Washington, DC, gets blown up all the time — in the movies. But what if it happened for real? Hopefully, we’ll never find out. But federal emergency officials have to prepare for the worst. Emergency responders would have to deal with shattered buildings, rubble-filled streets, downed power lines, fried cell phones, radioactive contamination, and casualties numbering in the hundreds of thousands. It would be a disaster beyond anything ever seen in an American city — certainly beyond anything that emergency managers have ever experienced. So, to help them figure out what to expect in such a catastrophe, researchers from Virginia Tech university have built an entire digital DC inside a computer — and have blown it up over and over and over. In an effort to make their model as accurate as possible, they start with the basic geography of Washington. Then they add in every building … Every road… Every power line… Every hospital… Even every cell tower. Next, the modelers populate their virtual DC with millions of digital people, known as ‘agents’. These agents are like members of the crowd in a computer-generated scene in a movie: They can move around the landscape and react to each other and their surroundings in human-like ways. But unlike CGI crowds, each agent pursues its own goals. they can switch between goals like ‘run’, ‘hide’, ‘help out’, and ‘find loved ones.’ Finally, the modelers simulate the effects of a 10-kiloton nuclear bomb that explodes on a Monday morning in May, just a few blocks north of the White House. They use standard models of blast effects to estimate casualties and damage to buildings. They use weather models to track the spread of fallout. But they allow the agents to react to the disaster as individuals. This gives the agents’ bottom-up behavior the kind of spontaneity that’s so often seen in real life — and that is almost impossible to model in any other way. It also gives emergency managers a much better handle on the surprising and obstinate ways that people often behave in a crisis. For example, if the authorities warn everyone to find shelter and stay there until the fallout dies down, the model shows that many people will ignore them. Instead, they will be out on the roads, desperately trying to pick up their kids from school, or find missing spouses. Many of them may even end up traveling toward ground zero. But the model also shows that the authorities can minimize this chaos if they move quickly to restore cell service: if people can verify that their loved ones are safe, they are more likely to stay where they are. This disaster-response simulation is just one example of what ‘agent-based’ models can do. More and more real-world decision-makers are using these models to plan new highways, improve the flow of crowds in public places, plan how to respond to epidemics — or even minimize the chances of a stock market crash.