Le mieux est l'ennemi du bien

I found a particularly interesting article today, linked here. It is a speech, given in 1954, that detailed the trade-off between good security and total security: There is something about this quest for absolute security that is self-defeating. It is an exercise which, like every form of perfectionism, undermines and destroys its own basic purpose. The French have their wonderful proverb: Le mieux est l'ennemi du bien--the absolute best is the enemy of the good. Nothing truer has ever been said . A foreign policy aimed at the achievement of total security is the one thing I can think of that is entirely capable of bringing this country to a point where it will have no security at all. And a ruthless, reckless insistence on attempting to stamp out everything that could conceivably constitute a reflection of improper foreign influence in our national life, regardless of the actual damage it is doing to the cost of eliminating it, in terms of other American values, is the one thing I can think of that should reduce us all to a point where the very independence we are seeking to defend would be meaningless, for we would be doing things to ourselves as vicious and tyrannical as any that might be brought to us from outside. This sort of extremism seems to me to hold particular danger for a democracy, because it creates a curious area between what is held to be possible and what is really possible--an area within which government can always be plausibly shown to have been most dangerously delinquent in the performance of its tasks. And this area, where government is always deficient, provides the ideal field of opportunity for every sort of demagoguery and mischief-making. It constitutes a terrible breach in the dike of our national morale, through which forces of doubt and suspicion never cease to find entry. The heart of our problem, here, lies in our assessment of the relative importance of the various dangers among which we move; and until many of our people can be brought to understand the what we have to do is not to secure a total absence of danger but to balance peril against peril and to find the tolerable degree of each, we shall not wholly emerge from these confusions. This same thought is echoed here, in a review of Worst Case Scenarios by Jeremy Waldron, who also references the One Percent Doctrine: The One Per Cent Doctrine: it’s a striking methodology and a liberating one, and many people think it’s the only way to respond to the threat of low-probability, high-impact events. With it, the endless evidence-gathering and analysis that characterises traditional intelligence policy gives way to clarity. Nothing any longer needs to be conditional. We no longer say, ‘If X has happened, then we need to do Y,’ with all our effort being devoted to finding out whether X has in fact happened or (in an uncertain world) what its probability is. Instead we say, ‘If there is the smallest significant chance that X has happened, then we have no choice but to do Y.’ If X may lead to a catastrophe that must be avoided at all costs (like a nuclear attack on an American city), then we need to swing into action immediately and do Y. No further questions. While this has obvious and large-scale implications on geo-political mechaniations, the problems with low probability, high reward analysis are pervasive. In the business sector, for example, I have seen many times the following situation: -We want to create a financial projection for the coming year, based on a series of events that may or may not happen. -Each of these events has a particular cost, and a particular benefit. -Probabilities of success are attached to each event. -The value of the event is multiplied by the probability of success, to achieve the "projected value" of the event. -The sum of projected values becomes the financial projection This type of analysis works well in "conventional" situations, where the probability of success is reasonable: many events with probabilities of success in the 40-70% range will usually pool and even out, coming close to the sum of the weighted values. In the case of very low probabilities, however, this analysis breaks down. If there is a $10,000,000 opportunity with a 1% chance of success, this is typically projected at $100,000. However, unless there are many of these types of events (there never are, btw), then the statistical pooling never occurs, and the 1% probabilites approach 0%. Therefore, the $100,000 forecased benefit is essentially a pipe dream, and including it in the final projection will only serve to create strife when statistics is ultimately stacked against success. To review the situation in a more technical matter (non-techies beware), consider the case of probability distribution. In every situation, stastics affords us the possibility that all probability density functions will extend to infinity in both directions, if only we collect enough samples. In practical terms, this means the following: -A car seat is 3 feet tall on average, with probabilites extending to infinity. -A car is 6 feet taall on average, with probabilities extending to infinity. Statistics tells us that there is a non-zero probability that the seat can be bigger than the car, if we build enought seats and enough cars. In reality, this is impossible for a large number of reasons. The tools that fabricate the seat and the car are of fixed size, the sizes of the raw materials are bounded, etc. Obviously then, at some point, the reality of the situation has deviated from the mathematics. The difficult question is, at what point does this occur? Historically, the answer to this question (from a mathematical perspective) was generally accepted to be 3 standard deviations. Capability indices (CPk) are based around this assumption, with 3 standard deviations equal to CPk = 1. More recently, this projection has grown to 4 staqndard deviation s (CPk - 1.33), and then 6 standard deviations (the new-wave 6-Sigma methodology). Included in this capability growth is the implicit assumption that the statistical mathematics of 6 standard deviation analysis still mirrors the reality of volume production. However, there are many statisticians who believe that 6 sigma is beyond the point where mathematics and reality diverge. Also consider the wisdom of Mark Twian: Figures often beguile me, particularly when I have the arranging of them myself; in which case the remark attributed to Disraeli would often apply with justice and force:There are three kinds of lies: lies, damned lies, and statistics. Where the point of divergence lies, however, is not nearly as important as the realization that there, in fact, is a point of divergence. While the mathematical and scientific community have devoted many years to the study of this divergence, the same cannot be said of the social structure. The One Percent Doctrine is still good policy in many places, including the current government. Doomsday scenarios, what-if's, and all manner of sub-finite analysis of low probability events leads a normally rational, analytical person to irrational fears - especially when personal and familial safety is incorporated. The effect is to tend toward the elimination of personal fredoms in favor of a highly controlled, totalitarian-style state. This regression is the greatest threat to our long-term well being. The terrorists stated goal is to destroy our way of life. The only way they can do that is if we help them.