JAIC wants to make smaller ’boutique’ data sets

The Department of Defense’s artificial intelligence hub generally thinks big, but now it’s thinking small in one key area: data.

The Joint AI Center wants to create “smaller” and more “boutique” data sets to fit specific needs across every part of the military. While much of AI’s recent development has been spurred by the new ability for computers to power through massive data troves, the turn in thinking at the JAIC was inspired by its goal to bring AI to “hyper-local levels” in the military.

“Smaller and smaller data sets are important,” said Nand Mulchandani, JAIC’s acting director, during an Institute for Security and Technology webinar Tuesday.

Hyper-local levels could include everything from tactical units or smaller components of the DOD that have specific use cases for AI. For example, if the Navy SEALs needed to train an AI model on data for how to optimize human endurance, the necessary data would ideally contain specific points from the elite part of the Navy’s Special Warfare Command. Part of the process for creating hyper-local data sets is devolving data management responsibilities down to component and service-levels.

The DOD’s cross-department data management remains “a little haphazard,” Mulchandani said, but the DOD’s new top chief data officer, Dave Spirk, and others are collaborating on data governance.

“Instead of thinking of data as undifferentiated mass … you have to start thinking about it like that book shelf,” Mulchandani said, referring to the interviewer’s bookshelf in the background. Being able to pick out datasets for specific use-cases will be a critical part of the DOD’s AI development, he said.

Mulchandani teased at a new acquisition vehicle the DOD will be unveiling in the near future that could allow for more use out of small data sets from the private sector. He outlined a system that would allow the JAIC to partner with small technology companies to run “pilots” on enclaved DOD data, allowing the center to work more efficiently with the private sector, which remains far ahead of the DOD on AI development.

The vehicle will make acquisition “incredibly easy,” he said.