From: Ben Goertzel (ben@goertzel.org)
Date: Sat Feb 08 2003 - 18:05:28 MST
> Training an AGI to be a "winner" in competitions before the AGI
> understands the wider context of game-playing, and the non-zero-summess
> of the competition within that wider context, would be a mistake. But
> once the AGI can understand the context, and how humans
> games/competition operate positively within it, then it would be
> counterproductive to *avoid* training how to win games.
Sure, I think we agree, then.
To make a human-childhood analogy -- which I wouldn't want to overstretch --
it's a bit like the difference between teaching a 3-year-old competitive
games, and teaching a 9-year-old competitive games. A 3-year-old does not
understand broader context, and having them do a lot of competitive stuff
really can cause them to develop an overly competitive outlook on life. But
a 9-year-old understands well that "it's just a game"....
Of course, there is an obvious problem in that our understanding of AI
developmental psychology is not going to be very good until AFTER we've
already raised a bunch of baby AGI's to adulthood.... This means that we
may well not allow the first N baby AGI's to reach adulthood. Murdering
them would be unethical ... so maybe we'll keep them as toddlers until we
successfully raise an adult AGI capable of properly bringing them up... ;-)
-- Ben G
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:41 MDT