From: Eliezer S. Yudkowsky (sentience@pobox.com)
Date: Tue Sep 03 2002 - 00:36:59 MDT
gts wrote:
> John K Clark wrote:
>
>> There will certainly be repair mechanisms so if a atom made an error
>> change it.
>
> Okay, but who is going to repair the repair mechanisms?
Suppose you have N repair systems, each of which is capable of repairing
any of the other repair systems. In this case all repair systems must
fail simultaneously in order for the system to permanently die. If you
add more repair systems over time, the chance of simultaneous failure can
decrease exponentially as the system grows linearly over time, which means
that the total risk, summed over the infinite series, can be bounded (and
small).
You wouldn't really want to use such a merely modular design - you'd want
the whole system to simultaneously be a system and a repair system, where
any surviving fragment can repair all the rest - but I think it
illustrates the concept.
-- Eliezer S. Yudkowsky http://singinst.org/ Research Fellow, Singularity Institute for Artificial Intelligence
This archive was generated by hypermail 2.1.5 : Sat Nov 02 2002 - 09:16:39 MST