From: James Rogers (jamesr@best.com)
Date: Wed Nov 26 2003 - 16:00:27 MST
On Wed, 2003-11-26 at 10:52, Eugen Leitl wrote:
> In my personal view, the screwup is directly traceable to the
> introspection illusion of a single sequential-process observer and not a large
> mess of concurrent processes occuring all in parallel.
A careful distinction has to be made here, and there is a shortage of
appropriate terminology.
"Parallel" is a term generally used to denote partitioning a machine
("algorithm") into multiple machines of the same information theoretic
order. Since this is machine equivalent to "serial" and cheaply
reducible as such for all intents and purposes, I do not view this as an
important theoretical distinction, though it does have pragmatic
ramifications.
Alternatively, "parallel" could be used to denote partitioning a machine
*by* information theoretic order. This second type of parallelism is
not tractably reducible to the first type of parallelism (only in
mathematics-land is it reducible), nor does it lend itself well to the
type of computing hardware that we can build.
No conception of this second kind of parallelism is possible in a
Shannon-esque model of computation, and algorithm spaces that scale well
using this second kind of parallelism can be extremely expensive using
an equivalent Shannon model. By the time reasonable conceptions of this
second kind of parallelism emerged in the post-Kolmogorov theoretical
environment, the fact that decades had already passed and that we still
can't build computational machinery that handles this natively for all
practical purposes has limited interest in it. One can simulate such
machinery on normal silicon but it is very memory expensive and exhibits
a lot of fine-grained parallelism.
Computer science as taught today still uses a model of the machine that
is essentially Shannon-esque, and it is the only model that our
programming languages address well. (In an abstract machine model sense,
all languages are roughly equivalent; they are targeted to the same
implementation concept of a computational machine.)
> It is also interesting to observe how vogues come and go, and
> how evolutionary lemmings deserted the field in droves when the
> field stalled due to hardware limitations and inability to understand
> that one has to evolve evolvable systems first, as brittleware
> doesn't take kindly to have its bits mutated under its feet.
Brittleness is related to the issue I mention above, and is a direct
consequence of using machines with purely axiomatic behaviors.
> > Have you had a chance to write up your views?
I'm working on a reference website that covers a broad range of
theoretical and pragmatic issues related to these topics at large.
Rather than hastily writing the same sloppy explanations for somewhat
complicated theoretical subjects for the nth time, I've been working on
well-written and well-organized body of work that covers these topics in
detail with lots of examples, cross-referencing, and elucidation.
Mostly so that I can point people to a link instead of writing about
it. I will post it to the appropriate places when I put it on the web.
-James Rogers
jamesr@best.com
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:43 MDT