[email protected] wrote:
>
> "How can silicon precisely mimic the behavior of flesh?"
>
> This is where functionalism resides.
Functionalism works fine as a general approach, but it has its limitations. The
problem lies in the assumption that there exists some kind of perfect algorithm
which can describe in 100% detail the functioning of a biological substrate.
Form and function are almost synonymous in the realm of biology.
> When you attempt to make a computer play chess, do you want precise mimicry of human behavior?
No, I want it to play good chess.
> People who are interested in robots want to achieve certain functional attributes but avoid certain human limitations. In any case, I suspect that there is nothing magic about protein and neurons.
But it is our "limitations" which make us human, and ironically enough if any
AI project started to display "human limitations" it would be a triumph for AI
research. But just because silicon components are for all practical purposes
precise and error free does not imply that computer intelligence will be in any
way superior or more lucid than our own.
I agree that there is nothing magical about proteins and neurons, but there is
astounding and awesome complexity.
> We do not put feathers on space shuttles.
Only because we tried that and it didn't work.
> I gave up on chess as a child when my older brother would always beat me, but here is a question about your chess network: maybe chess is really more than one game, in particular, maybe you need one sub-neural net that can be adapted according to examples that include 27-32 pieces on the board, another sub-neural net for when 22-27 pieces are on the board, etc. Also, some human players are overly aggressive. Maybe you need one sub-network trained for defensive strategy, another for play against a more conservative opponent. If you had such sub-networks, you would then need a system for deciding which sub-network to use at any given time,
Thanks for the ideas... yes, I have been toying with a lot of ideas myself,
like having an individual ANN for white and for black, or having one for
opening, middle and end-game.
> but I think it is clear that single neural nets like the one you are using are not flexible enough to deal with more than one complex problem at a time.
I am not so sure yet, it may be so... but I am designing Octavius primarily as
a research tool and a lot of work has gone into making the program flexible
enough to allow for a large variety of experimentation. Surprisingly, I have
found hardly any information on the internet specifically about neural
networks and chess. I have come into contact with a few people who I have
exchanged ideas with, including someone from Sao Paulo Federal University [a
pediatric surgeon and professor of statistics I believe] who wrote to me saying
that he was "very impressed" with my results so far.
> One last question......what is your method for ranking the value of individual moves?
As I have said, very little information exists about ANN chess programming, so
I have had to make up my own approach, which is being constantly refined.
Currently, Octavius works on the simple principle that any position reached
during a master game is superior to that of any other possible position which
could have been reached during that game. Octavius essentially outputs a
numeric positional evaluation based on his learning, obviously the highest
positional output is, in his opinion, the best move. I have also incorporated a
traditional alpha-beta tree-search to augment his play.
Cheers,
Luke Pellen
e-mail: [email protected]
ICQ#: 25510475
-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
'I don't want to achieve immortality through my work... I want to achieve
it through not dying.' - Woody Allen
-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
Chaotic Pearl: http://members.tripod.com/~vidagnosis/journal.html
This random quotation was generated by SIGGEN...
SIGGEN is an e-mail signature generator programmed by Luke Pellen
------------------------------------------------------------------------
eGroups.com home: http://www.egroups.com/group/charliehobbs
http://www.egroups.com - Simplifying group communications
charliehobbs
573 of 629 
