AI models are being used to predict conflict

Subscribe to enjoy similar stories.AS AN UNEASY truce holds between America and Iran, experts are struggling to predict what new phase the conflict may enter next. Might an artificial-intelligence model know any better? To find out, The Economist asked RAND, a think-tank, to see if its new AI forecasting system thought a popular uprising was in the offing in Iran. Integrated Strategic Forecasting (ISF), as the system is known, put the chance of regime collapse or replacement by the end of 2026 at 20%—higher than many experts would hazard.There are caveats.

The forecast was produced without classified intelligence. Its inputs and outputs were also not vetted by humans, as is customary for forecasts commissioned by government agencies, notes Anthony Vassalo, RAND’s head of prediction technology and a former senior official in America’s Office of the Director of National Intelligence. Even so, Mr Vassalo is bullish.

He describes ISF, which was completed in February, as “the better crystal ball” policymakers have long sought.And ISF is not the only game in town. After two decades in which attempts to build conflict-predicting computer models have yielded disappointing results, recent advances in machine learning and large language models have prompted many data scientists to take another crack. The idea is simple.

Models trained on past conflicts are fed indicators that may signal future strife, in the hope that predictive patterns invisible to humans will emerge. Inputs include data on crime, public health, labour strikes, weather, the economy and political developments such as democratic backsliding. Social media are widely mined to gauge discontent.