives to retain their people, particularly after their contract ends. But there are a lot of other factors that can contribute to quitting—things like fitness, exhaustion, personal or family stress, and other factors that wouldn’t be affected by the promise of a bonus check. It’s data the Marines can start to collect during recruitment, to predict the likelihood of recruits finishing a four-year tour. 

“How can we understand a Marine when they're a poolee, so before they go to boot camp and become a Marine, how can we use different factors to be able to understand if they're going to complete their first fully four-year term, right?”

The “retention prediction network,” as Clark calls it, is still in an experimental phase. The use of artificial intelligence in employee recruitment and employee management is a controversial area, ripe with potential for lawsuits. But officials in the military and national security space have had the opportunity to experiment with AI for management for years to detect or predict emerging insider threat issues that could affect national security. 

A 2019 experimental program from the Defense Security Service sought to help managers better predict when an employee might be facing problems—not to push people out of the service, but to better anticipate when an employee is having challenges that affect their performance, so that a conversation could occur early, before the resignation letter.

The value of the Marines’ prediction network is similar, Clark said. It’s not a replacement for human judgment, but a possible tool to allow recruiters or trainers to see something they might be missing. 

“You know, a prediction is a prediction. It's not always right, but what it does is it gives us information that we can then use to inform those recruiters where they, you know, currently, don't have a lot to go on,” he said.

The Marines are using a similar tool to better plan exercises and operations— a large language model specifically applied to the vast repository of text  the Marine Corps has amassed on various engagements over the years. The hope is to create a tool that can provide “critical information for Marines to make decisions on how to plan that next exercise, operation, deployment, whatever it is that they're doing. They're able to take that data and do a lot with it, whether it's summarizing decades of information into a concise report, or just find key pieces of information and trends within that data,” he said.