A screengrab of model encounter of an American and a Chinese satellite in space. (COMSPOC)
WASHINGTON — Artificial intelligence holds promise as a key tool in helping the Space Force achieve what service chief Gen. Chance Saltzman has called “actionable space domain awareness” — that is, timely warning about hostile adversary actions — and avoid “operational surprise,” according to experts.
“AI algorithms are really good at looking at a lot of streaming data and saying: ‘This is a change detection. … This is not nominal,” Lt. Col. Ashton Harvey, chief technology officer at the National Reconnaissance Office, said Tuesday.
Thomas Roberts, a graduate research fellow at the Massachusetts Institute of Technology’s (MIT) Astrodynamics, Space Robotics, and Controls Laboratory, agreed. Further, he told Breaking Defense today, there is plenty of observational data going back long enough in time to be able to use AI and machine learning to tell the difference between normal and “abnormal” behavior of specific satellites of interest to the Space Force and Intelligence Community, such as those owned by the Chinese military.
Indeed, Roberts himself said he has developed a software model [PDF] to do for satellites in geosynchronous Earth orbit that was presented at last year’s AMOS Conference on monitoring the heavens.
Avoiding “operational surprise” is one of the three “core tenets” of Saltzman’s new “Competitive Endurance” concept for how the Space Force can be successful in both deterring conflict in space and fighting to achieve “space superiority.” Saltzman first articulated his “theory of success” in a note to Guardians in May, and just last week circulated a “white paper” elaborating on the concept.
That paper, obtained by Breaking Defense, stresses the foundational role of timely space domain awareness in ensuring that Space Force commanders are not caught flat-footed by adversary actions — and, if necessary, are able to take action to prevent harm to US military forces.
“Actionable space domain awareness is a timely understanding of the operational environment that is relevant to decision-making. SDA is the combination of intelligence, surveillance, reconnaissance, targeting, cooperative reporting, environmental monitoring, and decision support tools,” the paper states.
“Timely and relevant indications and warnings will help us avoid operational surprise in crisis and, when appropriate, take defensive actions. In the event of combat operations, awareness of adversarial space dependencies allows space forces to disrupt space-enabled kill-chains that threaten the Joint Force,” it adds.
Benefits of AI
Speaking at the “Space & AI Summit” sponsored by Booz Allen, Harvey explained that besides using AI to sift through historical observational data to find what is normal and not normal behavior for certain satellites and space objects, new “large language models” such as ChatGPT also could be useful in providing early warning to national security space operators.
“There are a lot of opportunities to leverage large language models to look at unstructured data that otherwise we couldn’t pass into an algorithm very easily to potentially pull out insights — whether that be news reporting in foreign languages because it can read the foreign language and I can’t, or looking at Twitter posts via video and turning video to text that can then be searched — to be able to look for other left-of-launch indicators that might tell me I need to spend more cycles looking at something,” he said.
Maj. Sean Allen, who heads the Space Force’s new Space Domain Awareness Tools, Applications, and Process (TAP) Lab technology accelerator, said that service efforts to develop AI, machine learning and software tools for “event detection” have picked up in recent months following the appointment in June of Col. Raj Agrawal as the head of Space Force Delta 2 responsible for space domain awareness. This is because Agrawal shifted the focus from “closing kill chains” for US military forces in countering adversarial actions to looking for indications that an adversary was readying an attack.
“[Agrawal] said, ‘My priority to avoid operational surprise is to detect the start of a kill chain.’ And for whatever reason, that small change in language has converted very well, or translated very well, to software engineers, machine learning ops folks, data scientists, to say, ‘I can do event detection, that’s something that I can measure’,” Allen said.
Challenges To Implementing AI: Trust, Lack of ‘Spicy New Hotness’ Opportunities
Harvey said that one problem is that humans have “cognitive biases or preconceived ideas,” so although AI systems can identify “weird” or “non-intuitive things” that operators haven’t noticed, those findings are often pushed aside.
“In almost every domain, where the AI will find non-intuitive things, the human usually dismisses those as the AI being wrong. And in many cases, actually, it was just something we never imagined,” he said.
Pat Biltgen, who works on AI for mission systems at Booz Allen, said another issue is that operators also are prone to overreact to mistakes made by AI programs.
“Trust comes with time. The really tricky thing about AI is even when it’s right 99.9 percent most of the time … it’s like you see the one example that’s really obviously wrong, and you’re like, ‘This is never gonna work’,” he said. “Like none of us have ever made a mistake.”
Roberts said that part of the distrust comes from the fact that using AI for space domain awareness is relatively new.
“There’s a lot left to learn when it comes to developing AI-based models for satellite behavior analysis, which can make decision-makers feel uneasy about using these kinds of tools, especially for critical questions of national security,” he said.
To “accelerate development in this field,” his MIT lab is being funded by the Air Force-MIT AI Accelerator to sponsor a prize challenge for innovators in applying AI to find “patterns of life” for satellites, Roberts said. Interested participants have until March 17 to submit a proposal.
Allen noted that one of the issues for the military is access to commercially available data in the right formats that could train AI models to be reliable.
“I think a lot of the bottom of the data science ‘hierarchy of needs’ goes unmet in many cases: access to data, publicly available expert labeled datasets, so that people can throw some spicy new hotness against it,” he said. “Our community has done a poor job, and could do a lot better in solving the bottom of this pyramid.”
While Space Force leaders long have touted the need for injecting commercial space monitoring data into service domain awareness operations, that has been easier said than done. Besides concerns about data reliability, there also has been push back from those who fear enabling commercial providers might also reveal US space assets and capabilities that the Defense Department wants to keep secret — including some hidden behind Special Access Program status that even most Space Force operators lack clearances for.
