OnClustering

We naturally see things in groups: flocks and herds, villages and towns, constellations and asterisms, cells and organs. Clusters are fundamental to language. We use one word to apply to all of a type of thing: fish or fowl, red or green, female or male, young or old. Groups are somewhat arbitrary, and their boundaries can overlap or shift, depending on the features that seem most significant at a given time or for a given purpose. There are mathematical ways to cluster abstract data points, both top-down (slicing the whole into components and sub-components) and bottom-up (aggregating entities into families, families into clans, clans into nations, etc.). Groups may be sharp-edged or fuzzy, disjoint or overlapping, dense or sparse.

People cluster, based on common backgrounds and interests. Someone who fortuitously lands near the center of a dominant group can more easily speak to (or for) the "masses", write a best-seller, perhaps get rich or win an election. But while cluster centroids represent large raw numbers of individuals, they do not stand for as big a volume of possibility-space as do people who live in the deserts between clusters. Those outliers are critical, though they're often overlooked. They probe the wilderness and pioneer new trails between zones of stability. When radical change threatens or disrupts established groups, the adventurers offer a chance for survival.

Thursday, November 04, 1999 at 05:54:19 (EST) = 1999-11-04

TopicScience - TopicLanguage


(correlates: EmersonOnNatureAsAntidote, SandBox, ClusteringAlgorithms, ...)