And the problem is not that I can't get up and 
walk away; it's that I want to keep working here 
until it's obvious that AI Has Been Solved.
Meanwhile, time to take a break from the work 
documented below. Better respond to SJA and BenG.

1 Sun.10.OCT.2010 -- S-V-O Revisited 

Yesterday we had not uploaded new MindForth AI 
code to the Web for three days, and we had 
backtracked to a slightly earlier codebase of the 
AI, so we hastened to upload the 9oct10A.F(orth) 
version with its various enhancements, even though 
we had begun troubleshooting subject-verb-object (SVO) 
glitches and had not yet re-established SVO order. 
To upload the new code with an SVO challenge in it 
seemed actually like a good way to generate an 
eagerness to re-engage with a new coding session and 
to work on SVO revisited, now that great progress has 
been made on answering bi-directional be-verb based 
queries, that is, queries about the "YOU" other-concept 
and about the "I" self-concept. We even feel that S-V-O 
troubleshooting is much more straightforward and therefore 
easier than be-verb troubleshooting. So we rename 9oct10A.F as 
10oct10A.F and we delve into problems with SVO input and output. 


2 Sun.10.OCT.2010 -- 

Now we have something strange going on with neural inhibition. 
After we make a few be-verb input statements -- to make sure 
that the AI handles them correctly -- we type in an SVO 
statement like "cats eat fish". The AI erroneously answers, 
"FISH HELPS KIDS", but that glitch is not the main mystery. 
The main problem is that that the output word "KIDS" is 
showing up with -31 inhibition, even though the "predflag" 
variable, at zero, is not indicating a predicate nominative. 
There must be some left-over, vestigial code that is 
unwarrantedly inhibiting "KIDS" as a direct object. Oh, in 
NounPhrase there was indeed such code, with an "OR IF" 
statement letting either a direct object or a predicate 
nominative be inhibited. We comment out the "OR IF" code 
and we reinstated only the code for predicate nominatives, 
to see what happens. 

OK. We now have a clear and definite problem. After a few 
be-verb inputs, when we type in "cats eat fish", the verb 
75=HELP has acquired so much stray activation through 
association from be-verbs, that the AI outputs "FISH HELPS KIDS". 
Our obvious task here is that somehow we need to work on 
not letting stray activations build up too mightily. Perhaps 
we need to make better use of PsiDecay in order to bring 
stray activations down quite rapidly. Perhaps we need to 
implement some sort of clean-up routine so that, when a 
be-verb response has been made to a be-verb input, the AI 
will go around and make a special effort to damp down the 
aftermath of stray activations. In general, we are aiming 
here for a well-tuned AI Mind that activates concepts just 
enough to generate a logical thought while preventing excess 
build-up of stray activations. 

Mentifex
-- 
The un-credentialed AGI guy :-) -- but see:
http://www.chatbots.org/ai_zone/viewthread/240/


-------------------------------------------
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