WHEN Israeli troops pulled out of Gaza last week, the Israeli-Palestinian conflict seemed as intractable as ever. More than 100 Palestinians were killed during the incursion. Shortly after, a lone Palestinian gunman killed eight Jewish students in Jerusalem. As the dust settles, politicians in Israel, the US and the Palestinian territories are debating what to do next. Further attacks could make the situation worse, but talks might come to nothing. Which path should they take, and are there any other options?
What if there were a method, even a crude one, that could provide odds on what is likely to happen next? Since the 1960s, political scientists have used computers to identify the conditions that lead to violence. Thanks to the availability of online news reports, they now have far more data to draw on. And cheaper, faster computer power is giving them the ability to drill deeper into that data.
These changes have led to conflict simulations with promising levels of accuracy. If the specialists’ hopes prove justified – and many think they will not – we are about to enter an era of unprecedented political forecasting. “There is so much potential there,” says Philip Schrodt from the University of Kansas in Lawrence. “In five years time I will be able to say: ‘Here is a really great prediction about Israeli-Palestinian violence’.”
Advertisement
Political scientists say their work can help spot countries that are about to descend into violence, or to rate hypotheses about the future tactics of opposition groups. Schrodt says these simulations can be best compared to economic forecasts. “They’re not always right, but they’re right enough of the time that they’re taken seriously,” Schrodt says.
Existing methods have already produced good results. Last month’s riots in Cameroon, for example, were forecast by Monty Marshall at George Mason University in Arlington, Virginia, and colleagues in 2005. The prediction was surprising at the time because Cameroon had been relatively stable for years. Marshall’s analysis of African conflicts found that leaders who have been in power for more than 15 years are likely to face violent opposition. The model flagged Cameroon as vulnerable, in part because President Paul Biya has led the country since 1982.
“Cameroon has been relatively stable for years, but last month’s riots were forecast in 2005”
Results like that used to be generated by human researchers, who would trawl through news reports, surveys and other intelligence sources to identify whether countries were experiencing the conditions that tend to lead to conflict. These include poor healthcare, the level of democracy and the state’s openness to trade.
In recent years, computers have played a larger role in these efforts. The US government, for example, funds the CIA’s Political Instability Task Force, which produces classified computer forecasts of which countries violence is likely to erupt in. These attempts are likely to become far more detailed now. As well as focusing on individual countries, several research groups are drilling further down and examining political groups, such as Hamas.
In the vanguard of the search for improved accuracy is Steve Shellman at the University of Georgia in Athens, US. Over the past two years, his team has been analysing news reports from the likes of Reuters and the BBC on 29 Asian countries between 1980 and the present. Each story is broken down into its essentials of who did what to whom, and when.
The researchers start by defining the list of “actors” that the software should look for, such as heads of state and opposition leaders. They also tell the computer how to identify specific actions, by looking for phrases such as “held talks with” or “launched strikes against”. Then the software takes each story, line by line, generating data that links actions to actors. Some phrases cause problems. An “attack” could be verbal or physical, for example. In cases like these, the software tries to differentiate between the two meanings by examining the rest of the report.
The result is a quantitative record of the region’s recent history. The researchers can then mine this for trends. Using data on political groups in Bangladesh, for example, Shellman’s team has shown that repressive government policies, such as violent clampdowns on political groups, have side effects. When many political groups were repressed at once, formerly peaceful groups turned violent. So, while repression might seem attractive to governments, the model shows it often backfires.
Shellman then fed the data from all 29 countries into models designed to predict conflict. These are based on existing theories, such as the importance of democracy in preventing strife. Then he ran tests. By feeding data from 1980 to 2004 into one model, for example, he generated predictions for armed insurgencies in 2005 that could be checked against what actually happened. He did similarly for many of the years in between, and across all 29 countries, his predictions came true 8 times out of 10. Now he is working on predictions about events still to come.
Similar levels of success are being claimed by researchers led by V. S. Subrahmanian at the University of Maryland in College Park, where they have developed software that analyses the activities of political groups within areas of conflict, including Hamas and Hezbollah. Their model tracks links between a group’s circumstances and its subsequent actions, using data from as far back as the 1950s. The model found, for example, that when Hezbollah solicits support from foreign organisations and governments it is more likely to engage in kidnapping.
Aaron Mannes of the University of Maryland is also using the model to develop predictions for the coming year. His forecasts are not yet complete, but at the request of New Scientist he took a look at what the model says about the present violence in Gaza. One prediction is a likely Hamas decision to kidnap another Israeli soldier. Some thought kidnapping would decrease once Hamas assumed political power. But Mannes’s analysis found that the number of kidnappings increased with the number of social service roles taken on by Hamas.
US officials think such advice could be useful – hence the source of funding for some of these models – the US Department of Defense. But until such models produce forward-looking information that can be independently tested, many researchers remain sceptical. “There have been many attempts to predict the behaviour of foreign leaders,” says Gawdat Bahgat, an authority on Hezbollah at the Indiana University of Pennsylvania. “It is extremely hard because there are so many things we do not know. I would not put my money on their predictions.”
Uncertainties in the data are key problems. For instance, trying to include data on foreign support for some movements is tough to measure. “There are so many claims and counter-claims [in news sources] about what different groups are doing,” Marshall says.
Marshall has another concern. Some models, such as Shellman’s, focus on the actions of governments and opposition groups. But Subrahmanian’s focuses on predicting the behaviour of individual groups, some of which are viewed by the US as terrorist organisations. “The objective of these applications is biased – it is about learning to control these groups,” says Marshall. He says it is better to understand the entire political and economic picture. Attempts to focus only on your enemies can be counterproductive. In Iraq and Afghanistan, for example, it is not clear whether attempts to eradicate armed groups have reduced the level of violence, argues Marshall.
Mannes accepts that his model can be used by policy makers intent on controlling political groups rather than understanding them. But he says it could equally well be used to determine when best to negotiate. And, he adds, “if I were a policy maker I would want whatever tool I could get”.
Which raises one final problem: if policy makers do use these tools, modellers could become victims of their own success. Providing false leads to an enemy is as old as war itself. If a group knows that models are taken seriously, they will try and twist the results. According to Schrodt, this means the public is unlikely to ever know how much use governments make of the new forecasting approaches. “If the government has a really good model, they won’t want to talk about it,” he says.
