Page 138 - Biomimetics : Biologically Inspired Technologies

Bar-Cohen : Biomimetics: Biologically Inspired Technologies DK3163_c003 Final Proof page 124 21.9.2005 11:41pm

124 Biomimetics: Biologically Inspired Technologies

puzzling findings of neuroscience: the vast majority of synapses that have ever been individually
evaluated (e.g., by manipulating them, and monitoring their effects on the target cell, using multiple
patch clamps [Cowan et al., 2001]) have turned out to be very unreliable and only marginally
functional. This is exactly what you would expect to find if 99% of synapses are in a state of
minimal existence, awaiting the possible moment that they will be needed.
9
Humans live for roughly 3 10 sec. So, for example, if we acquire an average of one item of
knowledge during every second of life (86,400 knowledge items per day), and if an average of 300
transponder neuron synapses are used to implement each knowledge item, far less than 1% of all
synapses will ever be used (of course, not all cortical synapses are available for knowledge storage,
but most probably are, so this conclusion is still probably correct). So, the theory proposes that the
potential amount of cognitive knowledge that can be stored is huge.
In my laboratory’s computer implementations of confabulation, a startling fact (which is
consistent with the above numbers) has emerged: a staggeringly large number of knowledge
items is needed to do even simple cognitive functions. The theory postulates that the average
human must possess billions of items of knowledge. This has many startling and profound
implications, and assuming that the theory gains acceptance, many philosophical and educational
views of humans (and other animals) will likely be completely altered. For example, the theory
implies that children (and adults too!) probably accumulate tens of thousands, or more, new
individual items of knowledge every day. Thus, the process of reconsidering each day’s short-
and medium-term memories and converting selected ones into a more permanent form is a huge
job. It is no wonder that we must sleep a third of the time.
To appreciate the vast storage capacity of your cerebral cortex, imagine for a moment that you
are being asked a long series of detailed questions about the kitchen in your home. Describe all of
the spoons and where they are kept; then the forks, the drinking glasses, and so on. Describe how
you select and employ each item. Where and when you obtained it, and some memorable occasions
when it was used. Obviously, such a process could go on for tens of hours and still turn up lots of
new kitchen information. Now consider that you could probably answer such detailed questions for
thousands of mental arenas. Humans are phenomenally smart.
Another cortical property, which the theory’s hypothesized design of cortex imparts, is an
insensitivity to occasional random neuronal death. If a few of the transponder neurons which
represent a particular symbol randomly die, the remaining knowledge links from this symbol
continue to function. Newly created replenishment neurons (which the theory proposes arise
throughout life) which turn out to have the appropriate connectivity (once they have spread out
and connected up and reached maturation), can be incorporated into such a weakened link to
replace lost neurons; assuming the link is used from time to time.
If a link is not used for a long time, then as the transponder neurons of its source symbol slowly
get redeployed (see below) or die, the axons to the target symbol neurons of the link will not be
replenished and the link will become gradually weaker (other links having the same source symbol,
which are used, will not suffer this fate because they will be replenished). Eventually, the unused
link will become so weak that it cannot function by itself. Sometimes, when a link has become
weak, but is not completely gone, it can be used if accompanied by additional assumed fact inputs to
the same target symbol — a faded-memory recall trick known popularly as mnemonics. This is the
theory’s explanation for why we forget long-disused knowledge.
Another aspect of the hardware failure tolerance of cortex is the primary representation of each
symbol within its own lexicon. With tens or hundreds of neurons representing each symbol, the
lexicons symbols too have some redundancy and failure tolerance.
When new inputs to a cortical lexicon arise which do not fit any of the existing symbols well,
and continue to appear repeatedly, new symbols can be formed, even in adulthood. Depending on
how close to capacity the involved lexicon is, these new symbols may or may not displace existing
symbols. This lexicon rebuilding process is often used to add new symbols to lexicons when we
learn a subject in more depth (e.g., when we take Calculus III after having already taken Calculus