RE: Moravec's estimates?

From: James Rogers (
Date: Mon Apr 09 2001 - 16:04:09 MDT

At 05:19 PM 4/9/2001 -0400, Patrick McCuller wrote:

> I disagree. You're quantifying 'intelligence' in a way that I
> don't think is
>reasonable given the single axis of 'computation speed'.

Computation speed simply isn't a terribly relevant axis with respect to
intelligence, other than getting a good answer in a timely manner. A
wicked fast calculator is not even remotely intelligent. Good DSPs can run
circles around the best CPUs in terms of raw computational ability, but you
wouldn't want to use them for AI (or a lot of other applications). The
human brain is very slow computationally, but benefits from a large memory
(and super-wide memory bus) and efficient data organization where
intelligence is concerned. Memory matters.

>Chess programs make a
>MUCH more intelligent decision given more processing power.

This is an oversimplified and not particularly relevant model. For one,
this is an extraordinarily narrow problem space, requiring lots of
computation on a relatively small data set. Also, I would guess that
exponential increases in processing power yield a roughly linear increase
in performance for this game (you can see this in how silicon scales with
human ability at the game even though all humans are operating at roughly
the same processing speed).

It only seems to make much more intelligent decisions as its ability passes
through the distribution of human ability; minor improvements in the
performance metrics will yield apparently substantial improvements relative
to the human population, but not in absolute terms. In the same way, very
intelligent people are only "very intelligent" relative to the rest of the
population; from the viewpoint of something that is substantially more
intelligent than your average human, all humans appear to have about the
same cognitive ability.

-James Rogers

This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:36 MDT