From: Ben Goertzel (ben@webmind.com)
Date: Tue Jul 17 2001 - 05:31:27 MDT
> I know I'm sounding like a cynic here, but why do we need yet another
> programming language? I will admit, it's been a while (a decade)
> since I've
> looked at LISP, and I may be a lot more familiar procedural
> programming and with
> object-oriented programming in C++ and Java, but I still find it
> difficult to
> justify YAPL.
I have a similar attitude, though with a slightly different flavor...
I am sure that current programming languages are not entirely perfect for
implementing real AI.
However, I feel that in order to design a programming language that *will*
be close to ideal for implementing real AI, one must have a fairly detailed
design for the real AI itself.
With the Webmind project, after many years and many more man-years of
effort, we are just barely getting to the point where it's clear *exactly*
what kind of programming language would be ideal for implementing Webmind.
(But we are not going to make a detailed design or implementation of this
language at the moment, because it would take a long time and we've judged
that our priority now is to get a version of Webmind that really works and
demonstrates some powerful intelligence. For the next version, if we're
well-funded again when building it, perhaps we will go with our own ideal
language, who knows.)
I understand that Flare involves many features that are intuitively resonant
with Eli's ideas about AI. However, decades fo human experience in the
domain of computer programming languages teaches that it's a LONG WAY from a
language that seems intuitively sensible on the face of it, to a language
that is actually fun and productive to use.
Thus, I tend to think that a more likely path to success would be: FIRST
articulate the detailed design for the AI system. THEN, code at least key
parts of it in an existing programming language, or several. THEN, at
earliest, start to worry about how to create a programming language in which
these "key parts" would be expressed much more simply and elegantly and/or
more efficiently. Without some substantial code snippets of "real AI" in
conventional programming languages -- or pseudocode at least -- it's going
to be difficult to really get a new AI programming language right. And I
don't just mean a couple pseudocode examples whipped up for illustrative
purposes. I mean real code doing real things...
Of course, although I strongly feel this would be the best way to go, I'm
also well aware that my judgment has not always been correct in the past!
Perhaps this time Eli's intuition is right on target, and he's identified a
language that, when implemented, will be easy and interesting to use and
will mesh perfectly with his detailed AI designs when they come about.
Sometimes one gets lucky, inspired, etc. -- that's the wonder of the
universe...
-- Ben
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:36 MDT