The Washington AI Network welcomed Vice Admiral Frank Whitworth, Director of the National Geospatial-Intelligence Agency(NGA), and guests to the House at 1229 on Monday, December 4 for a special ‘cocktails and conversation’ on AI’s role in national security, intelligence, competition with China, AI’s use in warfare, and the promise of advanced AI models that can analyze visual data.
VADM Whitworth shared his insights on the complex and evolving role of AI in geospatial intelligence and national security, as well as the challenges and opportunities it presents. The conversation was released today as an episode of the Washington AI Network Podcast
Moderated by host and founder Tammy Haddad, the conversation touched on the NGA’s responsibilities in the visual domain of intelligence, the critical role of human expertise or “wetware” in training AI algorithms, and the promise of large visual models, inter-agency cooperation, ethical considerations and guardrails, AI use in warfare, including in the Ukraine conflict, and competition with China.
Guests spotted in POLITICO Playbook: Bob Woodward, Bob Costa, Phil Rucker, Josh Dawsey, Canadian Ambassador Kirsten Hillman, Elizabeth Falcone, Machalagh Carr, Teresa Carlson, Steve Clemons, John Hudson, James Adams, Evan Hollander and Eli Yokley, Jackie Rooney, T.W. Arrighi, Kathy O’Hearn and Mike Sarchet, Don Kerr and Polson Kanneth.
Highlights of the conversation are below. Transcript here.
Vice Adm. Frank Whitworth on guardrails and how AI and machines should support humans in their jobs – not replace them:
“We do have our own guardrails established in the wetware application of our tradecraft. And it’s things like training ourselves against things like confirmation bias or taking liberties with a normal checklist approach to whether you are there in positive identification. But to riff on your question, and this is where AI, I totally agree with the need for guardrails, provided we’re tempted to cut corners with AI. And right now, as far as humans, at least the humans at NGA, we’re building a certification program to ensure that will never occur, that [it] will be ethical. But I love also flipping this argument to suggesting that AI can actually provide guardrails.”
“Let’s say we had a pop-up urgent kind of episode, and we had people who have been working a long time, and they’re getting tired. That machine is not. So if that machine, during the course of the ML process, if we’ve treated that as a true digital apprentice, we’re going to count on that machine to double check if we’re getting tired, or if we might actually have a little bit of bias built in that we didn’t realize we did. I love the idea of AI actually being there to provide just another set of eyes as its own guardrail. This idea of singularity and letting the machine take over, those are human decisions. And so the way that we’re going to ensure that we approach this is very ethics bound, very certifications bound, totally in keeping with where DOD and the President now, by extension, with his executive order, 14110, we’re all in keeping with that.”
Vice Adm. Frank Whitworth on the anticipation of large visual models:
“In the visual sector for GEOINT, what I’m excited about, you’ve heard about these large language models. We anticipate, soon, large visual models. So instead of a visual detection that ‘This is a nice glass of water,’ it just tells you that it’s a glass of water. It might actually say, ‘Now it’s three fourths full.’ It will give you some context as to the behavior, or [that] it’s now moved about four inches to the right. That’ll be a phrase. So that’s kind of the equivalent in our world compared to what you’re seeing in such an exciting way with LLMs (large language models).”
Vice Adm. Frank Whitworth on competing with China:
“If I look at this from a STEM production rate and possible recruitment rate for people who are of Chinese origin, they’re way ahead. They are. The number of STEM-related graduates is probably over five times what we experience in the United States…This is concerning. We are a very STEM-oriented agency, and certainly people who understand AI and understand this tradecraft would probably benefit from being STEM graduates. What I like, though, about the United States is how experienced we are with critical thinking and with a tendency to tell the whole truth and nothing but the truth first, right up front. And this has been tested. I would tell you as somebody who’s been invested in multiple conflicts over the last 34 years, we have a really good tendency to get the bad news up the chain very, very quickly. And so during the course of the fog of conflict, the fog of war, we’re going to opt to get that information up and seek clarity and to help our decision-makers also seek clarity and know, and know before they make decisions. I don’t know, and I can’t tell you whether the Chinese actually have that advantage right now.”
Vice Adm. Frank Whitworth on combating AI-generated deep fakes and visual misinformation:
“The issue of validation is very important. I appreciate your asking that. So one of the things that we do have is a group of people dedicated to open source imagery. And there’s a lot of it out there. And as you know, an image can be very powerful. Especially if it’s misleading and, and the consumer doesn’t realize that. So what we do is we go through, and it’s typically from social media. It might be through press. But typically the stuff you want to double-check is that stuff that comes through unevaluated press or some social media that may not be as reputable. But if it’s really important, then we will actually put that product out with a validation index. So we’re looking for other indications of maybe some manipulation, some forensics in there. We’re looking for evidence on the ground that’s not quite right relative to what we know is on the ground. And I tasked them about a year ago, this team, I said, ‘Hey, don’t just put it out as a little product. Put a scale on there.” One, invalid. Absolutely not. We’re finding things inconsistent with reality, inconsistent with the truth. Five, valid. Everything seems to be checking. I have found that our consumers have really enjoyed at least having a hint of whether that’s something that is correct or not.”
Vice Adm. Frank Whitworth on humans, AI, and warfare:
“I don’t have any guidance to take a human out of the loop, or right now, I think ‘on the loop’ is a term that’s being used. But I take your point that there could be some forces out there that choose to take humans out of the loop with mission orders, autonomous kind of mission orders with vehicles. So, here’s where I am. And this kind of speaks to what I’ve termed as a reluctant RMA, a Revolution of Military Affairs, that deals with the unmanning of a lot of power projection. Not all of it, mind you, but a lot of it. And we, it’s reluctant because frankly, we’re really invested in those as humans in minimizing warfare. But if you got to go for some sort of warfare, we are really invested in doing it best as humans. And we are invested in being out there.
Vice Adm. Frank Whitworth on the role of AI (and U.S. help) in the Ukraine conflict:
“I can’t speak to whether the Ukrainians are independently applying AI. They’re very resourceful. They’re very IT savvy. Would not be surprised to find out that they are. I think we’ve got one or two reporters who are just there, and they could probably have more authority on that topic. I do know this. If there is a way to provide an advantage to the Ukrainians, that is our writ. That is what we do. Through Ucomm, we are providing a tremendous amount of information to ensure that they have what they need. We don’t do the targeting for them. They make their own independent assessments. They make their own independent decisions on what they will actually neutralize. But we have a responsibility, as the president has stated from the beginning of this, to ensure that, that we’re not holding back in, in the information that they need to make good decisions.”
Vice Adm. Frank Whitworth on MAVEN Smart System:
“And I like the direction that we’re moving in the MAVEN program to ensure that we are soliciting as many opportunities out there by smalls as we can. Now, that will have to be tempered by some security issues. We’ll have to make sure that as we protect these algorithms, that we don’t get too far afield. We don’t want to go back to being just a project that’s so flat and apparent that, frankly, you find your algorithms getting stolen.”