From: Eliezer S. Yudkowsky (sentience@pobox.com)
Date: Mon Jul 23 2001 - 13:24:40 MDT
Why is it, I wonder, that some people look at exponential growth and see
it continuing indefinitely, while others look at exponential growth and
try to forecast when it breaks down? Personally, I'll admit to using
Moore's Law, but everyone uses Moore's Law. The rest of the time I try to
think in terms of processes I can visualize, and lower bounds. For
example, one often hears folk asserting that Moore's Law will break down
in 2015 or whenever, but we know that currently visible technologies can
take us to 2005. This being the case, I tend to assume that the exact
*timing* of the development of these visible technologies will roughly
follow Moore's Law, both because I don't have the time to track individual
project dates, and because Moore's Law has been so powerful in the past.
Supposing that Moore's Law suddenly breaks down in 2005 and transistors
abruptly stop shrinking, I'm still not nervous, because I know that
computers will suddenly start having more and more symmetric
multiprocessing CPUs, rather than bigger and bigger CPUs, and everyone
will talk about how it was "obvious" that Moore's Law would someday switch
over to parallel power per dollar instead of serial speed per dollar. Or
we could see serial speed continuing to increase at the current rate even
as the number of CPUs also begin to double every couple of years, which
would also break down Moore's Law, albeit from the opposite direction. Or
FPGAs (field-programmable gate arrays) might become common and break down
our yardsticks.
Or things might simply continue entirely as they have.
The real question, from my perspective, is "Will computing power go on
increasing?" and the answer appears to be, with continuing inevitability,
yes. Will this computing power be in a form useful for AI? Yes. Is the
Singularity still on schedule? Yes. Is software, rather than hardware,
still the critical path? Yes. My attention won't really start to focus
on Moore's Law until someone argues that it's
critical-path-to-Singularity... although James Rogers(?) did sharply raise
the profile of "bandwidth" as the hardware characteristic with the slowest
exponent, something that could easily end up affecting AI.
-- -- -- -- --
Eliezer S. Yudkowsky http://intelligence.org/
Research Fellow, Singularity Institute for Artificial Intelligence
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:37 MDT