Mechanization of Cognition - Biomimetics: Biologically Inspired Technologies - page 58

3.A.6 Action Commands .................................................................................................................. 118
3.A.7 Discussion ............................................................................................................................... 123
Acknowledgements......................................................................................................................................... 125

3.1 INTRODUCTION

This chapter describes the state of the art in creating animal cognition in machines. It begins
with a discussion of the two fundamental processes of cognitive knowledge acquisition — training
and education . The subsequent sections then present some ideas for building key components
of cognition (language, sound, and vision). The main point of this chapter is to illustrate how
we can now proceed towards the mechanization of key elements of cognition. This chapter
assumes that the reader is familiar with the concepts, terminology, and mathematics of elem-
entary confabulation (as described in Hecht-Nielsen, 2005) and its hypothesized biological
implementation in the human cerebral cortex and thalamus (as described in the Appendix of this
chapter).

3.1.1 Mechanized Cognition: The Most Important Piece of AI

As discussed in Section 3.A.1 of the Appendix, human (and higher mammal) intelligence involves
a number of strongly interacting, but functionally distinct brain structures. Of these, significant
progress has now been made on three: cerebral cortex and thalamus (the engine of cognition — and
the focus of this chapter), basal ganglia (the behavioral manager of the brain — which manages
action evaluation, action selection, and skill learning), and cerebellum (the autopilot of the brain —
which implements detailed control of routine movement and thought processes with little or no
need for ongoing cognitive involvement once a process has been launched and until it needs to be
terminated). There are a number of other, smaller-scale, brain functions that are also critical for
intelligence (e.g., ongoing drive and goal state determination by the limbic system), but these will
not be discussed here.
Of all of the components of intelligence, cognition is, by far, the most important. It is also the
one that has, until now, completely resisted explanation. This chapter provides the first sketch of
how cognition can be mechanized. The approach is based upon the author's theory of vertebrate
cognition, which is described in the chapter's Appendix. This chapter is not a historical description
of ''how cognition was mechanized''; but is instead an ''initial plan for mechanizing cognition.''
Initial progress in implementing this plan in areas such as language and hearing (the subjects of
Sections 3.3 and 3.4) has been encouraging.

3.1.2 Lexicon Capabilities

This chapter considers some more sophisticated variants of confabulation that go beyond elemen-
tary confabulation. Each lexicon used in our (technological) cognitive architectures (collections of
lexicons and knowledge bases) will be assumed to possess the machinery for carrying out each of
these confabulation variants (or information processing effects — the term that will be used for
them here), as described below. Thus, from now on, the term lexicon implies a capability for
implementing a finite set of symbols, maintaining a list of the excitation states of those symbols,
and for executing the effects defined below. For the moment, lexicon dynamics will be ignored.
(However, in later sections, concepts such as consensus building and symbol interpolation , which
intrinsically require lexicon dynamical behavior, will be briefly mentioned.)
One very important detail that was not discussed in Hecht-Nielsen (2005), and only briefly
discussed in the Appendix (because it is not relevant to the biological implementation of elementary