CIA, Treasury CAIOs Talk Leadership, Governance Priorities

Chief AI officers from the CIA and Treasury Department outlined priorities in governance and artificial intelligence development amid various maturity levels for the technology in their agencies.

“There’s increased focus now in this space, so we really focus a lot on governing on how we are aligning our dollars in the most efficient way as government stewards to help us drive [AI] adoption,” CIA CAIO and Director of AI Lakshmi Raman said at AI FedLab in Reston, Virginia, Wednesday.

Part of the White House’s directives for federal AI development includes agencies having to name CAIOs and begin building out their AI offices. Communication has been a key component of this transition, said Treasury Deputy CAIO and Director of Domestic and International Cyber Policy Brian Peretti, who will begin a new role next week as CAIO and CTO.

“We’re trying to figure out how all this kind of fits together,” Peretti said. “As we’re trying to figure out what this construct is, how do we get to a place in which we’re sharing that dialogue back and forth? And the more that we can learn from practitioners — while we’re also a practitioner — hopefully gets us better aligned.”

Raman said the agency’s AI department had been in place with her role since 2020, so the agency already had a head start on defining AI strategy and understanding AI needs at CIA compared to other agencies.

“I started in January [2020] leading a small office to help us integrate, champion, drive activities within the agency,” Raman said. “In order to clearly understand what’s going on with AI and make it work, the first thing we have to do is pass a strategy, develop a strategy, create a strategy and adopt it in a corporate governance.”

Peretti noted the requirements from White House directives have required adjustments at the agency where the value of communication has been prominent.

“We have a decentralized cybersecurity platform across Treasury, and that is going to be, I think, a little bit of a change in this space here,” Peretti said. “Part of the executive order requires communication across the organization, a governance board, which requires that the organization understand what’s going on, that requires that we push and publish use cases. That means there has to be different communication that we had before.”

Raman said she is working hard to establish capabilities for the CIA to use AI and to integrate it into operations. She also emphasized the importance of working together with other top tech officials at the agency to further the intelligence mission.

“I have a very close partnership with our chief data officer because we really need to ensure that we have the data available in bulk for us to be able to leverage AI,” Raman said. “I have a very close partnership with our CIO because we need to ensure we have the right technology foundations in place to be able to enable AI at scale. I also have a very close relationship with our general counsel, our privacy and civil liberties officer, because whatever we’re developing needs to be adhere to all our legal requirements.”

AI use cases vary for agencies, but there are some common elements. Treasury, for example, is using AI for fraud prevention and financial recovery.

“You may have heard many stories about that [Paycheck Protection Program and] a lot of fraud that was out there people taking advantage of that,” Peretti said. “One of the things we saw was that as we do it within a part of our organization, we were able to save $375 million just by deploying AI.”

AI can process large amounts of data quickly, and other use cases are becoming clear, Raman said. CIA is exploring generative AI for these novel use cases as the technology develops.

“We’re thinking about [generative AI] in terms of how does it help as ideation aid, or how does it help as a search and discovery aid, for brainstorming?” Raman said. “[AI is] helping us generate counter arguments, you can imagine all of those areas are incredibly beneficial in the types of work that we do.”

Part of the White House’s strategy for AI also emphasizes its ethical and responsible use. Raman noted what this looks like of the intelligence community (IC).

“We need to be thinking about privacy and ethical risks and within the IC. We’ve actually had an IC AI ethics framework for quite a few years,” Raman said. “We’ve really been thinking about how we do this responsibly.”

In addition to safety concerns, workforce challenges have emerged around AI implementation. Peretti noted the discrepancy around skills that will emerge around all emerging technology as well as the challenges around upskilling and hiring workers who will be able to use future technology.

“There’s going to be a lot of dislocation of employees out there in which the current skill set may not be as critical as it was when we hire the folks,” he said. “Guessing where we’re going, trying to align our resources to get that skill set, is just going to be really difficult. But we’re going to have to kind of figure out how we, as a society, get to that space, which is different than the way in which we operate in.”

Culture in agencies can help ameliorate those roadblocks, Raman suggested. While future-proofing is impossible, it’s critical to adjust an agency’s bearing toward accepting and learning new technologies and ideas.

“It’s also about driving our culture inside the building to accept new technologies, new ways that these technologies might change the way they do their job,” Raman said. “We’re thinking about this from a human-machine teaming perspective [with a] human in the loop and that they are still incredibly important to this process.”