Information Technology Reference
In-Depth Information
from a computer that calculates F i to one that calculates F j . Mechanisms
that achieve such modifications are well-known as, e.g., long range inhibi-
tions and facilitations. However, I doubt that it will ever be possible to
establish a detailed account of the relations between individual synaptic
changes and the computational properties of the whole aggregate, the main
reason being that this is a problem that does not have a unique solution,
on the contrary, it can be shown that with just a few cells making up this
computer, the number of different solutions is, for all practical purposes,
infinite. On the other hand, I do not believe that such detailed knowledge
is of importance, as long as the principles are understood that make such
modifications possible.
Let me now briefly summarize some of the essential results of this dis-
cussion. Most of the neural machinery is functionally organized to estab-
lish from sensory information—whether about states of the outside world
or about internal states—relations between observed entities with respect
to the observing organism. This relational information modifies the modus
operandi of a computer system that computes new actions recursively on
the basis of the outcome of previous actions and, hence, on the basis of the
history of the stream of external and internal information. Figure 6 is a
graphical representation of this summary in form of a block diagram. I shall
call this whole system a “Cognitive Element”, for it represents a minimal
case of a cognitive process, or a “Cognitive Tile”, for it may be used in con-
junction with other such tiles to form whole mosaics—or “tessellations”—
which, as a whole, permit the high flexibility in representing relational
structures not only of what has been perceived but also of the symbols—
the “linguistic operators”—that ultimately are to convey in natural lan-
guage all that which can be inferred from what has been perceived.
The various components of this cognitive tile are quickly explained. X
stands for (external) sensory input, and Y for the output of the system as
seen by an outside observer. Hence, this elementary component is a
“through-put” system, as suggested by the small inset, lower right. However,
because of its internal organization, this element is quite a different animal
from a simple stimulus-response mechanism with fixed transfer function.
First, sensory information, X, is operated on to yield relations R S (X)
between observed activities with respect to “self” (note subscript s), and is
then used as input proper for the recursive function computer which may
be operative at this moment with any one of the functions F belonging to
range F. Its output is fed back over two channels, one being the recursive
loop with delay D to allow F to assess its earlier actions, the other carrying
all the relational information of the system's own actions R S (Y) as they
refer to “self”, and operates on F(F) in order to set the recursive function
computer straight as to this tile's internal goals and desires.
This element incorporates all those faculties which I considered earlier
to be necessary components of cognitive processes: to perceive, to remem-
ber and to infer. However, in this element none of these faculties can be
Search WWH ::




Custom Search