If it's Wednesday, this must be German we are coding in,
but no, we have had to drop back into coding the English
http://www.scn.org/~mentifex/mindforth.txt AI in order to
implement machine reasoning with inference in German AI.
Yesterday we obtained the following report from MindForth:
[the innate knowledge base KB about kids:]
583 : 528 -7 2 0 0 5 588 835 528 to KIDS
588 : 835 15 2 0 72 8 595 571 835 to MAKE
595 : 571 23 2 0 835 5 0 0 571 to ROBOTS
602 : 571 0 2 0 0 5 473 74 571 to ROBOTS
607 : 849 0 2 0 571 8 610 701 849 to NEED
610 : 701 0 1 0 849 7 0 0 701 to ME
[three words of human input come in:]
617 : 589 16 1 0 0 5 621 800 589 to BOYS
621 : 800 0 1 0 589 8 626 528 800 to ARE
626 : 528 -15 2 0 589 5 0 0 528 to KIDS
[AI creates an inference about boys:]
628 : 589 6 0 0 0 5 629 835 589 to BOYS
629 : 835 15 0 0 589 8 0 571 835 to MAKE
630 : 571 12 0 0 0 0 0 0 571 to ROBOTS
636 : 528 -13 2 0 0 5 641 835 528 to
641 : 835 -31 0 0 528 8 649 571 835 to MAKE
649 : 571 -15 2 0 528 5 0 0 571 to ROBOTS
time: psi act num jux pre pos tqv seq enx
Although the above report shows that the
English AI Mind _created_ the inference
(boys are kids --> BOYS MAKE ROBOTS),
the engrams are not yet stored robustly
enough to permit the AI to think the thought
stored as a silent inference in AI memory.
The inferential thought, "boys make robots",
cannot be retrieved until it has all the
essential parameters for its retrieval,
such as number for nouns and verbs and
person for the verb involved.
This particular coding of inferential thought
in the Mentifex AI Minds is for me (Mentifex/ATM)
as an independent AI scholar absolutely stunning,
because it is the first time in nineteen years
that the AI thinks up some thoughts of its own.
Previously the AI could answer questions by
retrieving known facts from its knowledge base,
but it could never before take two different
facts and put them together into a novel idea.
Here is the current (unfinished) Forth code:
: InFerence ( create silent triples for machine reasoning )
CR ." InFer: subjnom prednom = " \ test; 18dec2012
subjnom @ . prednom @ . CR \ test; 18dec2012
midway @ t @ DO \ search IdeaPlex to infer facts; 18dec2012
I 0 psi{ @ prednom @ = IF \ KB data? 18dec2012
I 4 en{ @ 1 = IF \ nominative? 18dec2012
seqverb @ 0 = IF \ only once; 18dec2012
I 6 psi{ @ seqtqv ! \ transfer; 18dec2012
I 7 psi{ @ seqverb ! \ transfer; 18dec2012
seqverb @ seq ! \ test; 18dec2012
THEN \ end of test for not-yet-declared; 18dec2012
CR ." InFer: t psi seqverb = " \ test; 18dec2012
I . prednom @ . seqverb @ . \ 18dec2012
\ LEAVE \ at first make only one inference; 18dec2012
THEN \ end of test for nominative; 18dec2012
THEN \ end of test for finding prednom facts; 18dec2012
-1 +LOOP \ end of backwards loop; 18dec2012
1 t +! \ increment time "t" by one for a gap; 18dec2012
1 t +! \ increment time to create an inference; 18dec2012
subjnom @ t @ 0 psi{ ! \ subj of inference; 18dec2012
5 t @ 5 psi{ ! \ pos=5 noun; 18dec2012
\ 999 t @ 6 psi{ ! \ test; remove; 18dec2012
t @ 1 + t @ 6 psi{ ! \ psi6=tqv; 18dec2012
CR ." InferB: storing seqverb = " seqverb @ . \ 18dec2012
seqverb @ t @ 7 psi{ ! \ seq is the verb; 18dec2012
subjnom @ t @ 8 psi{ ! \ enx; 18dec2012
1 t +! \ increment t for storage of verb; 18dec2012
seqverb @ t @ 0 psi{ ! \ verb of inference; 18dec2012
subjnom @ t @ 4 psi{ ! \ pre of verb; 18dec2012
8 t @ 5 psi{ ! \ pos=8 verb; 18dec2012
seqtqv @ 7 psi{ @ t @ 7 psi{ ! \ seq 18dec2012
seqverb @ t @ 8 psi{ ! \ enx; 18dec2012
1 t +! \ increment t to store direct object; 18dec2012
seqtqv @ 7 psi{ @ t @ 0 psi{ ! \ dir. obj 18dec2012
seqtqv @ 7 psi{ @ t @ 8 psi{ ! \ enx 18dec2012
1 t +! \ increment time "t" for an ending gap; 18dec2012
0 becon ! \ reset after use; 18dec2012
0 prednom ! \ reset after use; 18dec2012
0 seqtqv ! \ reset after use; 18dec2012
0 subjnom ! \ reset after use; 18dec2012
\ QUIT \ test; remove; 18dec2012
; ( http://code.google.com/p/mindforth/wiki/InFerence )
We will not code Wotan Supercomputer AI inference until
we first get it working in English. Of course, we will
also port machine-reasoning inference into the
http://www.scn.org/~mentifex/AiMind.html JavaScript AI,
and we expect it to look quite remarkable when the
Web-accessible AiMind.html shows that it can reason!
Mentifex (Arthur)
--
http://www.scn.org/~mentifex/AiMind.html
http://www.scn.org/~mentifex/Dushka.html
http://www.scn.org/~mentifex/mindforth.txt
http://www.scn.org/~mentifex/DeKi.txt
-------------------------------------------
AGI
Archives: https://www.listbox.com/member/archive/303/=now
RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424
Modify Your Subscription:
https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657
Powered by Listbox: http://www.listbox.com
