From: James Rogers (jamesr@best.com)
Date: Mon Apr 16 2001 - 11:50:05 MDT
At 09:30 PM 4/15/2001 -0700, Mitch Porter wrote:
>This, I don't understand. What are the implementation
>problems it gives us to solve?
>
>Back when I made my post in February, I thought
>that the architecture of an SI might consist
>(in these terms) of a UI core, and peripherals
>which translate domain-specific problems into
>a form the UI can solve. Is the creation of
>these peripheral modules the sort of implementation
>problem you have in mind?
I'm too busy at this moment to give this a proper response, but I'll make a
quick comment or two.
Some good work on universal predictors with finite resource constraints has
come out of Israel within the last ten years that offer good insight into
the problem. I can't give references right now because I'd have to dig to
find them, but they aren't AI related.
The biggest problem of UI is that it has a general computational complexity
of the form O(k^n), which is nasty because "n" could get large quickly for
our purposes. Practical tractability can be maintained using strategies
that keep "n" small (a general engineering rule, not just applicable here).
More later,
-James Rogers
jamesr@best.com
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:36 MDT