Center for Strategic Assessment and forecasts

Autonomous non-profit organization

Home / Science and Society / Analytical work: the experience of Russian and foreign experts / Articles
The algorithm can predict a revolution
Material posted: Publication date: 20-02-2014

2013 has provided the students of international conflict, a large amount of material. This year began the civil war in Syria, ethnic violence in China, and riots on the brink of revolution in Ukraine. For the laboratory staff Ward Lab at Duke University, each of whom is a specialist in predicting conflict, the year was like one big bet with many accurate forecasts, as well as those that didn't work out.

When in July the lab Ward, released its semi-annual forecasts, they said 97% probability of unrest in Paraguay that most of it was based on reports of Marxist rebels. Next month, the partisan movement became stronger, confirming the predictions. In the case of the Chinese armed clashes between Uighurs and Han Chinese, the prediction model showed a 33-percent likelihood of violence, even considering the fact that the cause of each case of aggravation of the situation has been concealed by the state media of the country. On the other hand, the unrest in Ukraine was no cause for concern until, has not yet begun active operation, therefore in the report the country was not mentioned at all.

According to the staff of Ward Lab, the aim of the project is not to make predictions and to test theories. If a certain theory of geopolitics can predict an uprising in Ukraine, then maybe she has a future. And even if these experts could predict every conflict, that would be only half the battle. According to the founder of the laboratory and principal research scientist Michael D. ward, "is a success only if it is not reached by the price predictions of many incidents that never happened. But this suggests that perhaps we are on the right track".

Forecasting the future of the countries was not always thus. Usually, predicting revolution or war has been a secret project, for the simple reason that any reliable prediction was too valuable to share. But as predictions lean more on data, it becomes increasingly difficult to keep secret, in connection with the entry into the new generation of publicly available models to predict who oppose the separate status quo.

The history of the forecasting of conflict begins at the Agency defense advanced research projects (DARPA), known as the research wing of the Pentagon. In the 1990s, DARPA wanted to try out the software-implemented approaches to the prediction of how the state will fall in the near future. Over this already worked for the CIA, whose chiefs of staff from each district were regular forecasts, but DARPA wanted to test whether a computerized approach more effective. They asked a simple question: will face if the government of this country with a serious threat to the existence in the next six months? When you pass the test the CIA analysts, they reached about 60 percent accuracy, so DARPA's new system set the bar at 80 percent, looking at 29 different countries in Asia with a population of over half a million people. This program was called ICEWS, a System of early warning about conflicts, and it almost immediately gained success by giving 80 percent by using an algorithm based on simple regression analysis.

Why to surpass the best the CIA analysts were so easy? To some extent, the answer is more linked to the human factor rather than technique. Imagine, for example, the Indonesian expert for the CIA. He wants to make accurate predictions, but at the same time, he is prone to some bias, which will never be identified in the data. He wants his work was exciting and important, helped him to earn the attention of his superiors; he wants Indonesia became significant in the world. Forecasts also are used to manage resources within the CIA, and this Indonesian expert may want to attract more resources, than is usually given to the Indonesian Bureau. By that time, when all will be considered bias, his progress will be a little more than the toss of a coin. On the other hand, the extras don't have to worry about internal politics or insult anyone's feelings.

This lesson was taken seriously in the field of conflict, and she began to become more transparent, even though the tools become more powerful. The ICEWS program was reclassified and placed back in the secret corners of the Pentagon, but now the most fun projects is a fully open program that publishes its forecasts for public viewing. From the data, researchers at Georgetown University are entered every significant political event of the last century into a database called GDELT, and leave all this open access for public research. This database was already used by some projects to analyze the Syrian civil war (eng.) and diplomatic gestures Japan and North Korea, considered the dynamics, which has not been studied ever before. And then, of course, there was a lab Ward Lab every six months, producing a new list of predictions and adjusting their algorithms as technology advances. This is a mirror of the same argument about openness/closure of computer programs – only now, instead of conflicts over program code and control information security tools, this rivalry over who precisely will be able to see the future.

Of course, the secret predictions still exist. Forecasting conflict is a lucrative business for consultants on national security, even if to check their work more difficult. And it is impossible to know what runs the Pentagon. Perhaps their forecasts each month exceed the predictions of the Ward Lab – but even if so, this would not give them a special advantage over publicly available information. And like the CIA analysts, they are closed source and are limited by institutional biases. While the publicly available forecasts are attracting more attention, increasingly subjected to checks and scrutiny, it is easy to sympathize their strengths. As he says ward: "you See, the world is confusing and complicated. I don't think will ever be a time when everything can be predicted quite accurately". Meanwhile, the trick is to continue to improve.

Source translation for MixedNews -


RELATED MATERIALS: Science and Society