Arthur,
Can it represent negatives? Time? Textures? Relationships?
Distinguish homonyms from context? Represent the concept of a homonym?
Represent itself? Can it handle deixis?
More importantly, do you have any prinicipled reason for claiming that
it will soon be able to handle any of these things, other than your
statement of optimism "If robot builders were to add sensory and motor
routines to Mind.Forth, the AI would flesh out its conceptual knowledge
and interact with the world."?
So far, what you describe looks like something I wrote in Basic on a
Sinclair Spectrum computer in 1982.
Richard Loosemore
A. T. Murray wrote:
In Vernor Vinge's classic paper on Technological Singularity:
And what of the arrival of the Singularity itself?
What can be said of its actual appearance? Since it
involves an intellectual runaway, it will probably
occur faster than any technical revolution seen so far.
The precipitating event will likely be unexpected
-- perhaps even to the researchers involved.
("But all our previous models were catatonic!
We were just tweaking some parameters....")
Very truly yours Mentifex here was "tweaking some parameters"
last Wednesday and suddenly the AI Mind in Win32Forth woke up.
It was not an accident. It was a milestone after years of coding
and debugging an AI Mind based on a linguistic theory of mind.
Mind.Forth thinks on the basis of "spreading activation."
When a pre-verbal concept in the AI Forthmind becomes active
enough to initiate the generation of a sentence of thought in
English, the various "Think" modules not only follow the
chain of activation snaking across the conceptual mindgrid,
but they also "midwife" or push the nascent thought as it
seeks out its own as-it-were birth-canal. As an example,
let's take the startling exclamation by an AI Mind of the
sentence: "Ben writes books."
When the concept "Ben" comes to mind, the AI has selected that
concept because it was momentarily the most active concept on
the entire conscious mindgrid. (Perhaps a camera saw "Ben".)
Having reactivated the concept of Ben, the linguistic AI Mind
lets activation spread from instantiation nodes on the
diachronic Ben-concept to various verbs that have been
associated in the past with knowledge of Ben's activities.
For whatever reason, the verb "write" comes to mind in
association with the concept of the persona of Ben.
And what does Ben write? Letters, AGI posts, poems,
rants, shopping lists, extortion notes, BOOKS!
You might think that it is easy to create software that
associates "Ben... writes... books," but until Wed.7.JUN.2006
it was not that simple. (Now any AI Lab anywhere can do it.)
In 2002 I published a book "AI4U" (to compete with "AIMA")
that contained my JavaScript Mind.html code as Appendix A.
The code was buggy and created English gibberish as output.
The output did not start out as gibberish, but it quickly
degraded into gibberish as the AI made spurious associations.
Mind.Forth -- the precursor to Mind.html -- had the same bugs,
but last year in 2005 I inserted powerful diagnostic routines
into Mind.Forth and I gradually removed the very worst bugs.
On 16 March 2005 I also made it possible to press Tab on
the keyboard in order to cycle through such display modes
as (currently) Normal; Transcript; Tutorial; and Diagnostic.
The Tutorial mode was scrolling voluminous messages so rapidly
down the screen, that the user had to kill the AI to read it.
On 9 November 2005 I consolidated the Tutorial horizontally.
The idea behind the Tutorial mode was to show the deep internal
thinking process of the artificial mind. By tweaking the same
code that created the output of the Mind, in Tutorial mode I
was able to show the range of possible thoughts available
to the AI and the forces at play in determining pathways
of thought across the conceptual mindgrid. One golden goal,
however, eluded me until now -- showing the slosh-over effect.
In Mentifex AI lore, the slosh-over effect is what happens
when you think by activating two concepts ("Ben... writes...")
and the spreading activation builds up so strongly on the
second concept (the verb) that it "sloshes over" onto the
third concept -- the direct object in "Ben writes books."
The tweaking that I did last Wednesday was simply to make
a subject-noun concept start out with an equal activation
at all its nodes and pass a spike of activation to verbs --
before the selection of the verb to go with the subject.
When the verb is selected, an increment of activation is
added to all its nodes, including any nodes on which a
modicum of "spike" activation has already been deposited.
When each node on the verb-concept fires its own spike
of relative activation over to an associated object-noun,
the stage is set for the AI Mind to pick a valid object --
not the spurious associations which engender AI gibberish.
There are still some bugs to be worked out, so that
Mind.Forth will home in unerringly on logical thoughts.
But it is already possible to download Mind.Forth and
Tab on into Tutorial mode where slosh-over is visible.
You won't be alone in your testing of Mind.Forth AI.
The recent access logs are showing many visits to
http://mind.sourceforge.net/mind4th.html without
even a referral. Some Netizens who search the Web
for "artificial intelligence download" find the AI.
To pre-empt a criticism that may occur to you, I
would like to state that what Mind.Forth achieves
is not comparable to a database or a look-up table.
The knowledge base (KB) of Mind.Forth consists of
concepts which interact by spreading activation.
If robot builders were to add sensory and motor
routines to Mind.Forth, the AI would flesh out its
conceptual knowledge and interact with the world.
Respectfully submitted,
Arthur T. Murray
--
http://www.blogcharm.com/Singularity/25603/Timetable.html
-------
To unsubscribe, change your address, or temporarily deactivate your subscription,
please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
-------
To unsubscribe, change your address, or temporarily deactivate your subscription,
please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
