The US Army is keen to integrate commercial private-sector AI algorithms into its operations and it's hoping industry can also figure out how to address the inevitable security concerns that will come from the move.

Young Bang - principal deputy assistant secretary of the Army for acquisition, logistics, and technology -
told an audience at Amazon Web Services' Washington, DC summit last week that it would be a waste of time for the Army to reinvent the systems that private industry has already created.

While the Army is overflowing with data that today's machine-learning systems are capable of crunching through, developing the algorithms to perform that work has been done already or is underway, so Uncle Sam may as well save itself the hassle and take advantage of that, Bang said.

"We have tons of data … but we're not going to develop algorithms better than y'all," he told the summit. "We want to adopt third-party-generated AI algorithms as fast as y'all are building them."

Bang said the Army is the largest user of AI and algorithms across the six branches of the US armed forces because, unlike the Navy with its ships or the Air Force with its planes, "our resource is our people." Those people generate a lot of data, ergo the Army ends up being the biggest user of machine-learning software to process all that information. That's why the Green Machine wants the industry's algorithms as a shortcut to handle that analysis.

Bang also spoke at a breakout session that provided some additional details on the Army's tech posture with regards to AI. 

The Army, Bang
said, was finally reaching a point where it had fully adopted modern software practices such as agile development and CI/CD - aka continuous integration and continuous delivery. With those practices virtually in place, the Army hopes it can use them with artificial intelligence to assist in data processing.

Addressing the risks

As well as the usual bias and hallucinations emanating from neural networks, depending on those networks' architecture and use, there are also security concerns - which range from skipping over safety guardrails with special prompts, to standard application security issues. These need to be taken into account before wiring these models up to programs handling sensitive info and/or assisting in the recommendation of action that's a matter of life or death.

Rather than assessing those risks itself, the Army thinks it can put out a request for information (RFI) to get the answers it needs from the private sector. 

"This is the Army saying we need your help," Bang told the summit audience. "We're trying to overcome things that could prevent us from adopting third party-generated algorithms. 

"We want you to identify certain controls, and we want you to come back and say 'here are some processes and tools we have that can help you,'" Bang added. The Army acquisition chief noted that the service is particularly keen on, among other things, having mechanisms in place to reject data and models that have been poisoned or booby-trapped so that they cause systems to unexpectedly misbehave down the line. That data could be in the training set or provided during inference.

It's not immediately clear when the RFI will be available for industry comment, and the Army didn't respond to questions. A spokesperson did tell Washington Technology that there would be multiple AI-related RFIs in the coming months, with the one Bang discussed debuting by the end of August at the latest. 

Defining the risks around the Army's use of machine learning has been a priority project this year, with a 100-day AI plan unveiled in April centering on defining adoption obstacles. Bang said that, once the 100-day plan is complete, the Army would be transitioning to a 500-day plan to operationalize everything it learned from these risk assessment efforts. 

That 500-day effort will include efforts Bang referred to as "BreakAI," though he didn't elaborate on what that may entail. ®

https://www.theregister.com//2024/07/02/us_army_private_sector_ai/