Since the 1970s, the United States intelligence community (mainly the Department of Defense and the CIA) has been trying to build computer simulations that will accurately predict wars, revolutions or, currently, terrorist attacks. Currently, the United States is spending about $40 million a year on this sort of thing. Politicians and government officials complain that these simulations don't work.
Actually, the models do work, but not in the way most people think. The problem is that the models can predict what large groups of people are likely to do, over a period of time (months or years). They cannot, and never claimed to, predict what individuals will do over shorter periods (days or weeks). Thus many models predicted that Egypt, Tunisia, and most Arab nations were unstable and headed for unrest and revolution. But because these models did not name a specific day or month, they were considered a failure. Models like this have long been used in marketing and financial markets. Users of these models know that they are simulating the movement of "markets" (large groups of people), and not in the kind of detail news directors and editors need to get a headline.
The models do not fail. In fact, the best of them constantly update their predictions as they monitor the news (economic, political, military) coming out of a nation. Moreover, the models, like weather forecasts, have gotten better and better over the last few decades. But don't call them failures because they didn't do what they were not designed to do, instead of what you wanted them to do.