Page:Sm all cc.pdf/110

 *Keep a record of the search; ideally, flag searched areas so that they are readily distinguishable from unsearched regions. Meteorites are occasionally found, but until recently they were considered to be far too rare to search for. Now they are being found in record numbers in Antarctica, and scientists have even found eight that they think have come from Mars and several that definitely came from the moon [Gleick, 1992b]. The new success is due to letting natural processes concentrate the meteorites: they are found on top of the Antarctic ice, in environments that are undergoing ablation rather than precipitation. Paul Ehrlich’s hypothesis was that some substances are able to kill parasites without damaging the host. He systematically investigated 605 compounds without success; the 606th, salvarsan, proved effective against syphilis. [Beveridge, 1955]
 * Search the most probable place first, then search places that are successively less likely to contain the object. Use specific criteria for estimating promising locations; do not just play hunches.
 * Search systematically. Haphazard searching can take just as long as a systematic search, yet search some areas several times and others not at all. At the end of a haphazard search one still cannot exclude the searched area. Most searches begin haphazardly, but once they become timeconsuming, one should pause and plan a systematic search. Systematic searches are not necessarily inflexible, inefficient searches.


 * Distribute your available searching resources -- whether time or manpower -- appropriately in the different regions. For example, if several areas seem equally promising but some are much easier to search, search the easy ones first. If you will have to quit searching after a limited time, usually a detailed search of the most promising area is more effective than a widespread superficial search. If a little-known phenomenon or a totally new type of observation is being explored, the initial search should probably be a broad reconnaissance rather than a detailed examination of a small subset. Detailed focus is higher risk, until the reconnaissance establishes which parts of the whole are most likely to reward close-up studies.

Consider the game of twenty questions, as employed in the old television show ‘What’s My Line?’ There are thousands of professions, yet the questioners were often able to guess the contestant’s profession. If one can design each yes/no question to cut the number of possible solutions in half, then twenty questions can sort out one choice from 1,048,576 possibilities (220). Twenty guesses or a million? Clearly a systematic search procedure such as this one can be extremely powerful. Unfortunately,
 * Use a convergent search procedure, if possible. Convergent searches employ feedback on whether they are getting closer to or farther from the object. This technique is feasible for questions such as “When did the equipment start giving strange results?” or “Where is the short circuit?” but useless for questions like “Where is the needle in the haystack?” When using a convergent search, it is better to overshoot than to undershoot; this is the tactic used in golf putting and in weighing (use large weights until overshooting, then smaller weights). The ideal search procedure eliminates half the possibilities at each step: