How do you predict the unpredictable?
Analysts have spent careers trying to anticipate dips and surges in the stock market, and meteorologists have likewise struggled to tame long-range weather patterns. But the imperfect science of foreseeing when terrorists or enemy factions might strike has proven an especially daunting challenge.
Researchers at the University of Arizona in Tucson have now begun work on a set of computer algorithms that may be able to make sense of mountains of intelligence data that would overwhelm human analysts. Known as the Asymmetric Threat Response and Analysis Project, or ATRAP, the effort is aimed at dispassionately sifting through everything from fingerprints to cultural influences to establish useful links and connections.
Jerzy Rozenblit, head of the electrical and computer engineering department at the University of Arizona, said the main goal of ATRAP is to avoid conflicts by identifying and resolving unstable situations before they blow up into bigger problems. Equipped with interlinking computer algorithms that analyze apparent patterns, the program could ultimately predict the courses of action for terrorists, criminal groups, ethnic factions and other destabilizing forces. That information, in turn, would help military commanders devise the best strategies for overcoming enemy combatants.
Rozenblit stressed that the program also could prove invaluable for civilian applications, such as countering identification theft, predicting the course of disease outbreaks and helping communities recovery more quickly in the aftermath of natural disasters such as Hurricane Katrina.
“The first part is to ingest massive amounts of data acquired by intelligence, sensors, satellite data, data from the Internet, images, you name it,” Rozenblit said. “And once we are able to vacuum it up, we need to not just ingest the data, but also digest it.” After that initial cataloguing, the next hurdle is taking the massive body of information “and trying to discern connections among the points to a level at which patterns might appear.”
One example of a bottom-up approach to determining links and patterns has come from Iraq, where GIs patrolling neighborhoods began noticing that as they approached specific intersections or areas, many of the inhabitants covered their ears. “Clearly this was an indication that there was a bomb that was planted, and they (the locals) would learn to cover their ears,” Rozenblit said. The bombers have since changed their tactics, but GIs who recognized the pattern at the time and immediately stopped their vehicle undoubtedly saved numerous soldiers’ lives.
Similarly, most fraudulent transactions yield a series of events that can be linked by data. Identifying a common denominator in the pattern could allow law enforcement officials to zero in on a criminal ring. “Once you know how that game is being played, you can look for indications that it may be happening again,” said Brian Ten Eyck, the university’s ATRAP project manager. And once a pattern is discovered, it could be fed back into the database to refine it and further draw out the “spider web of connections” that might link people, groups or actions.
A third element of the project will be its ability to visualize the connections in a meaningful, user-friendly way. A dynamic bar graph that vacillates between red and green, for instance, could convey the amount of fuel left in a vehicle, the density of forces in a geographic region or the status of an organization’s bank account.
If all goes well, ATRAP’s final product will be a sophisticated prediction of enemy courses of action. Calculating the likelihood of specific actions in advance would enable commanders to proactively counter with actions of their own. An initial $2.2 million for the four-year project has been funded by the Army Battle Command Battle Laboratory at Fort Huachuca, Ariz., which will handle the classified portion of the collaboration.
The system is designed to respond to changing conditions, Rozenblit said, just as IBM’s Deep Blue computer recalculated the playing field after every move by world chess champion Garry Kasparov. Even so, when Deep Blue beat Kasparov in a famous 1997 chess match, the computer had yet to fully master the game. “The number of possible moves is staggering large, but it is finite,” Rozenblit said. Given enough resources, a computer could eventually chart an unbeatable course because every move is bound by the rules of the game.
‘There are no rules’
It’s one thing to move a queen on the chessboard, however, and quite another to move security forces in an unstable part of the world. “In the situations that we deal with, there are no rules,” Rozenblit said. In the real world, “winning” also could mean creating a stable environment where multiple opposing sides all gain something or maintaining a truce that would lead to a penalty if one side disturbs the status quo.
Rozenblit said his participation in the project was motivated in part by observing the instability in Kosovo and Somalia during the 1990s, when the co-existence of competing ethnic, criminal and other factions thwarted attempts by U.S. military and U.N. peacekeepers to quell the violence and chaos. “The nature of conflict has changed — we are not operating in the classical ‘us versus them’ environments,” he said.
Rozenblit, Ten Eyck and their colleagues have since borrowed algorithms from the fields of mathematics, genetics and economics to help make sense of courses of actions that might appear bizarre or defy logic to most people but that ultimately serve a purpose. Genetic development models, for example, seek to explain how groups or individuals mutate in response to a changing environment to maximize their fitness.
Similarly, co-evolution supposes that an individual’s actions are geared toward claiming “king of the hill” status for themselves or a larger group. Other individuals could adapt by observing the winning strategy and adopting their own game plan to boost their self-interests. Eventually, though, the multiple glory seekers might reach some sort of compromise or equilibrium so that everyone benefits a bit. No one would reign as "king of the hill," but everyone could at least claim their own small reward.
Finding the best payoff
Game theory invokes what are known as zero-sum games that don’t reward anyone — like repeated tic-tac-toe stalemates between equally matched players. As with co-evolution, establishing some sort of equilibrium won’t maximize an individual’s success, but it could provide the best payoff for everyone. Cell phone providers and airline carriers often settle on a similar amount for equivalent plans or flights, respectively. Charging more would mean a loss of business to their competitors, while charging less would cut into their profits.
Nichole Argo, a principal investigator with the Jebsen Center for Counter-Terrorism Studies at Tufts University, said she considers the sheer magnitude of ATRAP’s data-analyzing power less impressive than the breadth of the project’s approach toward understanding conflicts such as terrorism and civil war. Instead of collecting vast amounts of data, Argo said, the real trick is gathering the most relevant bits. “Science’s problem tends to be asking the right questions,” she said.
In that vein, she said she was pleased to see the University of Arizona researchers gathering information that might fit competing theories. For example, Argo said that new psychology and social neuroscience research supports a co-evolutionary perspective as a better explanation for terrorism than traditional game theory, in part because it prioritizes individuals’ attachments to their social groups and recognizes how their motivations may change accordingly. An openness to testing data within those and other frameworks, she said, “is probably where they’re going to get the biggest bang for their buck.”
Rather than identify all possible links and patterns, ATRAP’s Ten Eyck said the project would allow people to corroborate their hunches and seek out more information. Future versions of the program will even have the ability to detect intentional deception by enemy factions or forces.
One key application may be in conducting “What if?” exercises and virtual disaster training within emergency response centers. The system is also designed for real-time situations that require swift data processing and has benefited from the advances in parallel computing that are allowing more computers to simultaneously tackle a given problem. The added speed could prove vital for military commanders or disaster responders needing to quickly analyze and respond to reams of incoming data.
In post-Katrina New Orleans, a volatile mix of criminal gangs, volunteers, police, hospital staff, the military, the media and residents all co-existed in a deteriorating environment. If the system lives up to its potential, ATRAP could one day be used by first responders and agencies such as FEMA to stabilize similar situations, maximize the chances for a quick recovery and perhaps help answer the most important question of all: “How do we prevent this from ever happening again?”