Probability and predictability

Some aspects of predicting the future, such as celestial mechanics, have been discovered to be highly statistically predictable, and may even be described by relatively simple mathematical models. At present however, science has yielded only a special minority of such "easy to predict" physical processes. Theories such as chaos theory, nonlinear science and standard evolutionary theory have allowed us to understand many complex systems as contingent (sensitively dependent on complex environmental conditions) and stochastic (random within constraints), making the vast majority of future events unpredictable, in any specific case.
Not surprisingly, the tension between predictability and unpredictability is a source of controversy and conflict among futures studies scholars and practitioners. Some argue that the future is essentially unpredictable, and that "the best way to predict the future is to create it." Others believe, as Flechtheim, that advances in science, probability, modeling and statistics will allow us to continue to improve our understanding of probable futures, while this area presently remains less well developed than methods for exploring possible and preferable futures.
As an example, consider the process of electing the president of the United States. At one level we observe that any U.S. citizen over 35 may run for president, so this process may appear too unconstrained for useful prediction. Yet further investigation demonstrates that only certain public individuals (current and former presidents and vice presidents, senators, state governors, popular military commanders, mayors of very large cities, etc.) receive the appropriate "social credentials" that are historical prerequisites for election. Thus with a minimum of effort at formulating the problem for statistical prediction, a much reduced pool of candidates can be described, improving our probabilistic foresight. Applying further statistical intelligence to this problem, we can observe that in certain election prediction markets such as the Iowa Electronic Markets, reliable forecasts have been generated over long spans of time and conditions, with results superior to individual experts or polls. Such markets, which may be operated publicly or as an internal market, are just one of several promising frontiers in predictive futures research.
Such improvements in the predictability of individual events do not though, from a complexity theory viewpoint, address the unpredictability inherent in dealing with entire systems, which emerge from the interaction between multiple individual events.