From: Gordon Worley (redbird@rbisland.cx)
Date: Sun Nov 25 2001 - 18:07:00 MST
On Sunday, November 25, 2001, at 06:31 PM, Simon McClenahan wrote:
> Let's say that the path to superhuman AI lies in enhancing current human
> intelligence. The SAI will probably still have human feelings, but more
> "enhanced", or "faster". Would persuasion techniques become more
> difficult to
> implement? Would it be more difficult, even for a SAI, to program an
> agent
> surfing the web trying to learn from all published human knowledge, but
> it reads
> mostly spam and advertisements?
Um, you want the SI to learn to do this NPL stuff (I'm familiar with the
concept under other terms)? I think you'd better read the objects to
creating an AI jail in this thread:
http://sysopmind.com/archive-sl4/0106/0063.html
> What about an unfriendly AI that learns all about NLP and uses it to
> convince
> humans, that it is friendly?
Again, the jail thread.
> To answer Jordan's original question, I believe that NLP and any other
> human
> psychological model is very relevant to an AI model, although at a
> higher level,
> or "chunk" in NLP-speak. If we send a stream of random data into an AI's
> processing cortex, effectively blinding or deafening or causing a
> hallucination
> without ver knowledge, would we be able to predict its behaviour? Would
> we be
> able to fix it?
Um, why would you do something like this? Just to see what happens?
Doesn't really read like a good idea. Besides, if the AI is intelligent
enough, ve should know what would happen and have blocks in place to
prevent such a stream from getting in.
> One of my favourite metaphors for God worship my
> father told me, "if there is a God and He's pointing towards
> enlightenment, why
> would we be sucking on his finger?"
God's wire-headed. :-)
-- Gordon Worley `When I use a word,' Humpty Dumpty http://www.rbisland.cx/ said, `it means just what I choose redbird@rbisland.cx it to mean--neither more nor less.' PGP: 0xBBD3B003 --Lewis Carroll
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:37 MDT