If we assume that world affairs are Gaussian, i.e., shaped like the bell curve, then nasty surprises can happen, but their likelihood is so rare that one might rationally argue that it is a safe bet they won't. That argument depends to some extent on the penalty to be paid, so perhaps in a nuclear world, war would still be risky, but one could argue that conventional wars can be contained.
But there is actually no justification for blindly assuming even that Gaussian distributions are "normal," (though that is what they are called) and even if they were typical across all arenas of existence, that would still in no way prove that such distributions characterized any particular class of endeavor, say, war.
The distribution of outcomes of social phenomena cannot be assumed to follow any particular pattern; it depends. Therefore, the level of uncertainty also depends: the likelihood of winning cannot be assumed.
The distribution of outcomes for any given social phenomenon might be fractal or, following Taleb (The Black Swan, p. 272), a gray swan - i.e., a surprise that you certainly do not expect but do realize might occur. We never see stock market crashes or the bursting of a housing bubble coming, but we all know that they do come; it's the timing, not the existence in principle of such a class of events that surprises us. And since fractal distributions are scale-invariant (more or less, within some uncertain but probably quite large range), all bets are off about the size of the next event. Nevertheless, realizing that the class exists and big ones can occur in what certainly appears to be a random sequence tells you something about risk in that particular field of endeavor. Participate in that class of events with your eyes open, if at all.
Then, there are the Black Swans - the inconceivable ones. Statistically, there does not seem to be much useful to say about a one-time event you haven't even thought of. A global affairs example might be Aum Shinrikyu's poison gas attack on the Tokyo subway.
The existence of wars is obviously predictable; indeed, we start them intentionally. The problem is in thinking that their outcomes are Gaussian. Maybe the side with more weapons in fact did typically win (historically); that does not prove such will continue to be the case in a rapidly evolving world. But the problem is worse. The historical distribution of wars, particularly given the difficulties of comparing wars now with wars in very distinct historical eras, is pretty small to use as the basis for judging distribution in the first place, but clearly some huge ones do occur, and the size and outcome of wars tend to be poorly predicted. War outcomes seem a lot closer to fractal than Gaussian - and thus bad bets.
The problem with getting the desired outcome through war is even worse than this, however. Put aside theoretical, mathematical considerations, and think about the social context, the endless interrelationships in which wars occur. War is the perfect catalyst for Black Swans. War is the perfect context for the generation of some new, unthought-of form of behavior. In the era of rapidly rising global social complexity and new technical capabilities for waging asymmetric warfare, those who want security will be wise to avoid chaotic situations, such as war, that maximize the likelihood of surprise. Chaos favors those who have the least to lose.