1 Fri.22.OCT.2010 -- Combining Inhibition and Slosh-Over.
Looking back to the beginning of the year 2010,
in MindForth AI we see several development tracks
which need now to merge and be better integrated.
In January of 2010 we added the "mfn" gender flag
and we began working on elaborate code for the
handling of be-verb forms. Then in September of 2010
we began using neural inhibition to cycle through
exhaustive responses to KB-queries. More recently,
in October of 2010 we have finally made subject-verb
activational slosh-over dynamically visible in the
MindForth Tutorial display mode.
Sometimes our work on one aspect of MindForth
distorts the functionality of other aspects of
MindForth. When we make sweeping changes for the
sake of cumulative slosh-over, we discover that
we have interfered with the ability of the AI to
make proper responses to queries like "What are you?"
and "What am I?" We then have the happy task of
going back into the free AI source code and
troubleshooting the simultaneous operation of the
slosh-over display and the function of inhibition
during responses to be-verb KB-queries.
Although we uploaded 18oct10A.F as a cleaned-up
(commented-out) version of 17oct10A.F, we now
rename the 17oct10A.F version as 22oct10A.F so
that we may continue our work and embed comments
that reflect the current calendar date of 22.OCT.2010.
2 Sat.23.OCT.2010 -- Making Minor Adjustments
It turns out that our Tutorial slosh-over
message in the uploaded 18oct10A.F version was
a little bit off, because the line of code
"spike @ 30 - spike ! \ from JSAI; 18oct2010"
was changing the "spike" value immediately
after the declaration of the report.
It seems time to get rid of the useless
"verbinc" (verb increment) variable that was
used on 27apr2007. That variable does not even
Between 9oct10A.F and 22oct10A.F we seem to have
lost the ability to tell the AI something like
"you are code" and to get a valid response when
we input "what are you" as a KB-query. In the
meantime we did all the work on slosh-over for
transitive verbs. Somehow we need to reconcile
the transitive and intransitive treatments and
we need to integrate the two functionalities.
3 Sat.23.OCT.2010 -- Tightening Up Activations
When we tell the AI, "you are code",
currently we get a report:
418 : 50 56 0 0 0 7 58 50 to I
422 : 58 63 0 50 50 8 109 58 to BE
427 : 109 42 1 58 50 5 0 109 to CODE
time: psi act num jux pre pos seq enx
As output we get, "I AM I". It may be that
the "I" concept is being stored with too
high an activation.
4 Sun.24.OCT.2010 -- Reconnecting VerbPhrase and PsiDamp
As we try to integrate our recent KB-query inhibition
code and our even more recent slosh-over display code,
we observe that the two functionalities are not mixing
well, and that transitive verbs used in the slosh-over
work are apparently not being psi-damped. So we inspect
our free AI source code and we notice that, for some
time now, the VerbPhrase module has apparently not been
calling the PsiDamp module to knock down the cresting
activation of a transitive (or other?) verb that has
just been included in a thought. Our fully commented
24may09A.F code indicates that VerbPhrase lost a call
to PsiDamp on 14jan2008 as recorded in the following
two lines of code.
\ psiDamp \ 29apr2005 Necessary for chain of thought.
\ psiDamp \ 14jan2008 Commenting out and using lopsi & hipsi.
Now let us try reinstating the call to PsiDamp from
towards the end of the VerbPhrase module. Let us remark
in advance that we are witnessing here the creaking,
rumbling process of the AI Mind taking shape like a
planet subject to earthquakes and continental drift
and asteroid impacts. The artificial intelligence
matures not linearly but zigzaggingly.
When we reinstate the call from VerbPhrase to PsiDamp,
we observe a limited improvement in the performance of
our AI Mind software. Now for the first time we see
the uneasy coexistence of the KB-query inhibition
function and the slosh-over pin-down function.
We are still getting erroneous cross-over activations
which interfere with our short-range goal of shifting
effortlessly back and forth between the inhibition
queries and the slosh-over displays. We must now
decide whether to upload our code that has made
progress but is far from perfect, or whether to
keep on improving the functionality. Probably we
should upload the code, because it has already
reached an important juncture.
12 years ago