Similarity cluster

From Issuepedia
Jump to navigation Jump to search


A similarity cluster is a group of ideas which tend to be found together, or a group of things that have many characteristics in common.

Because of this tendency, people tend to put labels on these groups as if they represent an unambiguous category, or to assume that the individuals involved in a cluster are in some way identical to each other, or to overgeneralize from some attributes being the same to a belief that all attributes must be the same (possibly even making negative value-judgements on individuals who do not share all the group attributes).


In an "attributional" similarity cluster, the "similar ideas" are attributes whose values tend to be highly correlated in certain ways – leading to a "clustering" of points when the entities possessing these attributes are plotted (using as many dimensions as necessary) along the axes of those attributes – but which are not completely dependent upon each other, resulting in a small but significant population of outliers.

That is: the dimensions are often erroneously conflated with each other, leading to concept-labels which assume that any given individual will be clearly classifiable as belonging to one or another of the clusters.


  • conservatism: conservatism consists of a combination of related ideas, but any given type of conservatism may exclude one or more of them
  • gender (an attributional similarity cluster): people are not necessarily all-male or all-female
  • religion: religion isn't a specific thing, but a set of related ideas


to file

  • 2008-02-06 Similarity Clusters by Eliezer Yudkowsky: "A dictionary is best thought of, not as a book of Aristotelian class definitions, but a book of hints for matching verbal labels to similarity clusters, or matching labels to properties that are useful in distinguishing similarity clusters."