From: hal@finney.org
Date: Wed Jun 27 2001 - 22:40:05 MDT
Brent writes:
> Let me ask you this: If an AI you were designing was asked:
> "What is the taste of salt like?" how should it answer? Would it be
> lying if it said it knew what the taste of salt was like? Will we
> ever be able to "eff" such ineffable sensations to others and to "AI"s
> - perhaps causing them to say something like: "Oh THAT's what salt
> tastes like!"?
I'm not sure this is a good example, because it is unlikely that anyone
will be designing AIs with a sense of taste any time soon. That is
probably about the least useful sense to build into an AI since the main
function of taste is for selecting healthful food, and AIs are unlikely
to get their energy from food.
Maybe a better example would be something to do with hearing. We already
have systems that do speech recognition, so hearing would be a very
useful sense for an AI. Plus as you may know Kurzweil is an expert on
speech recognition.
We might ask an AI, "what is the sound of my voice like?" It could
answer, your voice is moderately low pitched, similar to these other
people's voices. Or, your voice has a unique rhythm and cadence, but
it is similar to that of these other people.
Is this "effing the ineffable"? Or just a mechanistic set of
correlations? Or is there any difference?
I agree that it would be interesting to hear what Ray Kurzweil has to
say about this. No doubt he has given the issue considerable thought.
Hal
This archive was generated by hypermail 2.1.5 : Sat Nov 02 2002 - 08:08:19 MST