Re: Hardware Progress: $212/GFlop/s

From: Cliff Stabbert (
Date: Tue May 06 2003 - 18:36:06 MDT

Hi, you wrote:

TD> I realize that there is a lot of subconscious information people have,
TD> process, and use. But a super AI, given access to something like the
TD> Internet, would probably have access to all of the raw information it would
TD> need. If it was taught how to process that information, there'd be little
TD> need to train it further - it could teach itself. That's because the net
TD> contains ridiculous amounts of information (and myriad examples, unintended
TD> and otherwise) about things like social relationships, interactions, and the
TD> like. An SAI should be able to do a far better job than a research team
TD> could in learning how to deal with people if it was just given a basic
TD> framework that it could color in itself.

TD> Or maybe I am just naive. I just like reading this list, I haven't much
TD> education as far as cognitive science goes :)

IMO you are being naive. Look at it this way: would an intelligent
person, raised in isolation to the point where they could read, but
without social interaction -- would they be able to learn about
"things like social relationships, interactions, and the like" purely
from the internet?

As much content as is out there on the internet, to look at it as a
repository of all human knowledge is waaaaaaay premature. There are
numerous ways in which this is the case, a few of the most basic ones:
- it's content, not "information", until it's been processed
- there are vast areas of knowledge not covered and that cannot
  be easily covered with current technology -- lots of procedural
  knowledge falls into this category

As a larger point, I doubt Dani is married to the 2.5 year figure.
But it's useful to come up with some sort of guesstimates and stick to
them, so that trends can be tracked.


This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:42 MDT