RE: Flare

From: Ben Goertzel (ben@webmind.com)
Date: Tue Jul 17 2001 - 14:52:30 MDT


> To my mind, one of the past
> problems of AI is the tendency to write code for things that should be
> higher-level cognitive content.

That's often true of logic-based/rule-based AI, not so of neural net based
or evolutionary programming based AI

> Regardless, Flare may not be the perfect tool out of the box, but I'd be
> really surprised if it isn't (a) a much better starting point and (b)
> easier to improve.

I'm afraid you may be underestimating the vast amount of work required to
make a scalable, efficient programming language.

It can be done, to be sure.

For instance, Michael Korn, who we met at Extropy and who runs
InvestByAgent.com, and his team created their own LISP variant for building
their AI agent system. But it took them quite a while, and it was only a
variant of LISP, not all that fundamentally new.

I'm personally quite enamored of purely functional programming languages
such as Haskell, but, there is as yet no really efficient Haskell compiler,
alas.

> I'd love to hear about it. Everything you said about the need to handle
> large expanses of RAM is now very firmly enshrined in my heart, right next
> to my previous convictions about SMP and Beowulf. We have different AI
> models, and very different beliefs about the interaction of AI and
> programming language, but it still might save Webmind some effort.

Sure, I'll post something next week about what kind of programming language
would in my view be ideal for Webmind AI. Or sooner if the mood strikes me
;)

I think that having the right programming language could make self-modifying
AI go a bit smoother. But still, I believe that understanding the
conceptual basis of its own functioning is a much bigger task for a young
would-be-self-modifying AI to overcome, than understanding the particulars
of the programming language in which it's implemented. An easy language
will make the first stages of moving toward intelligent self-modification go
a bit faster, but it won't help with the hard parts.

Ben



This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:36 MDT