From: Michael Roy Ames (michaelroyames@hotmail.com)
Date: Sun May 19 2002 - 21:09:28 MDT
Justin Corwin wrote:
>
> The question is, how important is experimentation in providing experience?
>
Experimentation is going to be the most important way to impart experiences
to any AI.
Here's three becauses...
i) It is several orders of magnitude more difficult to provide
accurate/relevant/correct experience by hard-coding it, than by designing &
running the AI through a (long) list of learning experiments. This
'because' is intended to apply to hard-coding mental content of equivalent
complexity and depth-of-connectedness as experimental I/O would provide.
Giving the AI mental content less complex than that acquired through
experience would (of course) reduce the difficulty.
ii) Learning experiments will lead to richer and more relevant mental
content than 'given' content, because the *experimental failures are just as
important as the successes*. Hard-coded content is not likely to include
all the failures that the AI will make while experimenting.
iii) During the learning experiments, there will be an excellent opportunity
to detect bugs in implementation. Bugs will manifest with 'given' content
also, but they will appear all-at-once, rather than one-by-one at specific
points in the lessons. Experimental learning will allow you to pin-point
the bugs more easily.
I'm sure there are more reasons, but those are the first three I could think
of.
Michael Roy Ames.
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:38 MDT