From: Ben Goertzel (ben@webmind.com)
Date: Mon Jul 30 2001 - 06:25:30 MDT
> Yes, I strongly got the impression that WM nodes/links/etc. might have
> some of the properties of Flare.
Definitely. You could take Flare syntax and have it interpret to Webmind
schema, to be executed on the Webmind core.
However, in general, Webmind schema are not as efficient as C programs doing
the same thing -- far from it.
Where Webmind schema excel over C programs is when they refer to other
Webmind services at many points during their execution.
So, schema execution is *not at all* efficient as a general programming
language compiler, but it's efficient for the execution of the *particular
class* of programs that we believe are crucial to cognition (programs that
involve the combination of the activity of various WM modules).
Just to be clear: As of today, we don't have a sophisticated schema module
yet in the new C++ version of WM. The one in the Java WM was an interesting
prototype but very inefficient and not suitable except for toy examples.
> That's one of the little ironies of all
> this - Webmind would probably find Flare even more directly useful than we
> would, since Flare will probably correspond fairly well to the mindstuff
> of the Webmind AI. On the other hand, this same property means that,
> since Webmind's cognitive processes are Flarish, Webmind can be written in
> Java with only one added layer of abstraction.
Or in C++, as is the case with the current version
> SIAI, though, needs Flare
> just to get started. Frightening concept, but there it is.
Because your approach to AI is somehow higher-level than ours? I guess this
won't be clear to me until you explain more about your approach to AI. I
get the sense that you want to put causal inference of some sort at the
center of it, but I don't understand what kind of scheme you intend to use
to represent other types of knowledge (the many varieties of non-causal
knowledge), and I also don't understand exactly what mathematics you intend
to use to carry out uncertain causal inference (you've mentioned Bayesian
learning before, but as Pei pointed out in a private e-mail to you, there
are many known shortcomings of Bayesian methods in such contexts). Now
*this* is the e-mail thread I'd really like to have... but perhaps it's not
the time for it yet...
-- Ben G
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:37 MDT