Selection Power and Selection Labor - Human Information Retrieval - page 29

carry this distinction forward as an analytical distinction, while recog-
nizing the difficulty of substantive separation. For example, a searcher
requesting a list of documents in chronological order effectively instan-
tiates at the point of searching what once would have been a form of
description labor, or process. The contrast is traceable back to the fixity
of writing technologies compared with the possible fluidity of compu-
tation (Warner 2001, 33-46). The concern here is with the theoretical
possibilities for and constraints on selection labor, conceived as incorpo-
rating both description and search labor.
Theoretical minima for selection labor can be derived from serial pos-
sibilities. If items are examined serially and without regress, selection
labor increases with the number of objects examined in the collection. An
unchanged principle for discrimination is assumed, rather than a conver-
sational or dialogic alteration of the principle for discrimination, and this
would be closely analogous to the batch processing practiced historically.
An absence of meaningful organization of the discriminated objects and
a conflation and simultaneous occurrence of description and search labor
are also implied. If the choice between objects examined is reduced to a
binary contrast between acceptance and nonacceptance, the unit of labor
is somewhat analogous with the classic understanding of the bit.
Imposing organization upon a collection of objects separates descrip-
tion from search labor, with work invested in organization or descrip-
tion reducing labor and enhancing power in searching. Selection labor
is then distributed between description and searching. If description and
organization can partition objects into appropriate sets, Shannon's for-
mula for the information of a source would indicate that the number of
objects discriminated by search labor would rise more than the quantity
of search labor:
If there are N possibilities [for the choice of messages from a source], all equally
likely, the amount of information is given by log 2 N . . . . If it were possible to choose
questions which always had the effect of subdividing into two equal groups, it
would be possible to isolate, in twenty questions, one object from approximately
1,000,000 possibilities. (1968/1993, 214-215)
On this basis, the number of choices—broadly, units of labor—required
to discriminate between c.1000 and c.1,000,000 such possibilities (which
could correspond to objects, documents, or records for documents) would