Although the outbreak of war is usually viewed as an irrational act of violence, the risk of war can be calculated.
The outbreak of wars, riots, strikes, and unrest can actually be predicted – with sober scientific methods, political and social theories, but also data analysis and artificial intelligence. In order to make their models faster and more accurate, researchers have the software automatically evaluate articles from daily newspapers.
Unsurprisingly, one of the pioneers of automated conflict prediction is the US government, which launched a corresponding competition in 2008, which was won by Lockheed Martin. However, technical details about the system are not known. In the meantime, various states have followed suit, and the federal government operates two systems: one in the Foreign Ministry and one in the Ministry of Defense. According to the press release, the data for the model of the Ministry of Defense is “from publicly freely accessible sources”, which the Ministry “supplements with higher-ranked sources”. In contrast, the Ministry of Foreign Affairs has added a text mining module to its software, which combs through UN resolutions to understand how countries’ voting behavior is evolving.
The probability of hitting the computer models globally is around 80 percent. However, there are big differences. The probability that conflicts will break out again once there has been a bang somewhere is drastically increased for around ten years. It is more difficult to predict conflicts that arise where there has not yet been a bang.
Tackling the “difficult cases” of conflict prediction
Hannes Müller from the Barcelona School of Economics and Christopher Rauh from the University of Cambridge have therefore joined forces in the conflictforecast.org project to specifically tackle these “difficult cases” of conflict prediction. Their idea: They let algorithms evaluate newspaper articles – and then use this evaluation to calculate a conflict probability. The algorithm checks which keywords appear in the text and how often, and can thus assign it to the main topic. The second part of the model, which learned the relationship between the frequency of certain topics and the outbreak of armed conflict, consists of a random forest.
Irrespective of the methodological subtleties, the researchers, who also work with the Foreign Office or the Federal Foreign Office, are also struggling with a political problem: “Politicians often focus on areas that are already hot,” says Christopher Rauh. “In our opinion, however, it is much more important to intervene where nothing has happened yet, but something could happen.” In politics, on the other hand, there is a strong tendency “only to look where the fire is burning,” the researcher complains. “It’s difficult to force a rethink here.”
“We’re practically looking for a needle in a haystack,” says Hannes Müller. “But even if we are wrong in most cases where we warn of an impending civil war, it is still worth intervening in all such cases. You have to realize that there are currently 80 million people on the run who are facing wars. 30 million of them are children. A civil war is something like a nuclear accident on a political level.”