U.S. Army Is Considering AI Bill of Materials
To reduce the risks of harnessing artificial intelligence (AI), the U.S. Army is considering the applicability of an AI bill of materials (AI BOM). Young Bang, the principal deputy, assistant secretary of the Army, Acquisition, Logistics and Technology (ASA(ALT)) met with AI companies and fellow Army leaders at the Technology Exchange Meeting X in Philadelphia on May 25. The group examined the initial considerations of applying a bill of materials structure to the use of artificial intelligence, similar to how software bill of materials (SBOMs) are used, Bang said.
The National Telecommunications and Information Administration and other organizations, including the military, pioneered SBOMs to provide a detailed description of the software components in any software-based product to identify cyber vulnerabilities and reduce cybersecurity risks.
“We've been driving and pushing software BOMs and data BOMs,” Bang stated. “We're toying with the notion of an AI BOM. ... Just like we're securing our supply chain, with semiconductors, components, sub components, we're also thinking about that from a digital perspective.”
Bang, who has been instrumental in the service’s latest foray into AI, including with the inception of Project Linchpin, sees a similar possible application of BOMs to AI. The idea of identifying the components of AI could potentially help strengthen cyber risks and improve the data environment, he said. And AI BOMs represent the potential to improve the Army’s AI supply chain.
“And that's because we're looking at things from a risk perspective,” he said.
Naturally, industry is worried about the protection of their algorithmic solutions. Here, Bang clarified that the Army stance is not to access corporate intellectual property, or IP.
“I just want to make sure we're explicit about this,” he emphasized. “It's not to get at vendors' IP. It's really about how do we manage the cyber risks and the vulnerabilities. And so, we're thinking about how do we work with industry, as AI BOMs are a little bit trickier. And obviously data is a driver for both software and AI. But the AI side becomes a little bit different, because arguably, depending on what you request, you have the ingredients to potentially backward engineer their algorithm. And so we don't want to pinch on their IP.”
We're toying with the notion of an AI BOM. ... Just like we're securing our supply chain, with semiconductors, components, sub components, we're also thinking about that from a digital perspective.
Instead, the AI BOM application would be to look at the provenance of how algorithms were developed, Bang clarified.
“What are the features of the parameters you tested?” he said. “What are the data sets that you used? [It is] to ensure we have more trusted outcomes, so that there's no risk like Trojan triggers or poisoned data sets, prompting unintentional outcomes of the algorithms.”
“It is about securing our digital supply chain and the AI,” Bang stated. “Understanding how we did something might help us get better insights to the outcome. But initially, it's about reducing the risk, and that will help us with the trust and responsibility.”