From: Robert J. Bradbury (bradbury@www.aeiveos.com)
Date: Thu Sep 02 1999 - 04:25:08 MDT
Eliezer S. Yudkowsky" <sentience@pobox.com> wrote:
>
> Deep Blue involved some innovations with respect to the shape of the
> search tree - the ability to project some single lines 60 moves into the
> future, for example. But in the end it was simply brute force that
> crushed Kasparov. Deep Blue's feat cannot be compared to cognition; it
> was an autonomic process that happened to play chess.
Eliezer, if what you say is accurate, then I don't believe that much
of what I do, such as selecting a line of code in a program,
driving a car or even talking would not be "cognition" either.
[And I'm deeply saddened by that... :-;]
When I look I pick an instruction, I select from the available
list of instructions that do the job that needs to be done
(i.e. I select an add instruction when what I want to do is
produce the sum of two numbers, I don't select the subtract
instruction). When I want the car to make a right turn, I select
the alternative that is not making a left turn or going straight
or going backwards. When I speak, if I've had my coffee and
have not set the "lets put my foot in my mouth today" bias
switch to "on", then I select the word sequence that is designed (from
accululated experience presumably) to create in the communications
recipient some duplicate copy of the thought or concept (which
is itself a collection of words) that I am trying to convey.
If a huge fraction of what I do is *not* a brute force pruning
of a search tree, then I don't know what it is. So in that
respect Deep Blue and I are quite similar.
If by "cognition", you mean the awareness of sensory input,
classifying it, perhaps using it to amend the search tree,
then I would agree that Deep Blue did not do much of that
if any. Its authors were the individuals who setup the
data entry, search, pruning & selection algorithms. The
authors served in some respects as the cognition subroutines
for Deep Blue. [Clever little computer isn't it...]
The nature of much of what Deep Blue did would have required
little cognition. For the purpose for which it was designed,
the cognition part can be done by the position evaluation software.
Essentially the "Where am I?" or "How am I doing?" parts of
its code. Now if you had Deep Blue A & Deep Blue B
playing each other, and in each you created a feedback loop
between the assesment of the board positions and the parameters
involving search depth, pruning, and selection criteria
then I would argue that you have something very close to "AI".
[I don't need to wait for the singularity to feel the tidal
wave, I suspect I can produce them from where I sit now...]
Robert
This archive was generated by hypermail 2.1.5 : Fri Nov 01 2002 - 15:04:59 MST