Mechanization of Cognition - Biomimetics: Biologically Inspired Technologies - page 101

considering their function (rather than their biological implementation), each such thalamocortical
module will be referred to as a lexicon .
Each feature attractor module is hypothesized to be controlled by a single graded (i.e., analog-
valued) excitatory control input; exactly in analogy with an individual muscle (each muscle has a
single-graded excitatory control input that, by its value across time, specifies the muscle's contrac-
tion force history). The theory hypothesizes that properly phased and timed sequences of such
thought control inputs to each member of an ensemble of cortical modules cause them to carry out a
thought process. These thought processes are ''data-independent,'' much like computer operations
such as numerical addition and Boolean XOR. It is hypothesized that vast numbers of such thought
processes (and movement processes) are learned by rehearsal training and stored in a hierarchical
organization within knowledge bases linking yet other lexicons of cerebral cortex.
Confabulation is implemented in parallel by the neurons of a feature attractor module and is
often completed in a few tens of milliseconds. This is a ''winners take all'' style of dynamical
competitive interaction between symbols that does not require a ''referee'' or ''controller'' to be in
charge. The states of the involved neurons evolve dynamically and autonomously during confabu-
lation via the massively parallel mutual interactions of the involved neurons. The final state of each
involved neuron is either excited or active (a small minority of neurons), or almost completely
inactive (the vast majority). The term active (implying a momentary, maximally communicating
state) is deliberately undefined as it involves neuronal signaling details which are not yet known; as
does the term excited (implying a highly, but not maximally, communicating state).
If the outcome of a confabulation is a single symbol, the neurons representing that symbol will
automatically be made active and all other neurons of the module are inactive (not communicating).
However, if multiple symbols result from a confabulation (the outcome is dependent upon multiple
factors, including the time profile of the module control signal — see below), these will be at
different levels of excitation (but not active) and all other symbols will be inactive. Those few
neurons which end up in the excited or active state represent the symbol(s) which ''won'' the
confabulation competition. These symbols are termed the conclusions of that confabulation oper-
ation. Confabulations frequently end with no excited or active neurons — a conclusion termed the
null symbol — which signifies that no viable conclusion was reached. This ability to decide that
''I don't know'' is one of the great strengths of cognition.
The theory hypothesizes that the only knowledge stored and used for cognition within thala-
mocortex takes the form of (indirect, parallel) unidirectional axonal connections between the
population of neurons within one feature attractor module used to represent one symbol and
neurons used to represent a symbol in another feature attractor. Each such link between a pair of
symbols is termed an item of knowledge . The average human is hypothesized to possess billions of
such items of knowledge.
The theory hypothesizes that items of knowledge are immediately established on a temporary
basis when a novel, meaningful, co-occurrence of symbol pair activity occurs during a period of
wakefulness (assuming that those symbols are equipped with the necessary axonal paths with which
a link can be established). If this short-term memory link is selected for deliberate rehearsal during
the next period of sleep, it gets promoted to the status of a medium-term memory . If this medium-
term memory link is revisited on near-term subsequent sleep periods it then gets promoted to a
long-term memory , which will typically last years; as long as the involved tissue remains patent and
not redeployed.
It is hypothesized that each time a thalamocortical module carries out a confabulation which
concludes with the expression of a single active symbol (as opposed to no symbols or multiple
excited symbols), an action command associated with that symbol is immediately issued by a
specialized cortical component of that module (this is the theory's conclusion-action principle ).
Action commands cause muscle and thalamocortical thought module control signals to be sent.
In other words, every time a thought process successfully reaches a single conclusion, a new
movement process and/or thought process is launched (some of which may undergo additional