Miniature eurekula: how to design a functioning mind

Mentifex mentifex at my-deja.com
Thu Jan 20 00:34:39 EST 2000


Am 17. Januar 2000, Montag Morgen in der Zoka.  This morning
I created two new GeoCities Web pages
http://www.geocities.com/Athens/Agora/7256/flushvec.html and
http://www.geocities.com/Athens/Agora/7256/decay.html
and I wrote some new material into The Art of Computer
http://www.geocities.com/Athens/Agora/7256/acm.html Mindmaking.

Over the last two weeks, at the local library at ten cents a page
I have printed out the approximately 32 documentation pages of the
http://www.geocities.com/Athens/Agora/7256/mind4th.html AI source
code.  I have inserted the documentation print-outs into a blue
binder in alphabetical order.  Now, in an iterative process, I
am going through the documentation pages both on-line and in
hardcopy, further improving each documentation file in a quest
for uniform excellence.

I like the idea that my "acm.html" steps to coding AI are
becoming ever more thorough.
                             [spaeter, am Nachmittag zu Vaierre]
At Vaierre I am wrestling with the Mind.Forth SPREADACT problem.
Before I code anything, I am considering all the options.

There is no reason for the AI to parrot back whatever it has
just heard.  If the user types, "Cats eat fish," the AI could
conceivably respond, "FISH EAT BUGS."

It is obvious that the AI will operate on a chain of spreading
activation, but it is not obvious how that chain will snake
through the mindgrid.

Miniature eurekula:  One way to free up the AI from parroting
back whatever it hears, would be to make sure that the freshly
recorded concept-nodes were at a somewhat lower activation-level
than the preterite nodes suddenly reactivated through the
SPREADACT process.  This insight is a valuable technique for
computational linguistics, because it permits a general mechanism
of favoring reflective thought over parrotlike repetition of
input.  The technique is quite easy to code.

So now we have thought our way through to a technique of
having inputs cause reactivations at a higher intensity than
that of the inputs themselves.  It's as if to say,
reactivated preterite concepts shall be more important than
concepts of immediate input.

Well, I actually achieved something today, when my main impetus
was simply to start writing down some thoughts.

I am at a strange juncture where I feel that solving this
SPREADACT problem may turn out to be by far the most important
event in my life.  My Mind.Forth AI is chafing at the bit,
ready to burst out and proliferate across the Internet, if I
can only remove this SPREADACT obstacle at the heart of the
AI program.

--
Forth is the lingua franca of AI, embedded NC and robotics.
For details on Mind.Forth and the public domain AI project,
see the ACM Sigplan Notices 33(12):25-31 (December 1998),
"Mind.Forth: Thoughts on Artificial Intelligence and Forth."


Sent via Deja.com http://www.deja.com/
Before you buy.




More information about the Neur-sci mailing list