Booz Allen looks to independent research to differentiate its AI work

Booz Allen Hamilton, buoyed by recent wins in artificial intelligence work, is pursuing independent research related to AI, including on adversarial machine learning.

In a call with Inside Defense last week, Edward Raff, chief scientist at Booz Allen, said the company is focused on growing its base of knowledge in AI.

"We have a research agenda we've built up on questions we think are relevant to our clients and we think we have the skills in-house to tackle," he said. "Adversarial machine learning is one of them."

Another area the company has been researching for about three years is fairness and transparency as it relates to AI.

"There's other research we do that's just sort of opportunistic," Raff added.

Booz Allen has made artificial intelligence a key focus in recent years. Last year, Horacio Rozanski, Booz Allen's chief executive, said the contractor has sought to develop "premier offerings in artificial intelligence."

"We expect continued strong growth in demand as more and more federal agencies work to integrate AI into their missions," he said.

Earlier this year, the Pentagon's Joint Artificial Intelligence Center awarded the company a contract worth up to $800 million over five years to serve as the lead integrator for a marquee "Joint Warfighting" initiative that seeks to leverage AI and machine learning on the battlefield.

Raff said Booz Allen is seeking to differentiate itself by building its expertise, including by getting papers into peer-reviewed publications.

Meanwhile, he said the Pentagon is recognizing the importance of addressing the adversarial AI threat and security in AI.

"This is something we need to be thinking about now and not some after-the-fact Band-aid," he said. "Culture-wise, I think that DOD is doing a good job heading away from that trap."

Booz Allen's research team is "taking some of these adversarial questions and adjusting the lens to how we believe this will be really relevant to the DOD situation."

"How do we set all these knobs and parameters to be realistic to DOD?" Raff said the company is asking. "To me, that's really critical because if we just wait for the rest of academia to solve our problems, we could be waiting several extra years."