Re: [sl4] Re: More silly but friendly ideas

From: Vladimir Nesov (robotact@gmail.com)
Date: Thu Jun 26 2008 - 12:16:54 MDT


On Thu, Jun 26, 2008 at 7:45 PM, Stuart Armstrong
<dragondreaming@googlemail.com> wrote:
> (the challenge is understanding exactly what the term "slave" means).

Likely, there are setups that converge on the precise understanding of
informally specified goal, so that you won't need to start with a
precise definition, and will only need to supply enough information
for the system to be able to single it out from everything else. This
works the same way you distinguish between the improvement in
perception of the same concept and perception of a different concept:
the thingspace is sparse and concepts capture the clusters in it. To
point to a cluster, you don't need to know its boundaries exactly, you
just throw in enough evidence for the target cluster to be brighter
than any other. When the system collects more evidence about the
precise structure of thingspace, it will understand what you meant
better than you did.

-- 
Vladimir Nesov
robotact@gmail.com
http://causalityrelay.wordpress.com/


This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:01:03 MDT