When it comes to complex systems, the ability to predict future behavior becomes very difficult.
Taking games, for example, chess and other games have a limited set of plays and outcomes. As a result, a supercomputer is able to learn all these and defeat human chess masters. Games like poker are trickier as the plays and outcomes depends heavily on luck. But it is still confined to certain limits. If a graph is plotted, simple games come out to be more or less linear.
Opposed to such games is how human social interaction occurs. Phenomena such as the stock market are complex systems composed of human social interaction. In such systems there are no rules in which games are played. The graphs for such systems follow power laws: the progress is slow in the beginning but after an inflection point, the curve turns radically. In other words they are exponential. Predicting outcomes in such systems is virtually impossible.
Sometimes, one comes across some models which attempt to predict the behavior of a system based on past data. In order to come up with such models, the “outliers”, or data which deviate very far from the mean, are often left out.
However, it could so happen that in many systems, the entire behavior may be totally determined by the outliers. Also, it is difficult to choose an outlier and to understand how it could impact the system. So, much of the predictive modeling and “forecasting” turns out to be useless or even dangerous because they tend to mislead the human brain.
Weather predictions for the next few days are known to be pretty accurate. However as the days turn to weeks, the accuracy goes down dramatically.
It seems that to predict accurately, one needs to know and understand the rules of the game. As things get complex, take the climate for example, the rules are more difficult to understand. So, it becomes hard to pinpoint how climate will change in the future. However, we need to not mess with nature and risk causing any more global warming. This could be done as a matter of applying the precautionary principle.