Airmen and Guardians working side by side with researchers in the field of artificial intelligence at the Massachusetts Institute of Technology are part of a unique military unit that’s helping to steer some of the research studies while also attempting to guide the Air Force’s and Space Force’s wider adoption of AI.
Under the new research partnership, the Department of the Air Force and university jointly decided on 10 research projects to focus on, a departure from the typical top-down style in which the department advertises grant-funded topics, said Col. Tucker Hamilton, the Air Force’s director of the DAF-MIT AI Accelerator. Started in January 2020, when projects got underway, the accelerator also differs because it doesn’t focus solely on developing a military capability—it also benefits “the public good,” Hamilton said.
In fact, Hamilton doesn’t consider AI to be a capability at all but instead a “tool that enhances all other capabilities.” The work that the accelerator’s 12 Active-duty troops and four Reservists are taking part in is “meant to further the science of AI” in ways to be broadly applied—“not just in some military sense,” he said. “Everything that we decided on with them to pursue had to have a use for a military application as well as a use for a societal application.”
The research projects involve about 140 faculty members, researchers, and students from MIT and the federally funded, national security-focused MIT Lincoln Laboratory. The accelerator’s director on MIT’s side is Daniela Rus, who also directs MIT’s Computer Science & Artificial Intelligence Lab and served on the Defense Innovation Board advising defense secretaries.
From Hamilton’s perspective, “AI is ubiquitous right now,” he said. “Everything is being influenced by machine learning. So how do we, as a military, approach the technology?”
How it Works
Hamilton was an F-35 test pilot before becoming an MIT fellow with the Air War College, where he learned about the accelerator. Having the qualifications to serve as a program manager, he applied. The Airmen and Guardians assigned to the accelerator report to Hamilton, “so we work together as a military unit,” but they also embed with research projects.
Officers and enlisted members assigned to the accelerator full time come from career fields that complement the projects but also bring some prior understanding of machine learning, Hamilton said. They’re pilots, in part, but also weather, intelligence, and cyberspace operations officers as well as analysts in geospatial intelligence and operations research among others.
“We tried to, first and foremost, find the right people because right now, there are only a handful of people that truly understand this,” Hamilton said.
Three of the four Reservists taking part in the accelerator fulfill a special role. They’re “hugely important for this because,” as CEOs in their civilian lives, “they’re the ones that are actually running AI companies,” Hamilton said. “They have the ability to understand this technology more than most people in the military can understand it.”
The service members each embed with one or two of the research projects, ranging from the likes of AI-assisted autonomy for safe decision-making, optimization of training schedules, and personalized instruction in a foreign language.
“They help actually write some of the code, and they give [the researchers] perspective—like, ‘Well, this is how a pilot would use this type of technology in the field.’ Or, ‘This is what a pilot would be thinking’—or any kind of operator,” Hamilton said. “Our C-17 pilot—he’s working on our project that helps pilot training students.”
For the researchers’ part, “It gives them a vast amount of clarity on their efforts, on their research, on the direction that they’re moving—and also motivates them because they see an [eventual] outcome,” Hamilton said. “They see something that is like a practical application, which excites people.”
Space Force Capt. Jazmin Furtado works with the accelerator part-time from across the country at Los Angeles Air Force Base, Calif., looking into how existing AI research relating to space domain awareness could benefit the DOD, “especially the Space Force,” she said.
An Air Force Academy graduate, Furtado went to MIT for her master’s and eventually was assigned to Kessel Run, the Air Force’s software development hub, as a portfolio lead during its early pursuits of AI—where she was one of a handful of people directed to “do AI.”
“It was a very vague statement, but it was a very big goal and vision,” she recalled. “A lot needed to be done in terms of, ‘How are we collecting data?’” That was also when she first connected with the accelerator.
Now on the heels of a fellowship at SpaceX, she’s applying all that experience as a program manager in space command and control architecture for the Space Force’s Space Systems Command. She’s focused on “overseeing these enterprise data stores” and envisions helping to build a digital environment that’s already optimized for AI, which all relies on quality data. “In order for it to be actionable, it has to be accessible,” Furtado said.
Alongside helping to “mold” the research projects, the accelerator is also “accelerating the empowerment and implementation of machine learning”—a branch of AI—“throughout the department,” Hamilton said.
The work includes documenting methodologies for the wider adoption of AI—“the frameworks that are going to allow our Airmen and our Guardians to create machine learning solutions for their own organizations,” Hamilton said—as well as providing education courses taught by MIT personnel.
Hamilton believes that in the long run, AI will be best at “teaming with humans.”
“Maybe it is in a situation where you have a pilot flying, and they are being fed pieces of information that the computer is seeing that the human couldn’t decipher that helps them and enhances their performance,” he said.
In hopes of guiding acquisition organizations in their use of AI, for example, a group of Air Force and Space Force acquisition program managers are assigned as fellows to the accelerator for four months—dubbed “Phantoms”—and developing the first toolkit-style document describing, “This is how you need to think about machine learning when you acquire it, when you contract for it,” Hamilton said. “How should you think about this technology when you’re diving into it—when you’re trying to hire industry partners to solve a problem you have using machine learning? What are the things you should be thinking?”
Courses have ranged from a senior leaders’ course for general officers; to leaders’ courses for the likes of squadron commanders and civilian government executives; to a coders’ course “that’s very intense … for a select few,” Hamilton said. “We’re trying to teach people organic ability to code and to create machine learning algorithms and go through data.”
Thanks to such a close partnership with “a world-class academic institution,” Hamilton said, “We’re making advancements revolutionary to the entire field—the entire world.”