The Evolution
16
it might yield "toast," or perhaps "bacon" or "orange juice," as being
most likely to occur next.
But the whole-to-whole links emanating from "eggs" should be able
to contain more information than merely what, based on past experience,
is most likely to follow "eggs"; by varying their strengths, they should
also be able to reflect relative likelihoods. Thus if "eggs" has a proba-
bility of .0001 (i.e. "eggs" occurs once in every 10,000 percepts), then
during the course of a lifetime it should occur several thousands of
times, with each of these occurrences constituting a sampling of what is
likely given the presence of "eggs"; that is to say, "eggs" might develop
strong links with "toast," bacon," and "orange juice," while developing
no links at all with the vast majority of concepts, and in this way the
link strengths emanating from "eggs" could come to constitute a statis-
tical profile on what is likely to follow the perception of "eggs."
(Technically speaking, such profiles are samples taken from the environ-
ment under biased conditions, where the "bias" takes the form of the con-
ditional occurrence of the percept.)
Note that in what follows it will be assumed that the sum of all of the
whole-to-whole link strengths emanating from each concept equals one. The
purpose of this rather artificial assumption is to simplify matters by
seeing to it that a particular link strength corresponds to a particular
probability regardless of which profile it is taken from.
Using the above statistical profiles as a base, it is now possible to
analyze Type II and Type III intersections.