1.
Theory of
AI4U
Algorithm Steps: Code the English Mind-Module
/^^^^^^^^^\ English As One Syntax Among Several /^^^^^^^^^\ / visual \ / auditory \ / memory \ T / memory \ | _______asso-|ciative | ________ | channel | | /image \rec-|ognition | / FRENCH \ | | | / percept \---|---------+ \________/ | | | \ engram / |tag c|f __________ | | | \_______/ | o|i / JAPANESE \ | | | | n|b \__________/ | | | | c|e _________ | | | | e|r / ENGLISH \ | | | | p|s \_________/---|-------------\ | | _______ | t| flush-vector| | ________ | | | /fresh \ | ___|__ ____V__ | / \ | | | / image \ | / Psi \-----/ En \----|-/ Aud \| | | \ engram /---|----/concepts\---/ lexicon \---|-\ phonemes / | | \_______/ | \________/ \_________/ | \________/ |http://mind.sourceforge.net/diagrams.html shows a Theory of Mind.
The diagram above shows how English may co-exist along with several
other natural languages in the brain-mind of a polyglot human being
or robot. Note that there is only one "Psi" area for concepts --
because concepts are deeper and more fundamental than language
in a brain -- but there may be as many foreign languages as the
individual mind is ready, willing and able to learn. Note also
that there is only one auditory memory channel, which must keep on
file all the vocabulary for all the languages known by the mind.
2. Genesis -- Constructing the English Mind-Module
As a component of the AI Mind software, the English module is
simple and straightforward, with plenty of room for future growth in
functionality and logical complexity. The higher-level
Think Module
calls English as a matter of course in the early AI Minds. In more
sophisticated minds that speak not only English but additional,
foreign languages, the
Think Module might call English by default,
except when addressed in, and therefore trying to respond to,
some other natural human language.
Once English has been called, the mind-module has work to do.
It must immediately decide what kind of English sentences to generate --
subject-verb-object (SVO) statements; negations of SVO statements;
questions; commands; whatever. There will always be some cue to
trigger the type of sentence to be generated. The AI programmers
or their co-worker, the mind-design engineer, may choose to play God
and force the English module to always choose a particular default
syntax, such as Subject-Verb-Object, unless overridden and over-ruled
by special cues and circumstances. As long as the programmers put in
the various syntactic options, there may be a whole series of
default selections of syntax for the AI to think in. When all possible
English syntax has been hard-coded, a kind of saturation may set in --
but not for long.
When an AGI Mind is truly advanced, the hard-coded syntactic options
will be thrown out, and a spiral learning algorithm will be put in.
Then the machines will learn to speak English by trial and error,
the way a human baby does. The English mind-module software will
be more fluid and more adaptive to learning from trial and error.
The baby robot mind will perform trials by inserting a building-block
of syntax -- say, a noun or a preposition-plus-noun -- into a
pre-exisitng syntactic loop for sentence-generation. If the new syntax
works well enough, it will be retained indefinitely as a part of
the loop that spirals forward into the future. If the new syntax
is erroneous, problems will develop and the offending syntactic
node will be deleted from future spirals of the loop of nodes.
So the baby AGI Mind will learn language by means of the
insertion-and-deletion spiral that adds or subtracts syntactic
nodes in a syntax loop.
In the software code that follows, the English mind-module
may be still at a very early, even primitive stage.
There may be only enough functionality to demonstrate
the most rudimentary thinking in the artificial mind.
Before English may call more sophisticated patterns
of thinking, the syntactic sub-modules msut be programmed,
tested, debugged and included in the full AGI code-base.
All this work must be done on an ad hoc basis,
unless a clever programmer has released "self-modifying"
AGI code into the wild. Meanwhile, improvements in the
English mind-module code may be inserted below without notice,
and without any special changes in the text of this document.
http://mind.sourceforge.net/ai4u_157.html
is an overview of Mind.
// English() is called by Think and in turn calls
// subject-verb-object SVO or another syntax structure.
function English() { // ATM 12aug2002; or your ID & date.
if (recon > 0) Ask(); // If urge to reconnoiter or to learn
else { // is positive, then ask a question, but otherwise...
if (jux == 12) negSVO(); // If verb has 12="not" negative adverb.
else SVO(); // otherwise call the "positive" SVO syntax.
jux = 0; // Safety measure of resetting "jux" to zero.
} // End of _not_ asking a question.
} // End of English(); return to Think().
\ ENGLISH is called by THINK and in turn calls SVO
\ or any other particular English syntax structure.
: ENGLISH \ ATM 3aug2002; or your ID & date.
recon @ 0 > IF ASK \ If urge to learn...
ELSE \ If no novelty...
jux @ 12 = IF \ If negative 12=not adverb,
negSVO \ transformation of Chomsky;
ELSE SVO \ the "positive" S-V-O syntax.
THEN \ End of test for verb-negation.
THEN
; \ Return to the THINK module which calls ENGLISH syntax.
5. Analysis of the Modus Operandi
The English module source code shown above is beginning to grow
complex, because the AI Mind must select one of several possible
English syntax structures for a thought about to be generated.
A straightforward syntax module may be selected by default,
while the choice of a more complex transformation may hinge
upon such criteria as how much (full or incomplete) information
is available to the mind trying to generate a thought, or how
a higher deliberative process in the mind may try to couch
or conceal the perhaps limited information to be conveyed.
Higher up in the scale of the mental architecture, the Think
module may be faced with a question of which language to think in.
6. Troubleshooting and Robotic Psychosurgery
The
AI Debugger program may shed some light in general on how to debug
and troubleshoot programs in artificial intelligence.
7. English Resources for
Seed AI Germination and Evolution
