RE: "Real" programming and seed AI

From: Ben Goertzel (ben@goertzel.org)
Date: Thu Feb 07 2002 - 17:59:20 MST


I take a fairly dim view of complaints about hardware, OS's and programming
languages not being suitable for AI. Of course there is some validity to
the complaint, but it's often used as an excuse for lack of AI progress when
the real problem is lack of conceptual understanding.

One view is:
"If you say you know how to make an AI, put up or shut up. Write up the
design in equations. Write it up in pseudocode. Write it as a formal
program specification. Show us an AI design that looks workable, show us
why it can't run efficiently on current hardware, OS's and programming
languages, show us in detail why you think it *would* work on some other
kind of hardware."

Another contrasting view is:
"No, I don't know exactly how to make an AI, but I have an intuition about
it, and I could flesh out this intuition into a detailed design IF I had the
right kind of hardware/OS/programming-language to experiment with."

There is no doubt that if I had the right massively parallel hardware to
use, equipped with a suitable OS & programming language and
debugging/testing framework, then the process of arriving at the Novamente
AI design would have been vastly faster. On the other hand, if I had
started out the (Webmind, the predecessor to Novamente) project with a fully
detailed AI design, then I would have been able to figure out pretty well
from the start how to get it working on the hardware at hand (clusters of
SMP machines).

As it is, we started out implementing Webmind by implementing a really
beautiful distributed agents framework, which was basically an "AI OS"
running on top of Java, suitable conceptually for implementing and
experimenting with our AI designs.

Unfortunately, in spite of a lot of clever ideas, we failed to make this AI
OS adequately efficient. But experimenting with our AI ideas running on top
of it helped us to create a detailed mathematical AI design (Novamente).
Once we arrived at this, we saw how to implement our design on the
hardware/OS/prog-langs at hand. Not elegantly but workably. And that's
what we're doing now -- since April 2001 -- with a loooong way to go since
we only have small team and minimal $$ these days.

-- ben

> -----Original Message-----
> From: owner-sl4@sysopmind.com [mailto:owner-sl4@sysopmind.com]On Behalf
> Of Remco Gerlich
> Sent: Thursday, February 07, 2002 5:25 PM
> To: sl4@sysopmind.com
> Subject: Re: "Real" programming and seed AI
>
>
> On 0, Alan Grimes <alangrimes@starpower.net> wrote:
> > My biggest problem is that I don't have an OS that I would consider even
> > remotely suitable for work in AI.
>
> I thought that was an interesting statement. You seem to have ideas about
> the services that an OS suitable for AI should provide. Could you
> elaborate?
>
> [ Hi, I'm new here, I'm extremely interested in everything
> concerned with AI
> but quite skeptical about the singularity/etc thing, I've lurked
> for a week
> or so and think I'll like this list, I hope I'm welcome despite
> the skepticism ]
>
> --
> Remco Gerlich
>



This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:37 MDT