WASHINGTON — The U.S. military is exploring a range of artificial intelligence capabilities, from predictive maintenance of ICBMs to analyzing reams of satellite data.
This creates opportunities for AI startups like Virtualitics, which has won Air Force and Space Force research contracts and is offering an AI tool that could identify vulnerabilities in procurement programs by analyzing historical data and predicting potential issues before they arise.
Kyle Rice, chief technology officer of Virtualitics’ federal business, said he heard Space Force leaders talk about the need for improvements in how the service acquires satellites and other systems, and believes that AI tools can help.
AI can flag risks in procurements by analyzing historical data and uncovering patterns in past contracts, Rice said Feb. 13 at the Air & Space Forces Association’s Warfare Conference in Aurora, Colorado.
AI for predictive maintenance
Virtualitics, based in Pasadena, California, had a contract with the Air Force Nuclear Weapons Center to use the company’s platform to predict maintenance needs for the fleet of Minuteman 3 intercontinental ballistic missiles. That contract has ended but the company currently has an active contract with the Air Force Global Strike Command for predictive maintenance of the nation’s bomber aircraft fleet. The technology helps identify maintenance needs and potential equipment failures, manage inventory and human resources.
“The Space Force leadership is talking about procurement improvements, getting things done faster and how to streamline that,” Rice said. “Why that excites me from an AI perspective is one of the areas of AI that’s getting a lot of uptick around DoD is around project assessment.”
This is essentially using AI techniques to look at data about past programs. “We collect so much data on how we build things, in procurement documents, in financial documents, in delivery schedules,” Rice said. “And when you run models against those you can really get a good feeling for what are the attributes that make systems successful.”
By uncovering hidden trends and red flags associated with vendors, projects, or technologies, the Pentagon could save millions of dollars, he said. “It’s a super interesting AI use case,” he said. “You can determine that this project may be going off the rails because they’re doing certain things that are not associated with successful on time delivery.”
It’s an example of “capturing data so we can do some good stuff with it,” Rice said.
Can AI be trusted?
Rice spoke on a panel moderated by Maj. Gen. Stephen Purdy, who serves as the military deputy to the U.S. Space Force’s senior acquisition executive Frank Calvelli.
Purdy said the Space Force is eyeing AI for enhanced space situational awareness. Companies are developing AI systems that can scan vast amounts of satellite data, automatically identifying and tracking objects in orbit. This real-time intelligence is crucial for monitoring threats and safeguarding critical space assets.
Purdy also raised questions about whether the government can effectively trust AI, and rightly noted public wariness of AI chatbots that generate erroneous information.
Rice explained that most of the AI tools used by the military, such as those for space domain awareness, rely on structured sets of data so the result of the analysis is not made up. “What that means is that if you get a result, you can back it up, you can figure out why the models suggest these things,” he said. “Obviously, for certain use cases in DoD, that’s very, very important.”
By contrast, widely used AI applications like ChatGPT falls into the category of generative AI. “It’s a fabulous thing. There’s lots of good things you can do there. But a significant downside is that it is not an explainable technique. When you get a result, you don’t actually know how you got there. There’s no way to back that up.”
“For some use cases, that’s bad, and for other use cases it doesn’t make any difference,” Rice added.
The military mostly seeks AI for enhanced analysis, rather than content creation, he noted. Generative AI’s strength is that “you can generate something new. But sometimes what you generate is not actually accurate. That’s where you can pair that use with other types of AI where it is more deterministic.”
