CoupCast: An AI Predicts Revolutions

For many years and several moves, I have kept a small collection of issues of Newsweek magazine that were published in 1989. That was not only the year I moved from high school to college, but also one where the world had changed in close proximity. The population of Austria’s communist neighbors got rid of their rulers largely without bloodshed, the Soviet Union disintegrated, and in China a student demonstration was bloodily put down in Tiananmen Square.

I could hardly have guessed the significance that these revolutions would have for me personally and for the world, and yet I kept these magazines, because a world order that was believed to be immutable suddenly rearranged itself and many innovations came. They led me to come into contact with many people from these countries, they became colleagues and friends, and I have traveled to quite a few of these countries since then.

These changes had hit many experts and the world unexpectedly. Back then, the TV program didn’t start until 9 a.m., often with a break after noon, and ended at night. But this year, one special broadcast chased the other. Early in the morning, we were already turning up the TV to see the next breathtaking developments. Which country was next? Which dictatorship had now been deposed? Who was taking power?

My collection of Newsweek issues from 1989

However, it became clear from the statements of political analysts in the media and the actions of politicians in all countries that no one had foreseen this and thus no one had a plan of what to do. Many had come to terms with the situation of the world being divided into two power blocs, it had been predictable, one knew who one was dealing with. But that had changed abruptly. How to proceed, does the military intervene, what happens to the nuclear warheads, what do new rulers demand, how to help with a transition to democratic regimes, who pays for all this and and and. Questions over questions and no one had answers, spontaneous actions had to be taken, and what was valid this morning was old hat by the afternoon.

33 years later, an artificial intelligence from the University of Central Florida now promises to more accurately predict such revolutions and regime changes. CoupCast, as the AI is called, was started in 2016 by the nonprofit One Earth Future, and the Jan. 6, 2021, uprising at the U.S. Capitol now showed the system’s significance. Revolutions had arrived in the U.S., and researchers wondered how they could detect early signs of impending coups and assess the likelihood.

The researchers, who had been working on this for some time, collected data on the possible drivers of revolutions around the world since 1920 and tried to quantify them using parameters. They looked at a country’s democratic history, democratic “backsliding,” economic fluctuations, “social trust,” traffic disruptions, weather fluctuations, the profile of rulers such as their age and military background, infant mortality, gross national product, time between election dates, longevity of regimes and other factors.

Da KI was then trained with regression models in an autoregressive manner. First, a model based on data between 1950 and 1974 was trained to predict a coup risk for 1975. The predictions for 1975 were added to the dataset and the model was trained again to predict risks for 1976, and so on to the present. As a result, insurgencies and coups in Chad and Mali were predicted for 2021.

Until January 6, 2021, however, this AI model had barely considered the United States, but with the insights gained from the events, the model was adapted and expanded. For example, another nonprofit, PeaceTech Lab, plans to use Ground Truth to predict the likelihood of violence and riots around elections starting in 2022.

The researchers consider, among other things, the choice of words in posts on social media, but also the number of dogs on the streets before an election. More dogs on the streets could be a sign of expected violence, and people try to protect themselves from violent attacks through their dogs.

While the Pentagon, CIA and US State Department are the interested parties for this AI, the FBI and US Department of Homeland Security show little interest in it yet. The US military uses its own system to make such predictions.

Predicting revolutions and upheavals, however, remains a problem. Small data sets pose great challenges for AI, but because of the great dangers they pose, they can be helpful support in diplomatic and political contexts to make appropriate preparations that can avert humanitarian tragedies.

Leave a Reply