The Evolution
18
having two sources of information on what is likely, an optimized
memory is obligated to use both in order to form the best possible
view of what might occur next.
If one assumes that memory treats each set of probabilities as
of equal value, then to form a single profile telling what is likely
given the occurrence of both "Florida" and "eggs" requires the summing
of corresponding probabilities and dividing by two. The only impor-
tant difference between such a composite profile and each profile
taken separately is that, relative to the other concepts, "orange
juice" moves up somewhat. In the "eggs" profile "toast" is 12.5 times
as likely as "orange juice" (.5 versus .04), while in the composite
profile "toast" is only 10 times as likely (.25 versus .025). This
upward shift occurs because "orange juice" is the only concept on both
lists.
It is important to notice how small this upward shift is, in
particular that it is nowhere near enough to allow "orange juice" to
go to the head of the list. The reason for this can be seen clearly
by noting that the strongest link of all those emanating from "Florida"
and "eggs" is the one running from "eggs" to "toast." Accordingly,
"orange juice's" two sources of support in combination would have to be
greater in value than "toast's" one source of support if "orange juice"
is to be yielded as being most likely.
"Florida" ----------------> "toast" (zero probability)
----------------> "orange juice" (.01)
"eggs" -------------------> "toast" (.5)
--------------------» "orange juice" (.04)