JavaScript AI Mind Programming Journal -- Sat.2.OCT.2010

1 Sat.2.OCT.2010 -- The Royal Jelly Principle 

We have stumbled into a minor breakthrough in 
our Mentifex AI coding. Last January (2010) 
in MindForth we were coding elaborate schemes 
to answer who-queries and what-queries in the AI. 
Then on 5 September 2010 we developed a technique 
of using neural inhibition to simply answer the 
same what-queries for which we had written 
over-complicated code in January of 2010. 
We wanted to dismantle the complicated query-code, 
but we did not want to lose any of the improvements 
and advances that we had meanwhile incorporated 
into the code-base along with the complex 
query-response code. We decided to keep on coding 
and to remove one small item at a time from the 
complicated query-code. Then we decided to bring 
the JavaScript AI (JSAI) up on a par with MindForth. 

In coding the JSAI, we wished that we could keep 
just the query-subject variable from the overly 
complicated query-response code. It seemed a 
shame to work so many hours on query-response 
in January and then to abandon all the fruit of 
such hard work except for the variable, but now 
we see an AI breakthrough shining on the horizon. 
If we use "qusub" as the new name for a query-subject 
variable, we can start tagging each emerging 
thought-subject and each re-activated KbTraversal 
concept as a provisional "qusub", holding onto the 
"qusub" for one cycle of thought and not caring 
whether the "qusub" concept is actually used as 
the subject of a query. It is as if all 
thought-subjects are like honeybee eggs with 
the potential to mature into queens, depending on 
whether or not they are fed royal jelly. Likewise, 
each former thought-subject may or may not mature 
into the linguistic subject of a query-thought, 
depending on whether or not the dynamics of the 
AI Mind require a query-subject. If each briefly 
dominant thought-subject is tagged as both the 
"subjold" old subject and the provisional "qusub" 
query-subject, then our AI Mind software becomes 
implicitly and inherently more powerful and more 
pregnant with possibilities than we ever imagined 
it would be. 

We note in passing that we have devised a way to 
tag subject-concepts not by encumbering them 
internally, but by referencing them externally. 
Each subject-concept is momentarily and 
provisionally a "subjold" concept and a "qusub" concept, 
whether or not any use is made of that status. 
When Netizens say "and then something magical occurs", 
this hidden power of AI concepts is perhaps the 
magic being alluded to. 

2 Sat.2.OCT.2010 -- Debugging the WhoBe Glitches 

By inserting quite a few "alert" messages, we have 
determined that the JSAI was saying "WHAT" as its 
first utterance because some old code at the end of 
NounPhrase was directing the utterance of 54=WHAT 
when NounPhrase could find no candidate concept. 
Instead of just commenting out the offending code, 
we have added the word 109=HELLO at the end of the 
EnBoot sequence and mutatis mutandis changed the 
NounPhrase code to say "HELLO" instead of "WHAT". 
This method is a rather clumsy way of getting the 
AI Mind to say "HELLO" to human users, but at least 
it is a start. 

3 Sat.2.OCT.2010 -- Flushing out the Blank "aud" Fetch 

By inserting a diagnostic alert before every SpeechAct 
call, we have traced the origin of blank "aud" fetches 
to the end of the BeVerb module. There we simply 
knocked out the SpeechAct call, and the AI no longer 
created empty auditory word-stretches. Next we used 
the new "qusub" query-subject variable in WhoBe to 
cause the AI to ask much more sensible WhoBe questions, 
because the "qusub" variable was retaining the proper 
subject for enquiry. 

Mentifex
-- 
http://www.scn.org/~mentifex/AiMind.html


-------------------------------------------
AGI
Archives: https://www.listbox.com/member/archive/303/=now
RSS Feed: https://www.listbox.com/member/archive/rss/303/8660244-d750797a
Modify Your Subscription: 
https://www.listbox.com/member/?member_id=8660244&id_secret=8660244-6e7fb59c
Powered by Listbox: http://www.listbox.com

Reply via email to