The short explanation is that the computer geniuses I work with developed something called SOMA that, given data on a group's past behavior, can develop rules for probable future behavior given different conditions. There are many potential applications. An immediate one was for me (as the resident terrorism specialist) to examine the rules and see if they taught me anything I didn't already know about a group. The findings about Hezbollah discussed below are based on a case study we did on that group which I presented at the First Annual Workshop on Social Computing, Behavioral Modeling, and Prediction in April. At the link above you can find my presentation slides and the slides of other presenters. The conference is a window into how computer science is bringing new tools to social science, some of which have useful applications to counter-terror, counter-insurgency, and development. Just as the Cold War spawned new technologies that had profound impact on day to day life (like the Internet), so to will the present conflict.
I'm presenting a similar analysis of Hamas at an upcoming conference at the University of Maryland, the Second International Conference on Computational Cultural Dynamics.
Since the paper was published, we've tweaked our analysis and tools - but the key findings have stayed consistent.
Calculating the Margin of Terror
By Joseph Straw
How likely is it that Hezbollah in Lebanon would launch suicide attacks or engage in kidnappings during an election year? Professor V.S. Subrahmanian, head of the University of Maryland’s Institute for Advanced Computer Studies (UMIACS), says that the right data combined with the right software can answer that question.
Subrahmanian has taken a decade’s worth of open-source data on political and terrorist activity in the Middle East and plugged it into his own behavioral modeling software program—which he calls a stochastic opponent modeling agents (SOMA) framework—to create the SOMA Terror Organization Portal (STOP).
The program, funded by the Air Force Office of Scientific Research (AFOSR), is currently in use by four defense agencies. Via a simple, Web-based portal, users select a country, a terror group, and an environmental variable (usually political), and receive a percentage probability for activity during a given period. For some groups and countries, users can select a specific activity and motivation, such as kidnapping for ransom versus kidnapping for political demands.
Subrahmanian cautions, however, that the program’s data inputs and the Boolean logic behind the modeling are extremely coarse. Consequently, the program’s predictions carry a statistical margin of error far beyond that of a typical public opinion poll. Thus, STOP does not offer hard statistical predictions, but ranges, for example, a 68-87 percent chance of activity in a given period.
Despite the margin of error, the modeling has generated some valuable findings, says terrorism scholar and UMIACS researcher Aaron Mannes. For one, it has shown analysts that when a political group with ties to terror runs candidates in a national election, attacks spike in the period around the election. For that entire year, however, attacks by the group drop.
STOP’s data universe relies on existing research by the university’s Center for International Development and Conflict Management. It considers 31 terror groups operating in countries from Morocco east to Afghanistan, and considers roughly 150 action variables within those countries.
James Jay Carafano, counterterrorism expert at the Heritage Foundation in Washington, D.C., says Subrahmanian’s work is pioneering, although not groundbreaking. Separately, the CIA has modeled international political activity for more than a decade in hopes of predicting instability. The field is still maturing, but research indicates that the technology holds great promise in forecasting group action.
“Technology is increasingly making soft sciences, like anthropology or political science, into hard sciences. People are quantifying a lot of these things, making them interactive, and producing results,” Carafano says.
Subrahmanian explains that while free will makes it impossible to predict individual action, it is possible to look at group dynamics and make predictions with some degree of accuracy.