Dawn Meyerriecks, Deputy Director of Central Intelligence for Science and Technology, delivered a SINET "think forward" presentation in which she shared some lessons learned and a wish list for innovation.
Meyerriecks began by describing how the US Central Intelligence Agency (CIA) works hard to baseline its science and technology investments against porfolio best practices. They're trying to ride the waves rising in the commercial space. They only want to invest tax dollars in things no one else is investing in.
So, considering what the CIA does will help you understand their investment. Human intelligence, "HUMINT" matters a lot to the CIA. There's nothing like understanding what an adversary's leader's intent is. The CIA, she said, putting it into a business context, does "competitive intelligence" for the US.
CIA works, of course, closely with IARPA, the Intelligence Advanced Research Projects Agency, a kind of Intelligence Community analogue of the better-known DARPA. IARPA invests in operations, collection, analysis, and anticipatory intelligence. (That fourth area has a particular interest in machine learning.) And then there's In-Q-Tel, the CIA's major interface with the venture capital community. "In-Q-Tel is very good at aligning start-ups with problems the CIA is actually interested in." These organizations and others represent some of the ways, Meyerriecks said, "we're trying to reduce the impedance of doing business with us."
She touched briefly on two problems she'd welcome industry's help in solving. "How we aggregate other people's results without reingesting their data, and how we do so with high confidence, is important." She would love to see entrepreneurs take this up. And machine learning is of great interest. But here it's important not to oversell, which, she said, far too many vendors do. According to their due diligence with In-Q-Tel, she observed, "only one in thirty companies that say they're doing machine learning are actually doing machine learning."
Why machine learning? "We love machine learning because it enables our analysts to apply their longitudinal experience. Romantic vignettes are not the way we should influence policy. You've got to be able to show your work." Machine learning offers the prospect of empowering analysts to do just that.