Can Computer Models Help to Quell Insurgent-Drive Strife and Instability?

Two scientists who have been using computers to help assess the behavior of insurgent groups told a AAAS-organized seminar that, despite their limitations, quantitative methods could play an important role in helping military commanders and political leaders make more informed decisions.

While acknowledging that computer models are still in their infancy and can provide only hints of possible outcomes during the often chaotic conditions in combat zones, the specialists said the Pentagon has considerable interest in the use of cross-cultural research to help combat insurgencies.

V.S. Subrahmanian, co-director of the University of Maryland's Laboratory for Computational Cultural Dynamics, described how computers can be programmed to automatically and quickly extract relevant data from thousands of news reports on a topic and offer a probability estimate that a particular action might happen. He spoke at a 14 June Capitol Hill seminar organized by AAAS's Center for Science, Technology and Security Policy.

Subrahmanian's team has looked, for example, at reports of suicide attacks by the militant Hezbollah group, based in Lebanon. Preliminary results by the researchers suggest that when the group is engaged in education and propaganda activities in a major way, there's a 46-47% probability it will carry out suicide attacks. When it is not engaged in such activities, the probability of an attack rises to about 80%.

Subrahmanian's team also did an automated analysis of 1555 recent stories in the Afghan media to assess the perceived strength or weakness of Afghan President Hamid Karzai on a scale of minus 1 to plus 1. The analysis, which searched for phrases containing both opinions and statements of fact that can influence opinions, showed Karzai's overall rating was mildly positive for most news sources.

The intensity of opinions can influence how a group might act during times of stress and conflict, Subrahmanian said. Behavioral scientists would like to find ways to accurately predict how a group might respond, he said, and do so in a matter of hours or days rather than weeks or months.

"You would like to find the pressure points where you can exert influence to change behavior," said Alexander Levis, an engineering professor at George Mason University in Fairfax, Va., and former chief scientist for the U.S. Air Force.

The Pentagon, primarily through the Air Force Office of Scientific Research, has been funding research by Subrahmanian, Levis and others in an effort to better understand the potential cultural, religious and social impact of military actions in combat zones. The Pentagon is proposing to spend more than $120 million over six years on social science modeling projects. The goal is to provide a new suite of tools that analysts can use to help assess the emergence of insurgent groups, how they interact with the local population and how they might respond to military or economic interventions. Such models would go beyond the purely military objectives that are at the heart of computerized war games.

Levis said his current work has been funded by both the Air Force Office of Scientific Research and the Office of Naval Research. His open-source academic research builds on similar studies pursued during the past 10 to 15 years by the intelligence community, he said.

The key, according to Levis, is to find ways to fuse the insights from different quantitative models, drawing on the strengths of each. "We need to make sure that the tools play together when used properly," he said. But he stressed that none of the tools will allow analysts to make firm predictions.

"We're not predicting outcomes," Levis said. "We do comparisons of the effectiveness of alternatives."

In one project, Levis and his colleagues developed a computer model that offered U.S. forces in Iraq options for securing two old silk roads in the Diyala province that been attacked by insurgents using improvised explosive devices, or IEDs. The roads, one controlled by the Kurds and the other controlled by Sunni and Shia groups, had long been routes for both legitimate trade and for smuggling of drugs and other covert goods.

When IEDs began to be used along the Sunni-Shia route, the movement of covert goods shifted to the north along the Kurdish-controlled route, Levis said. Suppressing the attacks on one of the routes led to increased attacks on the other, he said.

Rather than deploying additional U.S. forces to help control the IED incidents, the computer model suggested that U.S. commanders should turn over security along the roads to Iraqi forces and allow them to ignore the smuggling of goods along the routes.

"We wanted to maintain the economic activity on the two routes," Levis said.

The preferred option also included educational efforts among the local populace to reduce support for the insurgency, the re-establishment of services, and external financial support to help improve the local economy. The model projected that all of the actions, taken together, would take at least six months of concerted effort to significantly reduce the number of IEDs. It projected that support for the insurgency would decline below 50 percent after about 21 months.

"I have absolutely no clue whether it would take 21 months or not," Levis said. But the model does tell an analyst that the recommended actions would take "not one or two months [but] a whole bunch of months," he said.

Such information can be useful to an analyst who is offering options for policy makers. But Levis cautioned that the numbers must be used appropriately. "This is not a view graph to be shown to the commander," he said, as a guide for planning troop withdrawals after 21 months.

Some of the computer tools have been given to defense and intelligence agencies for testing and user feedback. And some data has been provided to field commanders. Subrahmanian said his team "shipped a bunch of information on several tribes in the Pakistan-Afghanistan borderlands" to the U.S. Army's 10^th Mountain Division before it deployed to Afghanistan in 2006. "So we have valuable results as well as much work to be done," he said.

One long-term goal of Subrahmanian's research is to develop a three-dimensional "virtual experience environment," a cultural and behavioral analogue of a computer war game, where U.S. decision-makers could decide how best to play out the actions of multiple groups in a region. The virtual environment would resemble the real landscape of the region being modeled, with characters behaving according to rules of behavior drawn from relevant and timely data collected by the researchers.

The modelers are just starting to grapple with the complexities of such a virtual environment, Subrahmanian said. They can do visual reconstructions of the landscapes, but they have little real-world content and few real-time news and information feeds. Among the key challenges, he said, is extracting useful knowledge from perhaps a thousand times more data than the team has been using.

While some former military officials have been skeptical of the computer-modeling work, Subrahmanian noted that "we're not the only ones building these immersive experiences." Even Hezbollah has a war-fighting video game that it sells to potential recruits, he said.

Earl Lane

25 June 2007