April 5, 2011 at 4:43 PM ET
The ongoing military campaign against Afghan insurgents may get a boost from new computer software designed to zero in on the locations of weapons caches and warlords.
"The idea is to say, look, this is a large area, where do you target your resources," Venkatramanan Subrahmanian, co-director of the Lab for Computational Cultural Dynamics at the University of Maryland, told me today.
The software, called SCARE (Spatio-Cultural Abductive Reasoning Engine), combines data on terrain, road networks, tribal affiliations and past attacks with a computational analysis technique called geospatial abduction to help locate the enemy.
Geospatial abduction is a way to infer unobserved geographic phenomena (such as where explosives are hidden) from a set of known observations and constraints, such as the locations of past attacks and the roads that have been used to move around large caches of bomb-making materials.
SCARE was first used to analyze attacks in Iraq involving IEDs (improvised explosive devices) and predict the locations of IED weapons caches. Subrahmanian's team optimized the second generation of the software for use in Afghanistan.
Applying the technology to Afghanistan presented several new challenges, including the varied terrain, the vast area that had to be covered (360 by 270 miles, or 580 by 430 kilometers), and the influence of multiple tribes.
"There are inter-tribe rivalries, so clearly tribesmen who carry out certain attacks are more likely to want to seek refuge with parties they trust, which happens to be their own tribe," Subrahmanian said. "So we had to understand the geography of the region both in terms of terrain and roads as well as tribal affiliations."
Other constraints on the insurgents include a desire to carry out their attacks near their home bases or weapons caches, but not too close to them.
And since the software is designed to find large weapons caches and high-level insurgent leaders, "We use an assumption to restrict movements of the insurgents to the road," Paulo Shakarian, a US Army Captain and PhD candidate in computer science at the University of Maryland told me.
In an evaluation of the software's effectiveness in Helmand and Kandahar provinces, it pinpointed the insurgents and their weapons to regions of less than 39 square miles (100 square kilometers), that contained an average of 4.8 villages and a density of high-value targets 35 times greater than in the provinces as a whole.
"What that tells a commander of international security forces in Afghanistan is that four or five villages are the places he can zero in on and bring his other assets to bear," said Subrahmanian.
For example, the commander might decide to get aerial imagery of the villages with an unmanned aerial vehicle (UAV).
"With a UAV, you have to plan out the route, there is only so much fuel, there is only so much of a camera lens to look at things, so you need to be able to reduce the area you are searching for quite a bit in order to make that platform effective," Shakarian noted.
But while the technology helps pinpoint where the insurgents are likely to be found today, won't they just adapt? No, they probably won't, Subrahmanian said, since they are still forced to operate under the same constraints: terrain, roads and tribal affiliations.
"Even if they read our paper, all they know is that we know they have to operate under certain constraints, which they know we know already," he said. "So we don't see this as giving them any kind of advantage of any sort."
What's more, Shakarian added, the software depends on the data input to the system and that information is closely guarded.
In addition to military applications, the technology, which as cost about $300,000 to develop, is being tested for use in identifying animal hosts for certain viruses that spread disease in Africa, Subrahmanian noted.
"The idea is you see where the outbreak occurs and then try to infer back from that locations of the animals that host the viruses that cause the diseases and presumably the public health organizations could look at those regions and decide what action to take," he said.
A paper on SCARE-S2 has been accepted for publication in the 2011 International Conference on Innovative Applications of Artificial Intelligence.
More stories on military software: