From: Eliezer S. Yudkowsky (sentience@pobox.com)
Date: Thu Apr 13 2000 - 14:48:34 MDT
Brian Atkins wrote:
>
> This is a _good thing_ ?? Ack, I was just beginning to hope near-future
> (5-10 years) computing power would be capable of running an AI. If we
> have to simulate some kind of quantum effects to get a real brain then
> that is going to be a pain in the ass.
I didn't say it was _good_. I said it was _coooolll_.
Besides, you gotta think of it as a hidden variable, not something you
can push on. If we find out that the brain uses quantum whatever, then
it is _good_ that we have _found out_, not _bad_ because it makes AI
harder. AI hasn't gotten any harder. It was always that hard. Now we
know it, that's all.
Besides, I don't think the quantum whatever is necessary to
intelligence, just qualia, and I see no reason a seed AI needs it. My
current thinking (as of the last year or so) is that regardless of
whether human intelligence uses 10^17 qops or just 10^17 flops, it
should be possible to build a reasonably intelligent AI on 10^15 flops
or less. I'm not sure that human intelligence uses neurons as the basic
level, except possibly for memory storage. Certainly, if William
Calvin's hypothesis about columns is correct, then 10^15 ops should be sufficient.
--
sentience@pobox.com Eliezer S. Yudkowsky
http://pobox.com/~sentience/beyond.html
Member, Extropy Institute
Senior Associate, Foresight Institute
This archive was generated by hypermail 2.1.5 : Fri Nov 01 2002 - 15:28:00 MST