jfox at ricochet.net
Thu Mar 5 12:45:37 EST 1998
James Yegerlehner wrote:
> Mentifex Maker of Minds wrote in message ...
> >Our offer to the comp.robotics.misc engineers is: If we show that
> >a robot AI is possible in Mind.forth, will you take it from there?
> My $0.02: if you want to have a successful project involving collaboration
> of many people on such a project, you have to use a language-independent,
> distributed software technology: either CORBA or DCOM.
That is an interesting idea. A distributed version of mentifex
using COBRA could have massive intelligence. But I think the
suggestion that Forth version will be offered is intended more
as a demo or educational implementation to provide a working
system not an end-all implementation. A distributed software
technology version like one based on COBRA would be great but
it will need someone who understands both COBRA and Mentifex.
I think Mentifex is inherently distributed object oriented and
have suggested to Arthur that this is the way I would implement
it. But I would have to learn COBRA myself and would not
suggest to Arthur or anyone else that they should have to learn
that step before learning the Mentifex concepts. So far all
I know of is text and a Rexx demo program so I would be very
happy to see some Forth based demo code. It is just one small
step but I think it would be a good third step or so.
> By dictating the
> language (especially Forth), you exclude 90% of the prospective
I don't think Arthur would want to limit the implementations of
the concept to any particular language or OS or whatever. I
think portable demo implementations will help teach the concept.
If Forth code is offered I expect some people will experiment
and provide some documentation and expanded demo programs,
perhaps even with a COBRA interface. As you say that way other
people could provide implementations in whatever language and
they could all work together in the same distributed robot mind.
There is a portable ANS Forth and some people have expressed
interest in providing a COBRA interface there for this sort
of portable distributed open interface stuff.
> There are also two other advantages:
> 1. Clearly defined interfaces between
> the different pieces of the system, and
> 2. the software can be distributed
> over networked hardware. I suspect an
> AI project of any sophistication is
> going to need multiple CPUs, and in order to grow,
> it has to allow the
> addition of new compute power. You get that with distributed objects.
I have said the same thing to Arthur. However since I suggested
that we will provide distributed objects as one of the features
of the Forth AI expert system shell on our Forth Hardware
multiprocessors we think a Forth implementation of mentifex
will be the easiest way for us to go down that path. Maybe
someone can help by providing a COBRA interface demo implementation.
Since we can provide distributed objects with very little code
and get performance on a Forth program that is state of the art
on hardware that is almost cheap enough to give away I have
felt that Mentifex will be a very good demo program.
I think Mentifex is a natural for a distrubuted implementation
for a scalable environment. If we do a good machine Forth
implementation on a machine that does Forth primitives in 2ns
we get state of the art performance, yet we can do this with
100 or 1000 times less silicon than conventional designs. Since
F21 is designed to provide a node with Forth engine CPU, memory
interface, network coprocessor, audio and video coprocessors,
parallel port, rtc, etc on a chip that can be manufactured for
a couple of bucks it will provide the basis of an SMP nearly
ideally suited to a Mentifex implementation.
Jeff Fox jfox at dnai.com
Ultra Technology Inc. http://www.dnai.com/~jfox/
More information about the Neur-sci