Re: Moravec's estimates?

From: Jimmy Wales (
Date: Mon Apr 09 2001 - 16:18:32 MDT

James Rogers wrote:
> The effect of CPU speed isn't terribly important in terms of pure
> intelligence. Running at half-speed, you'll get half the number of results
> in a given period of time but the intelligence of the results doesn't
> change. OTOH, memory appears to have a scale-invariant logarithmic
> relationship to any reasonable AI implementation, and is therefore arguably
> a more critical resource limitation.

Let me make sure I understand this by paraphrasing.

One potential problem with Moravec's estimates (indeed, any estimates
based on MIPS) is that even if we achieve human-level MIPs in 5-20
years, we've only got one piece of the puzzle. One other major component
is having enough memory (RAM). And so we need to look at Moore's Law
type considerations for RAM/$$$$ as well.

O.k., so has anyone done a Moravec-style analysis to figure out how much "RAM"
would be needed to replicate human performance? What have trends been in RAM
prices? What trends are on the immediate horizon, based on reported plans of
manufacturers plus reported breakthroughs "in the labs"?

I'm still interested, though, in direct examinations of the original question.
Are there good reasons to think that Moravec's estimates are better or worse
than Kurzweil's estimates? Kurzweil's estimates are 200x that of Moravec's, and
since we are getting really close (<5 years) to building a machine that exceeds
Moravec's numbers, we're either *almost there* or still *several doublings* away.

Your point about memory is well taken, because it may imply that we've got
another hardware problem, even more severe, that may take much longer than
20 years to solve.

*              *
*      The Ever Expanding Free Encyclopedia     *

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