Data Sets, Security Tools Top Needs for AI Development, JAIC Chief Says

While the Department of Defense’s (DoD) Joint AI Center (JAIC) moves forward on its Joint Common Foundation (JCF) and other initiatives, Acting Director Nand Mulchandani underscored the continuing need for data sets and security measures as the basis of AI efforts.

“Data is the essence of AI,” Mulchandani said at the September 8 Billington CyberSecurity Summit. “Having labeled, curated data sets that are actually useful for both training and testing is incredibly important. Having the tool sets and libraries available to actually do AI development, that’s completely missing today.”

Mulchandani added that other efforts, such as democratizing the availability of coding data, are also important to AI efforts. Fortunately, JAIC’s JCF effort and its foundational infrastructure are “designed to solve that bigger problem off of that.” Through JCF, JAIC aims to provide an AI development network that will make it easier for DoD to test and scale emerging tech capabilities.

Securing these capabilities, however, has proved to be a struggle. “We’re dealing with tactical edge AI deployment,” Mulchandani explained, “and one of the big concerns that we always have, and our adversaries have as well, is getting access to that model or getting access to your training and testing data.”

JAIC must strike a balance that ensures security without hampering productivity, he explained. For example, control-oriented security products like firewalls can cause self-inflicted denial of service.

“We have to think of this in a more systemic, logical, controlled way, as opposed to bolting on and thinking of security systems post fact onto an architecture that’s already built and out there,” Mulchandani said.