Page 133 - Biomimetics : Biologically Inspired Technologies

Bar-Cohen : Biomimetics: Biologically Inspired Technologies DK3163_c003 Final Proof page 119 21.9.2005 11:41pm

Mechanization of Cognition 119

The Layer V neurons which become highly excited when a symbol wins a confabulation cause a
very specific set of actions to be executed (or at least to be considered for execution; depending on
the function of the action nucleus that receives the Layer V efferents). This is the origin of all
behavior — each successful (one winning symbol) confabulation causes the launch of a set of
associated action commands. These actions can be part of a movement process, part of a thought
process, or both.
During development, the genetically determined program for creating the brain is, barring
problems, executed. This program causes the development of axons from neurons in Layer V of
each cortical module portion which proceed to genetically directed subcortical action nuclei (of
which there are tens). In other words, genetics can ensure that a module has the Layer V neurons it
needs to launch those actions, which that particular module should be empowered to execute. Thus,
each of us has a range of behavioral potentialities which are in this sense predetermined. Undoubt-
edly, this is how various talents and personality traits are transferred from parents to children. This
is part of the ‘‘nature’’ portion of the human equation.
Given the behavioral potentialities established by the genetically directed wiring of the axons of
the Layer V neurons of a module to action nuclei, the big question is how exactly the correct ones of
these Layer V neurons end up getting ‘‘wired’’ from the population of neurons representing each
symbol. Given the exact specificity of effect each Layer V neuron produces, there is no room for
error in this wiring from each symbol to the action commands it should launch. Since the local
geometrical arrangement of the symbol-representing neurons and action command neurons within
their respective layers is random, and their local axonal wiring is largely random, this wiring from
symbol representing neurons to Layer V action-command-generating neurons cannot be genetically
determined. These associations must be learned and they must be perfect. Figure 3.A.8
illustrates the theory’s hypothesized mechanism for implementing these precise symbol to action
associations. This figure will be referred to extensively below.
The learning of symbol to action command associations is almost certainly a totally different
learning process from that used in development of module symbol sets or in the establishment of
knowledge links. This symbol to action association learning process is hypothesized to take place
extensively during childhood; but also very frequently during adulthood. Cognitive lexicon devel-
opment, cognitive knowledge acquisition, and symbol to action command association learning
together make up the most ‘‘glamorous’’ parts of the ‘‘nurture’’ portion of the human equation (there
are a number of other, quite different, learning processes that go on in other parts of the brain; e.g.,
learning to sense when we should use the toilet).
Notice that in Figure 3.A.2, every cortical layer of a module is mentioned except Layer I (the
most superficial). Layers II, III, and IV are primarily involved in symbol representation, precedence
principle interactions among feature detector neurons, and the receipt of afferents from thalamus.
Layer V is where the action command output neurons reside. And Layer VI is where the cortical
efferents to thalamus arise. The theory hypothesizes that Layer I is where the wiring between the
symbol representation neuron sets and the Layer V action command output neurons takes place
(and quite possibly some of the wiring for the feature attractor module function as well). It is well
known (Paxinos and Mai, 2004) that the neurons of Layer V (typically these are of the pyramidal
category) have apical dendrites that ascend to Layer I and then branch profusely. Further, neurons
of Layers II, III, and IV typically send large numbers of axon collaterals to Layer I (and also
frequently have apical dendrites too — but these will not be discussed here). Further, the basal
ganglia (BG — a complicated set of brain nuclei known to be involved in multiple types of action
learning (Paxinos and Mai, 2004)), specifically, the BG substructure known as the striatum sends
signals in great profusion to Layer I of cortex via the thalamus (see Figure 3.A.8). This radiation is
principally concentrated in frontal cortex (where most behaviors seem to originate), but other
cortical areas also receive some of these inputs.
Given the random nature of cortical wiring, the only way to establish correct symbol to action
associations is via experimentation. This experimentation is carried out (starting with the simplest