Army Builds AI Development Toolkit For All Domain Ops

WASHINGTON: As the Army develops AIs for different missions, from artillery targeting to helicopter maintenance, it doesn’t want each project to reinvent the wheel. So the Army’s Artificial Intelligence Task Force is working with Carnegie Mellon University and contractor VISIMO to create a shared toolkit of reusable algorithms, test data, and development tools. It’s meant to be a “common platform” or virtual “workbench” on which Army units can build a wide variety of AIs.

[Click here for more from the Army’s AI Task Force]

From a cold start in August, with no code on hand, the team had a functioning bare-bones version – what’s called a Minimum Viable Product – ready in five months, a pace the AITF’s chief data scientist, Lt. Col. Isaac Faber, called worthy of a Silicon Valley startup in an interview.

“Without some efficiencies, it will just be science experiments,” a series of one-off tech projects that don’t add up to greater combat capability, Faber said. “If we can consolidate… it really allows us to achieve the scale that’s going to be required for MDO and JADC2,” he said, referring to Multi-Domain Operations and Joint All-Domain Command & Control.

So, “how do I enable this growing data workforce [of] AI scholars… cloud technicians…software engineers?” he asked. “What tools do they have available to them?… What’s the platform that they log into to get access to the tooling?”

The common foundation, said AITF director Doug Matty, is “bringing some powerful tools…that you can reuse and build upon those projects that other folks have already worked on.”

This month, the platform will undergo a “validation exercise” using a sample, unclassified data set – specifically Twitter posts about COVID-19 – for test purposes.

“There’s about a half-dozen organizations that are actively evaluating the system right now,” Faber said, although he wasn’t allowed to name them. “As we look forward this next year, expect a lot more adoption across bigger parts of [Army] Futures Command and the larger enterprise.”

How will the Army’s common data platform work with the all-service Joint Common Foundation being created by the Pentagon’s Joint AI Center? JAIC and the Army’s AI Task Force work together “very closely,” Matty told me, with the Army playing a leading role on the JAIC’s helicopter predictive maintenance project in particular. That will continue with the Army common foundation and the Joint Common Foundation.

“We understand that if we’re going to be successful with AI, it is inherently joint, and that’s why we have such close ties to the folks in the JAIC,” Matty said.

“We anticipate close integration with the JCF. Our development teams work together on a daily basis,” Faber added. “But right now, we’re focused on Army problems and Army clients.”