From: polysync@pobox.com
Date: Mon Jul 08 2002 - 13:12:30 MDT
> A 50/50 chance. Yes, those are your odds. And yes it could do things vastly
> worse or better than my examples. And you may be able to catch a poorly
> formed AI. But a sane intellectually superior AI of either variety can tell
> you all the things that would make you feel better and you would be no closer
> to knowing.
Do what other industries do: peer review and redundancy. Didn't you write
that story about Eliezer's AI discovering Ben's AI was out to rule the world?
If some number 'X' of research groups independently created 'X' AIs, and they
all interviewed and approved of each other, what would you think then?
Which reminds me of something I meant to ask a while back: Suppose a Friendly
AI is exactly "cloned", and the two go off on their own ways. Several seconds
or mega-years later they meet up in a bar (or on IRC). Could one FAI reach the
conclusion that the other was no longer Friendly?
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:40 MDT