The U.S. Space Force is looking to artificial intelligence/machine learning (AI/ML) to aid space domain awareness and other mission areas, a top service official said on May 24.
The Space Operations Command’s transformation organization, led by Royal Canadian Air Force Brig. Gen. Kyle Paul, “is working through how do we implement a generative AI type of capability inside the headquarters to help us get after our business practices,” Space Force Lt. Gen. Stephen Whiting, the head of Space Operations Command, told a Mitchell Institute for Aerospace Studies’ Spacepower Forum. “But we’ve also established a ‘Top 10’ AI/ML needs list across all our operational missions and working with partners like Space Systems Command and Air Force Research Lab. We’re working to field capability, and we’ve seen some of that. Some of that’s in our space domain awareness system where we have a lot of data and AI/ML can help us parse through that data. We’ve also seen it in some predictive maintenance activity, predictive maneuver type capability. We really are working hard to put practical instantiations of AI/ML across all our mission sets.”
While AI/ML hold promise for military operations, challenges include ensuring that the large language models do not contain inherent biases that lead to faulty conclusions and cyber vulnerabilities that may lead to corrupted data.
“We need to have insights as to how the AI/ML is architected, and how it’s coming to its answers,” Whiting said. “We still have to have humans in the loop while we’re testing that AI/ML to prove to ourselves that it’s working and then cyber mission defense teams making sure once we have demonstrated to ourselves that this AI/ML works, that it’s not being corrupted.”
Lt. Col. Daniel Kimmich, the materiel leader at Space Systems Command’s Cross Mission Data branch, said in a May 24 phone interview that AI/ML is a top priority and one that needs increased funding.
“We are very much in our infancy,” he said. “I think the challenge that we face right now is making the data accessible for industry partners to help us truly take advantage of what the models can inform. We have some budding efforts with Air Force Research Lab. They’re in the midst of making sure our data is tagged appropriately and made available for our sensors, but, of course, one of the biggest challenges is classification. As we build the models, there’s certainly a need to make sure they’re being provided continuous access to data, as it evolves. Certainly, our biggest challenge is getting industry access to that information.”
Project Maven has been the signature DoD AI/ML effort, which has aimed to process relevant drone imagery rapidly to reduce targeting to firing timelines against fleeting targets from hours to minutes. At the Intelligence and National Security Association’s spring symposium in March, Phillip Chudoba, NGA’s associate director of capabilities, said that Maven is the “only performant computer vision, AI/ML capability in the DoD” (Defense Daily, May 23).
“Maven is positioned to rapidly deploy AI to meet DoD requirements for those real-time geospatial situational awareness needs,” he said.
The 18th Airborne Corps at Fort Bragg, N.C., has used Maven features in the corps’ Scarlet Dragon exercises, which began in December 2020 under then corps commander Army Lt. Gen. Michael Kurilla, who now heads U.S. Central Command as a four star.
“I would say they’re probably five to six years ahead of us in terms of when they started,” Kimmich said of Project Maven. “I certainly hope we catch up sooner than five to six years down the road. They’ve invested significantly. They’ve put in the requisite infarstructure to bring industry with them. We’re trying to do the same thing.”
Lt. Gen. Michael Guetlein, the commander of SSC, “has demanded that we stand up a Space Domain Awareness TAP [Tools Applications and Processing] Lab, essentially an environment in a facility that would be akin to Project Maven that would enable our industry to have access to the information they need to help us build these models and access the information they need,” Kimmich said.
Last week, Guetlein briefed AI/ML companies in Silicon Valley on how they might satisfy Space Force needs. Those AI/ML firms included Anduril, C3.ai Inc. [AI] and Microsoft [MSFT].
AI/ML “is near or at the top of our [SSC’s] list,” Kimmich said. “With that said, I think we need Space Force to equal that perception with the corresponding funding. We haven’t seen the inlays for funding specific to AI/ML and, from a Space Force perspective, to remain competitive/to remain relevant/to outpace our competitors, this is absolutely where we need to be investing our next dollar to harness its potential for protecting and defending our assets and deterrence, if necessary.”
Thus far, SSC has “a fledgling [AI/ML] effort for maneuver detection,” Kimmich said. “From the space domain awareness perspective, we have metric observation so we’re able to determine where spacecraft are located. We have light curves from our telescopes that provide characterization and information about attitude and orientation and possibly even characterization of that spacecraft–what it might be intending to accomplish. And then, there’s a third component–the RF [radio frequency] characteristics, both the transmission from that spacecraft as well as the radar cross section. Those three components have different means by which we can apply models to understand when, for example, if the RF signal/the band changes, then maybe a command was sent, and that spacecraft is going to maneuver, and we can look at patterns of life and decipher what that spacecraft may be doing and get ahead of it.”
“If there’s a means by which there are algorithms loaded into our high-value assets, information provided to them would allow them to autonomously maneuver and not wait for a command from the ground from the C2 [command and control] center itself,” Kimmich said.