Page 74 - Biomimetics : Biologically Inspired Technologies

Bar-Cohen : Biomimetics: Biologically Inspired Technologies DK3163_c003 Final Proof page 60 21.9.2005 11:40pm

60 Biomimetics: Biologically Inspired Technologies

in cortex. It also induces what might be thought of as a probability distribution on the expectation
symbols. Cognition must very often conduct a multi-stage process of gradually promoting hypo-
theses (expectation symbols) which gain significant support from incoming knowledge links and
demoting those which fail to gain as much support. Thus, the effect CK is very important in
cognition. In the brain, ‘‘CK processing’’ is continuous in time and happens very rapidly.
The set of symbols of a lexicon having nonzero excitation levels I(l) following a W, WK, or CK
effect is termed an expectation. Expectations are considered to have a short life (i.e., after a ‘‘short’’
time has elapsed after a confabulation the state of the lexicon, its collection of I(l) values, becomes
indeterminate). Note that this is a generalization of the term expectation used in Hecht-Nielsen
(2005). The term active is still reserved for the case of a single confabulation conclusion; and highly
excited will still mean that the expectation has multiple elements.
Another effect is freezing, denoted by the letter F. Freezing a lexicon causes each symbol with
positive final excitation (i.e., after a W, WK or CK) to have its final excitation I(l) value preserved
for a longer time. During this (still rather brief) period of time that follows F, only those symbols
which are members of this expectation can receive further knowledge link inputs. In other words,
the input excitations of symbols not in the expectation stay at zero during the frozen period. So, for
example, if further new link inputs arrive shortly after an F has been invoked, and then a W is
commanded, an expectation symbol (if there are any) which obtains the highest positive I(l) value
will be made active.
As we will see later, building and using expectations is one of the most important elements of
cognitive information processing. By using sequences of confabulations to ‘‘whittle down’’ ex-
pectations, constraint knowledge of various kinds can be applied to rapidly home in on a final
conclusion. In effect, each expectation represents the set of all ‘‘reasonable conclusions’’ that are
worth considering further. When the expectation is finally reduced to one conclusion, via succes-
sive freezes and confabulations, the final, decisive conclusion is found (or if the final expectation is
empty, then the answer is ‘‘I don’t know’’). Almost every aspect of cognition is implemented by
such sequences of such ‘‘deductive’’ confabulation steps (although this is not deduction in any
formal sense, because it is based on the undecidable (but usually reasonable) assumption of
exhaustive knowledge).
Finally, consider a lexicon which, when last used for confabulation (within the past few hours),
yielded a decisive conclusion and which, subsequently, has not been erased. If this lexicon now
receives a W, WK,or CK but no knowledge link inputs, that symbol which was its last conclusion
will, in isolation, become active. This is a sort of temporary symbol storage mechanism that the
theory terms working memory. If the lexicon has been erased, an expectation containing some of the
symbols which resulted from recent past uses will be expressed (with the chances of appearing
depending upon how many times that symbol was frozen in succession when it originally
appeared).

3.1.3 Discussion

Technological cognition will be inherently limited without the other functionalities that brains
provide (see Appendix Section A.1). Further limitations arise because of the lack of on-line
memory formation mechanisms (short-term, medium-term, and long-term memory processes)
and the lack of a capability for goal-driven delayed reinforcement learning of thought and
movement procedures (see Section 3.6). Yet, despite these limitations, there are probably many
high-value early applications of pure cognition that will be possible. Pure cognition is the focus of
this chapter.
Language is almost surely the faculty which accounts for the dramatic increase in human mental
capability in comparison with all other animals. It is in the language faculty, and in the language
faculty’s interfaces with the other cognitive faculties, that almost all distinctly human knowledge
is centered. Thus, language is where the mechanization of cognition must start (see Section 3.3).