Forget Mind.REXX; go with Mind.Forth 
Author Message
 Forget Mind.REXX; go with Mind.Forth

Hello, anyone following the port of 1994 Mind.rexx to Mind.Forth:


in e-mail for the public-domain source code of 26nov1994 Mind.rexx,
so I intend to post the 107K main Mind.rexx source code at the URL
http://www.*-*-*.com/ ~mentifex/mindrexx.html as a native ASCII file.

The question was asked with a view toward re-implementing REXX
(itself!) in Forth and then running Mind.rexx in the Forthrexx.

Mind.rexx was not robust enough to deserve such re-implementation;
http://www.*-*-*.com/
is already far enough along that IMHO it is better to concentrate
on the native Forth version rather than resurrecting old Mind.rexx.

Anyone who wants to port either Mind.rexx or the new Mind.forth into
any computer language should start by doing one easy thing and one
difficult thing.  The easy thing is to set up an auditory memory
channel array similar to the Mind.forth SHORTTERMMEM Screen #13.
When the human user types an English message into the keyboard,
each ASCII character, after capitalization, goes one by one at
time points t.1 to t.n into the auditory memory array ear{ }.

The more difficult thing is to write the COMPARATOR Scr #11 code,
because Mind.rexx and Mind.forth use a unique method to recognize
words coming in from the user.  The recognition scheme is intended
to mimic what must be happening in the human brain.  Each character
(as a quasi-phoneme) coming in from the keyboard slightly activates
all previous instances of itself in the array ear{ }.  Mind.Forth
STRING-EFFECT Screen #10 makes sure that only those words with an
unbroken string of serial activations get recognized.  It is hard
to write the COMPARATOR Scr #11 code, and even now there is a tiny
bug in Mind.forth with the problem that double letters (as in "see")
cause Mind.forth not to recognize the reoccurrence of a known word.

Mind.forth is at the stage right now of setting up the array psi{ }
to hold eight or six{*filter*} or up to 64 basic concepts as a BOOTSTRAP
sequence so that it can be seen quite readily whether or not the
AI program is really thinking when it converses with a human user.

Each concept is just a numbered node in the array psi{ }, with a
numeric flag pointing to lexical entries in the English array uk{ }
and with further flags in uk{ } leading to word-recall in ear{ }.
No new code has yet been written beyond Internet Web Release #16.

In November of 1994 I was shocked to discover that Mind.rexx would
not work unless the *temporal* relations among concepts played a
role in the re-formation or re-statement of a piece of knowledge
from the rudimentary knowledge base (KB) of the Mind.rexx program.

As Mind.rexx did, Mind.Forth will store an incoming sentence both
as a short stream of quasi-phonemic memory in the array ear{ },
and as *temporally* related nodes among the concepts of psi{ } and
uk{ }.  For an input like, "Horses eat...(what?)," the idea was that
the heightened local (in time "t") activation of the temporally
nearby nodes for the noun "horses" and the verb "eat" would kind of
"slosh over" or "spill over" in a kind of "temp{*filter*}cloud" clustered
around the verb "eat" so that the concept/word "hay" would be
activated -- exactly as happened in Mind.rexx on Sat.26.Nov.1994.

Unfortunately, either such an arrangement (i.e., based on temp{*filter*}
proximity) is "cheating" in the sense that it crudely constitutes
a "look-up table" and therefore no thinking is going on, or, almost
as sadly, we humans do not really make the free conceptual leaps
that we think we do, but rather our memory activations race back
along the *temporal* tracks of our memory channels and only jump
from one concept to another when there was already not a physical
but a *temporal* juxtaposition inside the cluster of concepts --
typically mediated by a verb playing the central role in an idea.

But all is not lost in such a nature of things, and we humans
still rule the roost, because our conceptual meanderings are still
NOT hardwired; they are only predisposed (rigged?) in favor of
past (preterite) temp{*filter*}proximity.  Let us hope that our Logic
overrules any false assertions arising from temp{*filter*}proximity.



Tue, 05 Jun 2001 03:00:00 GMT  
 
 [ 1 post ] 

 Relevant Pages 

1. Forget Mind.REXX; go with Mind.Forth

2. Mind.VB initial translation/port of Mind.Forth PD AI

3. Mind.rexx port to Forth

4. Mind.rexx port to Forth

5. Mind.Forth Is Dead. Long Live Mind.Forth!

6. Mind.rexx AI public domain

7. Porting Mind.Forth to APL

8. Porting Mind.Forth to Smalltalk

9. Mind.rexx AI public domain

10. Porting Mind.Forth to Clipper

11. Mind.rexx AI public domain

12. Mind.Forth AI update

 

 
Powered by phpBB® Forum Software