From: Ben Goertzel (ben@goertzel.org)
Date: Wed May 04 2005 - 17:44:41 MDT
Hello to all --
It's been quite a while since I issued a serious "call for programmers" to
this list.
So, here goes ;-)
The purpose of this email is twofold: an update on the Novamente AI project
(see www.agiri.org, and http://www.goertzel.org/research.htm) and a renewed
call for help.
First the update: Progress on Novamente has been nonzero but frustratingly
slow.
Cassio (my lead collaborator) and I have been spending a lot of time on
Biomind,
our bioinformatics startup (www.biomind.com), and have been doing lots of
great
stuff, but not so much AGI stuff as we would have liked. We have also spent
some time
building a narrow-AI NLP system for a government customer here in
Washington, which may be very useful as a kind of interface to Novamente,
but again isn't really an AGI-approach to language processing...
I've been gradually polishing the Novamente book manuscripts, and those are
actually nearing completion, miraculously enough....
Izabela (my lovely new wife) and I did a lot of work fine-tuning
probabilistic
inference in Novamente; Luke Kaiser and Moshe Looks made a programming
language
(Sasha) that compiles into Novamente nodes and links; and Moshe integrated
procedure learning and pattern mining using the Bayesian Optimization
Algorithm
into Novamente....
At this stage we have a lot of nice AI components integrated (or
semi-integrated ;p) with the NM core but still no f**king AGI, and we still
haven't made a serious effort to hook everything together in an AGI
context -- embarrassingly enough.
However, things started moving in a more AGI-productive direction at the
start of 2005 when we got a small investment enabling us to bring Ari
Heljakka on board full-time to work on NM/AGI stuff. Ari has been mostly
working on the AGI-SIM simulation world and has gotten it to a state where
it runs (though it's still quite incomplete) and can be worked on and
improved by others.
It's become quite clear to me that we will NEVER get an AGI built at this
sort of rate -- doing AGI in our spare time while working on commercial
projects, and using part-time volunteers to help out now and then, just
isn't gonna cut it. We need to raise some dedicated AGI funding. And to do
that, we need to not only have me publish the NM books (finally) but to make
an at least moderately wizzy demo of NM doing something cool ....
In this light, AGI-SIM is intended to serve a dual role: first to provide a
framework in which we can teach Novamente and in which it can potentially
interact with other AI systems, and second to provide a framework for
building a wizzy demo to show off Novamente to potential
investors/philanthropists...
For information on AGI-SIM (an open-source project), see
http://sourceforge.net/projects/agisim/
Our goal for 2005 is to create a wizzy demo of Novamente doing some simple
learning in the AGI-SIM world, and then take this demo to robotics firms and
try to get a co-development deal in which they fund us to
1) simulate one of their robots in AGI-SIM
2) make Novamente successfully cognize in a simulation of one of their
robots
the goal being of course to then actually make NM control one of their
robots (probably via Wifi not directly putting NM servers onboard).
[To operate within AGI-SIM, NM will need some simple vision processing, but
it can be relatively low-bandwidth compared to human vision, so this won't
be a very big deal.]
AGI-SIM is a Linux-only system at the moment but could be ported to VC++
pretty easily, as the underlying CrystalSpace graphics engine runs in C++.
It is still at a fairly primitive stage and needs a lot of work, I'm trying
to recruit volunteers for the project.
Ari has recently shifted from AGI-SIM work to working on reimplementing
Probabilistic Term Logic inference in Novamente (the prior implementation
was too researchy and incomplete, now that we understand the theory
thoroughly we also understand how to make the implementation "right"). He
will then have a lot of work to do on the Novamente side to get NM ready for
AGI-SIM learning/interaction.
So one thing we could REALLY use right now would be a couple people to focus
attention on AGI-SIM itself. This is not AI coding per se, it's server-side
C++ coding. It's not mainly graphics coding per se either, because the
graphics is done in the CrystalSpace simulation engine. But it's definitely
not *easy* coding, this demands pretty good C++ programmers. At the moment,
just compiling AGI-SIM is a pain, due to the numerous dependencies on
various open-source packages...
If you want to help out w/AGI-SIM (or Novamente, of course ... I'm just
suggesting AGI-SIM at the moment because it's a current "pain point" and
it's more approachable than NM-core coding ... plus it's open-source which
may be a bonus from the volunteer's perspective) please let me know.
Thanks
Ben Goertzel
This archive was generated by hypermail 2.1.5 : Tue Feb 21 2006 - 04:22:56 MST