From: Philip Goetz (philgoetz@gmail.com)
Date: Thu Apr 27 2006 - 17:00:58 MDT
On 4/27/06, Eliezer S. Yudkowsky <sentience@pobox.com> wrote:
> Ben Goertzel wrote:
> >
> > The concept of "algorithm" is applicable here, but the concepts of
> > "linear" and "exponential" and other orders of complexity need to be
> > handled with much care...
>
> I agree. Human short-term memory is small, and evolution tightly
> complexity-bounded in its algorithms, so why not compare each of 7 items
> to each of 6 other possible items? (Since I'm not sure I buy that we're
> considering all possible sets.) If you had 20 million items in
> short-term memory, you'd spend your processing power differently - you'd
> start asking hard questions about which comparisons were worth carrying
> out.
I would guess that that is in fact what human brains do already - a
very large proportion of our computational power is spent deciding
what to pay attention to. Evolution isn't complexity-bound; it's
energy-bound. Complex algorithms that are energy efficient are
likely.
Given the fact that humans have very large long-term memory stores
relative to what they can hold in "working memory" at one time, the
limitation on working memory is not brain volume. The limitation is
so small that I would guess it is either that
- the method of activating a memory is such that activation of one
memory interferes with activating others - very likely, given what we
know of neural networks
- the computational complexity, as well as the number of
interconnections, needed to deal with the 5 or so chunks in working
memory requires on the same order of neurons as merely storing the
billions(?) of other chunks in long-term memory - also likely
- Phil
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:56 MDT