US Working on AI to Predict Ukraine’s Ammo and Weapons Needs
But a Pentagon watchdog is already unhappy with the way materiel is being tracked.
FRANKFURT, Germany—In a large office building amid closely-cut grass, one of the U.S. military's top data minds is developing machine-learning algorithms to predict Ukraine’s ammo and repair needs, rather than just react to them. But an older problem persists, according to the Defense Department’s inspector general: the Pentagon isn’t doing enough to keep track of what’s going where.
This is the joint operations center of the International Donor Coordination Center, or IDCC, where officials from the U.S., Britain, Ukraine, and a dozen other countries track the transfer of donated weapons and supplies, right down to individual bullets, IDCC officials told reporters this week.
The process begins with a request from the Ukrainian Defense Ministry for, say, armored vehicles or bullets. IDCC officials check to see if there’s a donor country or entity that has the item. The coalition officials then design a process to get that materiel to Ukrainians, who then take it over the border. Along the way, the coalition officials document what was requested, what was donated, and what was received.
The requests also allow the IDCC to calculate and track how quickly the Ukrainians are using up materiel.
“Understanding their usage rates is important to understand how quickly they need to be replaced,” one coalition official told reporters.
The IDCC also works security issues, such as when an arms shipment from Macedonia was picked up on social media. Officials adjusted the timing of the shipment to avoid drawing public attention. They said that while some of the shipments were very conspicuous at first—think armored vehicles on flatbed trucks—the shipments have since become much harder for the outside world to detect. IDCC officials have figured out ways to ship items without flight numbers or other tracking indicators that might tip off Russian intelligence trying to intercept or destroy the aid in transit.
The requests from the Ukrainians for more supplies are both constant and urgent, according to the officials. That’s partially why the next step is take the large amounts of data that the coalition is collecting and develop AI-driven techniques to anticipate those needs ahead of time, rather than just respond as they come in, said Jared Summers, the chief technology officer of the 18th Airborne Corps, who is working with the IDCC in Germany.
It’s the sort of thing big companies like Amazon do to ensure they can meet demand. But large retail outfits have a few advantages that the IDCC does not. They can add sensors to shipments to get a complete data picture. Summers said the IDCC is planning to work around this shortfall by developing predictive models from existing data.
“You can start to see failure rates and break-down rates once we get enough of that data in. And we can actually build that model without having to invest in sensors,” he said.
Knowing when an item might break is key to knowing when to ask for a replacement. He said the models to predict Ukrainian needs were “in development” and expressed cautious optimism that they would be in place by year’s end.
That sort of predictability could make a vital difference to Ukraine and its allies as they seek to make sure that Ukraine can continue to accept aid and fight through next year and beyond. It might also help to satisfy some of the concerns from lawmakers and others on accountability for the aid that’s been given and asked for.
But there are some big differences between how the IDCC tracks and keeps data for its operations and how the Defense Department Inspector General would prefer it to do so.
In July, the DOD Inspector General’s office said the Pentagon wasn’t using the year-old analytics hub that’s supposed to simplify the gathering and sharing of data. Instead, the Defense Department was using journal vouchers. “The use of summary journal vouchers is a concern because journal vouchers have the potential to limit the transparency of the funds, particularly if the summary journal vouchers do not trace back to the supporting transactions details,” the report said.
On Tuesday, the IG office issued a new report, reminding the Defense Department to follow its own rules on accounting procedures, and cites a March DOD memo that “states that Advana will be the authoritative and only source for reporting the Ukraine supplemental funds and that DoD Components are required to update Advana weekly with their execution of direct funds.”
The new report noted that the Office of the Secretary of Defense has “made improvements to the functionality of Advana to increase the traceability, transparency, and reporting of Ukraine supplemental funds execution.” But the analytics hub still wasn’t being used widely or consistently enough. The result is that parts of the Defense Department are working with different numbers on what sort of aid has been allocated, used, and so on.
Summers said “Advana is an important piece of the overall required ecosystem to support data centric operations,” said Summers. “Also, when you deal with tactical data links, which we do… that’s also near real time and driving additional technological requirements to deliver at the speed necessary to achieve decision dominance.”
The coalition needs to use all that data it has as quickly as it can get it, rather than run it through some analytics platform before they can see and execute on requests.
“The preference is to be able to take that information directly from the source to go and drive our decision making in our processing instead of taking it out of the system to somewhere else to work something else to then have it fed to us,” he said.
Summers said he’s currently working on it with Craig Martell, the Pentagon’s chief digital and artificial intelligence officer. “We talk with his team, many multiple times a week, if not on a daily basis on a variety of topics,” he said.