From: Carlo Wood (carlo@alinoe.com)
Date: Fri Apr 05 2002 - 21:04:51 MST
While reading I might write a remark every now and then,
so that I won't forget that later on :).
Reading the URL in the Subject, I must say that the
given list of layers:
* source code and data structures.
* sensory modalities.
* the concept level.
* thoughts.
...
Where 'thoughts' are described with
"By applying a series of concepts to a single target,
it becomes possible to build up ..." [...] "The archetypal
example of a thought is a human "sentence" "
I must say that I definitely miss a layer there.
"thoughts" that we experience as 'words' are only
the result of tagging/labelling real thoughts,
a (usually poor) *translation* of the real thoughts.
This translation is by no means really necessary, it
is typical for a human because we use words for communication
a lot, but I'd consider sentences as a side-track.
The real thoughts then do NOT consist of anything like
"a series" of concepts applied to ... That is 'one
dimensional', just like language. Real thoughts are
multi-dimensional: one finds the *balance* of many
concepts at the same time, given their relationships.
Certain concepts are considered fixed, others are free
to be manipulated (usually those that will be our own
final actions/decisions) and many are weighted with a
statistical chance. The chance that certain assumptions
are true are very important in real thoughts, as are
the evaluations of results which are also used as
weights (almost every 'parallel "reasoning"', by lack of
a better word, contains 'likeliness' and 'evaluation' (how
important do I feel it is, compared to other things), in
fact so often that I think it is a specialized part of the
brain that deals with this. Essential here is that we can
in no way speak of a series of applied concepts but always
of a process that mostly feels like finding the maximum or
minimum in a multi-dimensional equation of variables where
all variables are coupled and "slowly" move simultaneously
until we find a balance were changing any variable a
little bit seperately causes the outcome to get worse
(in other words, a local minimum/maximum).
Further more, there is another layer missing:
There is a low level analyses layer that some how
finds a variable (concept) -transformation of the variables
involved such that the result is as symmetric and low dimensional
as possible. Many n-dimensional problems can be simplified to
one with less variables by looking at it differently. I believe
that this has to do with elimination of (strong) correlations that
you are aware of somehow.
Example:
Suppose you are at a place X and want to go to a place Y,
having to cross a road in between:
X
----------------------------------------------------------
----------------------------------------------------------
Y
If the road is very busy with traffic, you walk like this:
X
.
.
-----------------+----------------------------------------
[XX] [XX] | [XX] [XX] [XX] [XX]
-----------------+----------------------------------------
.
.
.
Y
In other words: you cross the road perpendicular.
When there is absolutely no traffic whatsoever in sight (an
empty country road in a flat and bare sand environment),
you will ignore the road completely and walk in a straight
line from X to Y, crossing the road diagonally.
But, somewhere inbetween, when there is a little traffic
but not too much - then you walk just like light travels
through a plate of glass, "breaking" your path twice at
the edge of the road:
X
.
.
-----------------+----------------------------------------
\ [XX]
------------------ +--------------------------------------
.
.
.
Y
How do you determine the angle in which you cross the road?
By reasoning at the level of 'words'? By applying a series
of operators? No, you *solve* an equation with as variables
the angles of the three lines involved and the *danger* level
of each path.
Minimize: Total-Path-Length * evaluation(annoyance-that-it-takes-so-long) +
Time-On-The-Road * evaluation(danger-level-on-the-road).
The mathematical equation happens to be precisely equivalent
to the one for light breaking in a plate of glass, where the
'annoyance-that-it-takes-so-long' and 'danger-level-on-the-road'
represent the 'breaking-indexes'.
Solving the path happens in two steps (after realizing the
concepts involved):
1) Simplifying the problem to one with three STRAIGHT lines,
where the first and the last are parallel, and only two
fixed variables that have to be relatively compared and
determined (the annoyance that it takes longer against
the extra danger when you cross the road more diagonal).
2) Actual parallel solving that equation. In your head you
vary the variables a bit until you are satisfied, and then
you use one of those variables as action: the angle of the
first part of the path.
-- Carlo Wood <carlo@alinoe.com>
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:38 MDT