Saturday, July 17, 2004

Spammed by Maxwell’s Demon

What we think of as the norm is, after all, just what we’re used to. We shouldn’t be surprised, then, if some oddball thing we happen on turns out on closer examination to be exemplary. Since our existence is only possible because we are located in the bull’s eye of a nested set of improbabilities, inductions on everyday experience are risky. The Firesign Theater may be overstating matters to claim that everything you know is wrong, but everything is certainly likely to be anomalous. Many of the regularities we mistake for laws of nature are just ground rules and only apply in this one ballpark and then only for the duration of a game that may be called at any time for bad weather. I’ll mention some highfalutin instances in a moment, but the phenomenon is already captured by the puzzled look of a soon-to-be run over pedestrian who doesn’t understand how a taxi could be bearing down on him since the light is red. And then there’s the case of the extremely virginal girl I knew in high school who, when informed that her best friend was pregnant, cheerfully informed everybody that there must be some sort of error since her friend wasn’t married. Unlike Janet, Leibniz wasn’t, so far as I know, a Christian Scientist; but his belief that all curves have simple formulas has some of her endearing naïvite. It could not last, not for Janet, probably, and certainly not for the mathematicians, who having laboriously defined a few pathological functions in the 19th Century had to admit by the 20th that almost all functions are pathological and that the exceptions are the rule.

What I’m talking about only seems to be a rare pattern. Some more examples: in ordinary life, we assume that knowing more about what’s going on in some competitive situation will improve our chance of prevailing. That expectation may or not be valid in these parts and for the nonce, but it certainly can’t be generalized. Years ago, the game theory folks devised games in which having more accurate information puts a player at a disadvantage. Subsequently, they realized that these monsters were not curiosities at all. Having more information turns out to be disadvantageous in most possible games. Again, Kurt Godel proved the existence of propositions whose truth could not be decided by deductions from the axioms of the system in question. Naturally, it turns out that essentially all possible true propositions are in this set.

Many apparently general procedures only work under very particular circumstances. One example is the simplex method of solving linear programming problems, the single most commonly used operation’s research technique. The simplex method is the means by which an enormous number of governmental, business, and military decisions are made because it can often quickly find the best available solution to real-world problems. But the method, which works splendidly in practice, shouldn’t be that efficient. Tested on randomly generated inputs, it takes a long time to find the answer. The utility of the method depends on something about our world and how we deal with it. Which really shouldn’t be too surprising. We know that the performance of other optimizing techniques also depends on the circumstances. It has been clear for a long time that natural selection only works in a small subset of possible environments; and, more generally, the efficiency of any learning algorithm depends on what there is to be learned. We can’t know in advance what method to use. It’s not just that it’s better to be lucky than good. You’ve got to be lucky to be good.

Meanwhile. All of these thoughts were occasioned as I set about the task of deleting a day’s worth of emails for advanced degrees, penile enlargements, photos of hot middle aged women, real estate deals, and work-at-home schemes from my computer. I suddenly realized that the spammers aspired to be Maxwell’s demons, agents that make a profit without doing any work. As you’ll recall, a Maxwell demon is a character in a thought experiment. The demon, acting as a gatekeeper between two gas-filled chambers, lets fast molecules go from left to right while only letting slow molecules go from right to left until one chamber is much hotter than the other. In this fashion, in apparent disregard of the 2nd Law of Thermodynamics, a temperature difference is created that can be used to drive a piston. The demon gets something for nothing, exactly what the Viagra salesmen are trying to get.

Now a Maxwell’s Demon is not really possible if only because it costs something to find out about the velocity and location of the particles and compute when to open or close the gate. But one of the reasons it costs to do the computation is rather subtle. Holding the intermediate results of your calculations in your head isn’t so expensive, but you not only have to remember things. You have to forget them, and that requires extra energy. When I encountered this argument years ago in an article in Scientific American, I never expected that it would turn out to have a real world application, but it does. It explains how the computer spammers succeed. They can impersonate Maxwell’s Demons by finessing the energetic costs of forgetfulness. They aren’t stuck with the vast amounts of garbage information they generate in the course of their operations, but only because they can get us to take out the trash for them.

No comments: