Conflict is inevitable, but with the right computational tricks, we ought to be able to see it coming. Social scientists have been trying to build that particular crystal ball for almost a century. Now database records, eyewitness accounts, and other data points are making such predictions look possible. And just in time. After the September 11 attacks, figuring out where the next trouble will be has taken on new urgency. Here's a report from forecasting's front lines.

Correlates of War, U. of Michigan
Encompassing every major conflict from 1815 to the 1990s, this database quantifies size of conflict, countries involved, and political interactions. Findings: Small nations are more likely to go to war than big-leaguers. The more alike two countries in a dispute are, the better the chance for shooting.
How good is it? Not very. The project is more dedicated to developing an etiology of war.
The catch: COW's so-called structural data doesn't account for rapid, chaotic action.

State Failure Task Force, U. of Maryland
This database includes nearly every revolution, war, violent regime change, and genocide from 1955 to 2000. The model uses statistical techniques like neural network analysis and genetic algorithms.
Findings: Nations with low openness to trade, weak democracy, and high infant death are vulnerable.
How good is it? Good. The model reproduces - retrodicts? - history with 80 percent accuracy.
The catch: Other researchers have found holes in the statistical methodology.

Integrated Data for Events Analysis, Harvard U.
An AI that can read. Fast. Its algorithms parse the lead sentences of 80,000 Reuters news articles each month to figure out who's doing what to whom.
Findings: Patterns of innocuous events such as nonviolent protests can signal impending violence.
How good is it? Good enough to discern week-to-week patterns. Virtual Research Associates sells the engine to early-warning groups.
The catch: The English-language press has its limits. Other groups are broadening the data set.

Country Indicators for Foreign Policy, Carleton U.
This project combines a structural database of 100 performance indicators from 196 countries from 1985 to 2000, events data from the Swiss Peace Foundation, and reports from on-the-ground observers.
Findings: Five countries in West Africa and three in Southeast Asia are at significant risk for violence.
How good is it? Probably very. CIFP uses simplified stats to help policymakers and NGOs avert trouble.
The catch: If you ensure the predictions don't come true, how do you know you were right?

By Adam Rogers

 Copyright 2001 by Wired Magazine.