The Evolution
27
scalpel, one would not then go on attributing higher probabilities to
kitchen knives and pocketknives—they would no longer have higher
than normal probabilities.
Accordingly, if a memory optimized for anticipation and disambig-
uation perceived "hospital" and "blade" in immediate succession, ideally
it would use the whole-to-whole links emanating from "hospital" to
disambiguate "blade" and would then for some time afterward attribute
elevated probabilities to all those concepts having whole-to-whole links
with "hospital”; but such a memory ideally would not go on to attribute
higher than normal probabilities to the various forms of "blade"—
instead the part-whole links emanating from "blade" should be made to
cease to exert influence as soon as "blade" itself had ceased to be
immediately present in memory.
As it turns out, what is true for the part-whole links of "blade"
and the whole-to-whole links of "hospital" is also true for part-whole
links and whole-to-whole links in general. Part-whole links, such as
those that colors, shapes, and textures have with the concepts they are
a part of, fairly consistently supply information that is invalidated
by accurate disambiguation. Whole-to-whole links, in contrast, in general
supply information that is not invalidated by accurate disambiguation or
anticipation. Accordingly, in an optimized memory a set of whole-to-whole
links, once activated, should remain activated for some time afterward,
with a resultant heightened expectation being established in memory for
all those concepts singled out by those whole-to-whole links. Part-whole
links, in contrast, should not give rise to any form of persisting acti-