From: Ben Goertzel (ben@webmind.com)
Date: Wed Jul 18 2001 - 04:49:41 MDT
>
> True, but cheaper ram doesn't have anything to do with the 4GB limit on
> cost effective Pentium based PCs. The vast majority of systems (mother
> board / BIOS combos) simply can't handle more the 4GB. Some of the very
> high end systems might, but they are much, much more expensive
> than systems
> that can handle 2-4GB. So even if RAM was significantly cheaper still it
> won't be cost effective to go beyond 4GB until new BIOSs and
> mother boards
> which can handle this become generally available.
Correct. And all this is very frustrating, tactically.
The situation is subtle. It's clearly true that by far the biggest obstacle
standing between us and a real AI is not hardware, but the lack of a
demonstrably correct detailed design. (I think the Webmind design is
correct, but I don't delude myself that this has been convincingly
demonstrated yet.)
And yet, it's also true that to create a thinking machine seems to require
much more hardware than to actually run one. The reason is that one has to
do a lot of experimentation, a lot of testing, a lot of parameter
optimization, etc. So that in order to take one's high-level AI design and
turn it into a detailed design, does seem to require a fair bit of computer
power -- since we lack the science to *deduce* a detailed real AI design and
hence have to use some trial and error experimentation in the design
process.
The bottom line is then that hardware limitations slow us down a bit. And
it's a constant struggle between spending time overcoming current
limitations so one can experiment with one's ideas effectively, and spending
time on pure design and analysis and waiting for the hardware to come about.
Personally, I spent about 8 years doing the latter, and started doing the
former in 1997 when it seemed to me, optimist that I am, that the hardware
was *almost there*...
ben
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:36 MDT