ATM:
One of the very biggest problems with my AI software
for execution on von-Neumann serial machines is that
what we are trying to do (emulate the human brain) really
requires massively parallel ("maspar" processing. I cut
a lot of corners in trying to "fake it" w.r.t. maspar AI.

For instance, when subfunctions of my Think() module
search back in time for previous instances of concepts
which will go into the generation of a sentence of thought,
they generally use the most recent single previous instance --
whereas a better imitation of the brain-mind might let
multiple recent (or even long-term) associations
compete for inclusion in a truly maspar phenomenon.

In fact, maybe I should Web-publish a list of the
shortcuts that I am taking as I attempt to rush a
proof-of-concept AI out onto the marketplace of ideas.

The list of Mentifex AI-or-bust shortcuts would include:
- faking maspar with serial-execution software;
- pretending that regular English is phonemic English;
- treating short-term-memory as if it were long-term;
- hardcoding initial English vocabulary and simple syntax;
- using Rejuvenate() instead of a really large memory;
- limiting the sensorium to audition as a single sense;
- searching for associations instead of a direct access;
- and various other shortcuts perhaps yet to be recalled.

However, I feel quite strongly that the AI problem-space
is so huge that "the first true artificial intelligence"
has to take the risk of all these potentially show-stopper
shortcuts or else AI will not get off the ground for decades.

[...]

One thing that intrigues me about Python is the possibility
of using Python to make what I suppose you might indeed call
"agents," but not in the sense of an AI that will shop for you
or find the best airfares for you and then report back to you.

I would like to branch out into maybe AI in Perl (or Python)
that resides steadfastly on a server with no intention of migrating,
and AI in Python or Java or Telescript or whatever other language
will permit the entire life-history AI to "pull up stakes" and
migrate (in true metempsychosis) from site to site on the 'Net.

http://sourceforge.net/projects/tre is a Python AI project
where I noticed a while back that the Python programmer had
written a "home-made HTTP daemon" in Python. I would like
my AI to have its own HTTP daemon so that the AI will have
the ability to learn by reading all available Web pages.

Then I want the AI to be able to transfer itself to other sites,
or to enter and control a robot where AI access is permitted.

Thanking everyone for their patience,