EL SEGUNDO, Calif. (AFNS) —
What, exactly, is generative AI and how can it improve Space Force Guardian readiness in an era of Great Power Competition?
These were the foundational questions driving the inaugural U.S. Space Force Generative AI Challenge, an eight-week combination of AI crash course and problem-solving hack-a-thon that engaged 350-plus participants, delivered 11 learning sessions, and produced three finalists that will present their generative AI solutions — and compete for first place — at this year’s Space Power Conference, Dec. 10 – 14 in Orlando, Florida.
As AI applications like ChatGPT gain popularity among a general populace looking to save time and effort on everyday tasks, the Department of Defense is embracing the use of more specific and/or sophisticated iterations across a range of applications. Recognizing that the rapid pace of change in this area requires the DoD to be agile and adaptive in its approach, DoD’s current focus is on experimentation and rapid deployment to achieve the best solutions for the nation.
Enter the Space Force Generative AI Challenge
What began as an idea at the senior leadership level developed quickly and organically through grassroots volunteers including Capt. Jake Schneider, chief of portfolio, platform, at Space Systems Command based in El Segundo.
“Word kind of spread through the grapevine that ‘hey, we want to do a Space Force GenAI challenge’ and as someone with previous AI experience, I was in,” Schneider said. “Before you know it, this network of volunteers kind of happened and we started making the phone calls and getting the right people in the right spots.”
As the program evolved and readied for launch, organizers recognized that interest in AI within the ranks extended well beyond super coders and other like-minded “techies.”
“A lot of people had very little background in AI but they had a huge interest and wanted to learn more,” Schneider said. “So, we organized an educational series that happened one or two times per week over Teams where we brought in experts from the field.”
Some of the sessions were introductory. “We’re talking, ‘explain it to me like I’m five,” Schneider said. “What is generative AI? What are the differences between AI and Generative AI? How can we use AI to automate tasks, improve decision making, or create content?”
Other sessions delved more deeply into the technology itself, covering the mechanics of retrieval-augmented generation (RAG), open-versus-closed source query platforms, and the computational processing power required for more advanced applications.
“In one way or another, we were all there to learn,” Schneider noted.
The hybrid-virtual and hack-a-thon like competition kicked off on Sept. 12, with in-person events and live-streamed feeds from El Segundo, Washington, D.C., and Colorado Springs, Colorado. Registered teams were challenged to create solutions to Space Force problem statements leveraging generative AI. Problem sets spanned a wide array of disciplines, from mission-focused topics like space battle management and launch to general business administration and quality of life issues such as childcare.
Participants were equally varied, representing DAF, OSD, SAF, JS/J6, CYBERCOM, SPACECOM, INDO-PACOM, CENTCOM, SOCOM, NRO, MDA, DARPA, NSDC, Army Futures, AFRL, the Marine Innovation Unit, FORSCOM, and Army DEVCOM.
“We had participation from just about every branch of the military,” Schneider said. Military, civilian and contractor personnel, with some participating from as far away as Ramstein Air Base, Germany.”
At the end of the eight-week challenge, three proposals were elevated to semi-final status. Two of the three finalist teams are from SSC: they include members of SSC’s Atlas Office and Space Launch Delta 45. The third team comes from U.S. Air Forces Europe – Air Forces Africa. All three are headed to the Space Power Conference in Orlando for selection of the final winner. And all three demonstrate the wide range of applications for generative AI solutions, from health and fitness to data management to language translation. Here’s a quick re-cap of the three:
The generative AI solution developed by the Atlas team customizes fitness plans to the specific needs of Space Force members, taking into account unique factors such as irregular shift and work schedules, remote deployments, and limited food options. The app aligns with Space Force Fit goals and would include continuous tracking and appraisal of fitness programs customized by location and integrated with Guardian Resilience Team training.
For example, a senior master sergeant recently PCS’d (permanent change of station) to Colorado Springs and adjusting their cardio levels to an additional 6,000 feet of altitude could use Atlas’ proposed gen AI solution to get recommendations for local hiking and running trails. A lieutenant coming back from a knee injury could use the gen AI app to generate a workout plan that focuses on rehabilitation and allows their physical therapist to monitor recovery progression. Or, a new Specialist 2 living in the dormitory and limited to a meal card plan, could use the app to review the their dining facility’s daily menu and incorporate it into a holistic nutrition plan.
The team from USAFE-AFAFRICA tackled a more specific challenge: how to more quickly and efficiently translate communications in NATO’s official language (English and French) to accommodate the 28 languages and 32 dialects native to NATO members – a critical need, they reported, as many member countries do not distribute important announcements from the U.S. unless it is in their native tongue. Currently, each translation must be done individually by a linguist with specialty in a particular language.
USAFE’s generative AI solution would leverage BABEL, a free and open-source JavaScript transcompiler, to facilitate high-quality, secure, and multilingual translations for rapid communication with NATO allies. It would be the first-of-its-kind secured translation capability, featuring a simple, responsive design that would securely generate information via DoD infrastructure, reduce linguist man-hours to seconds, and replace the need for commercial software licensing at a cost savings of approximately $100,000 per squadron.
Rounding out the three semi-finalists is the Velociraptor team from SSC’s SLD 45 in Florida. This team’s generative AI solution addresses the current and future reality of launch, in which personnel numbers remain static but launches, particularly commercial, grow exponentially. More launches mean more data points to compile, review, analyze, and archive.
SLD 45’s solution leverages Mattermost, an open-source, self-hostable online chat service, coupled with a tactical data cache. Its capabilities include real-time audience transcription with generative AI summaries that capture and transcribe live on-console communications. The team cited four key benefits: AI-powered automation reduces manual tasks and enhances mission efficiency; real-time transcription with keyword detection improves situational awareness; automated crew logs and after-action reports streamline after-action reporting; and applications can be used for training and skill development.
source: www.spaceforce.mil