Intel Agency Teaches Machines to Read Maps of the Future

By Duncan Scot Currie

Duncan Scot Currie is the Director of the Source Mission Integration Office Imagery at the National Geospatial-Intelligence Agency (NGA). His team is responsible for managing equities in performance, policy, open source research, and the systematic integration of all geospatial intelligence sources into operations. Currie was previously the Director of NGA’s Commercial Imagery Data and Programs Group from 2009-2012, where he was responsible for executing contracts related to the acquisition of domestic and foreign commercial imagery, data, products, and services.

Geospatial intelligence, or the collection and analysis of the physical features of the Earth, has long been an integral aspect of both strategic national security efforts as well as tactical combat. The U.S. National Geospatial-Intelligence Agency (NGA) has been at the forefront of confronting national security challenges, from terror groups in Iraq, Syria, and Afghanistan, to overhead imagery of the secretive North Korean ballistic missile and nuclear programs. Much like other members of the U.S. intelligence community, the NGA is seeking to further modernize its efforts by bringing in cutting edge technology. The Cipher Brief’s Levi Maxey spoke with Duncan Scot Currie, the Director of the Source Mission Integration Office at the NGA, about how the agency seeks to move forward in the years to come.

The Cipher Brief: In the last 15 years, budgeting for the NGA has shot up. What kind of advances in technology has this spurred as far as geospatial intelligence collection and analysis?

Duncan Scot Currie: We have made some investments to make all the sensors more efficiently work together. We’re moving to the cloud – the entire intelligence community is moving towards the cloud – and our data is going with it. So, all our library of commercial and national images has to be moved. The value of having it in the cloud is now you can process that data constantly with algorithms and find patterns and extract value so analysts don’t have to put eyes on it. Those are the investments we’ve been making over the last 15 years.

The future is machine learning, computer vision, and that kind of technology. How do I teach these things how to read with a level of accuracy that I am confident these algorithms can represent what a human could do if they were just counting planes, looking at position changes, or whatever? If I can trust that algorithm, then I can take those people who are doing that function and move them over to some task of higher order. Our desire in the future is to replace 75 percent of the routine functions that imagery analysts do with machines, and move those people to focus on doing more anticipatory analysis. That’s the whole strategy going forward from the NGA perspective.

TCB: What are some of the broader national security problem sets that geospatial intelligence is useful for? Is this primarily for identifying troop movements. Can you do this with smaller groups, such as terrorist organizations?  

Currie: We support the war on terrorism constantly, particularly in Syria with the ISIS threat. We are also looking at the proliferation of weapons in North Korea and the nuclear program in Iran. Geospatial intelligence is applied to pretty much every aspect of the national security problems the U.S. faces. We’re less involved in cyber as perhaps the NSA is, but you can essentially characterize almost anything if you lay it on a reference framework with a geospatial base. It’s kind of a foundation or base map of everything – even in the cyber realm – when you need to determine where threats are physically located. That’s how we work together with the rest of the intelligence community.

TCB: How does imagery and other geospatial intelligence get to warfighters? Is there a real-time feed that is constantly updated that can be queried?

Currie: We can get our data from satellites to the warfighter within minutes. The long-haul question is how long does it take for somebody to look at the image, extract the value, figure out what happened, and get it out? That’s the long haul. If you can bring that timeframe down to a minute using an algorithm, then you can push it out and save all that time. But getting it down off the satellite to image processing and then out to troops can be done in a few minutes. But it’s raw intelligence – analysts on the ground have to sit down and look at it.

If I could process the algorithm on the satellite and then downlink it into theater, that analysis would already be done. So how do I push the processing and algorithm as close to that satellite as I can, get that value extracted out of the imagery quickly, and then downlink it in theater? Everybody’s trying to reduce that timeline. It doesn’t do any good to have five-minute revisit image compilations of the ground if it takes 3 hours to move it.

Our whole thing is, how do I get that timeframe tightened up? The Department of Defense recognizes this too. We’ve got enough satellites. Where we are failing right now is the ground team’s ability to make sense of it and integrate it all so we can get it out to the people who need it in a timely fashion in order to make decisions effectively.

TCB: There are always assumptions behind the algorithms of data analytics. Is there a way to go in and look at the data underneath to see how it came to the decisions it made?

Currie: We’re helping build the analytic models. If we see something and it fits a pattern that we expect, that automatically triggers further collection. Essentially, if we see something happening and know that five things will usually happen next, we can go out and start collecting on those five things right away. It works very quickly.

So, yes, a lot of what we’re doing is trying to train the machine to recognize patterns and make sure the patterns we’re establishing are correct. And those patterns aren’t static, they can change if people change their doctrine or learn that we know something they didn’t think we knew. They can change their behavior and we have to start all over again. It is a kind of analytic warfare. How can we build enough algorithms to so we can get ahead of the curve and reduce that timeline from sensor to shooter?

TCB: Are anti-satellite systems a major concern when it comes to blind spots in geospatial intelligence?

Currie: We understand where we have single points of failure. If we can, we fix those to ensure we have redundancy on the ground. We’ve got investments we’re making, and everybody else who is related to command and control or processing of satellite data is doing similar things. Sometimes there is safety in numbers. Say an anti-satellite system takes three of them out. If I have 300 then what’s the big deal, right? So that’s kind of a numbers game. At some point they can’t take everything out, right? They may take out some real big and important ones, but you’d hope that you’ve got enough others that you can still cover collection.

The other project we have going on now is called GEOINT Broker, where we kind of move toward looking at what is out there and bring it together in a way in which it’s all complementary and it all becomes resilient. We need that resiliency, and that’s part of what we need to do institutionally from the national security perspective.

TCB: So, if you have 300 satellites out there, are they all talking to each other? How do you secure those data links?

Currie: So that’s all commercial. Each one of the commercial satellite providers has to come up with their own methods case for that. You come to us and tell me what you’ve done to protect your data. What have you done to make sure we can count on all that data and that you’re not introducing error or problems into our data, or that you’re not going to be attacked or cyber attacked and take our entire infrastructure down? What have you done to protect against that?

TCB: A lot of NGA’s operations are classified, but not all, such as assistance during natural disasters. Are these the testing grounds for commercial tech to see if they are capable of contributing substantially?

Currie: We do a lot of that, certainly Digital Blowup is one of the preeminent ones that we have that does a lot of that disaster support. There’s a consortium of commercial companies that post to a site all their imagery so we can make best use of that or whomever can make use of it. We found over the years that there’s tremendous value not just on the humanitarian side, but also on the warfighter side. It’s all unclassified, they can share it with all of the coalition forces. The other piece that primarily is unclassified is the whole foundation mission or the mapping mission we do for safety of navigation, maritime, for the Navy, Airborne for the Airforce, topographic for the Army, etc. All that is now moving to unclassified production and 90 percent of that data is from commercial centers.

TCB: Where do you see NGA moving in the future?

Currie: We’re trying to do as much as we can within our own budget to stop some of the legacy behaviors that we’ve been finding for years, trying to free up as much money as we can to do some of these more futuristic kind of functionality of moving towards anticipatory intelligence. We’re making those trades in our own budget. We can make a strong case for getting some resources to help us in that regard, especially on the processing piece to get the traditional eyes of the images that I talked about before.

Related Articles