Chess using Neural Networks/Fuzzy Logic.
Kumar Chellapilla
Hi,
I have been reading this newsgroup for a couple of weeks and would like to
know if any Neural Network and/or Fuzzy Logic based approaches have been
tried for computer chess. Any good references or landmark publications
would be appreciated. Any thoughts on their success and future would also
be appreciated.
Any information will be appreciated,
Wishing you a wonderful day,
Kumar
----------------------------------------------------------------------------
KUMAR H. CHELLAPILLA,Grad Asst - EE Dept,Villanova Univ.,Villanova, PA
19085.
Res: 377 Avon Road, Sugartown Mews, #D-107,Devon, PA 19333.
tel: of: (610) 519 7222 res: (610) 688-7006 FAX: (610) 519-4436
email: kum...@ece.vill.edu WWW : http://www.ece.vill.edu/user/kumarc
Utku Ozcan
Kumar Chellapilla wrote:
>
> Hi,
> I have been reading this newsgroup for a couple of weeks and would like to
> know if any Neural Network and/or Fuzzy Logic based approaches have been
> tried for computer chess. Any good references or landmark publications
> would be appreciated. Any thoughts on their success and future would also
> be appreciated.
Sir,
I am an MSc student on microelectronics. In my undergraduate thesis, while looking
for articles in IEEE JSSC (Journal of Solid State Circuits) for BiCMOS SRAM,
I found a research on a chess chip. *I think* it is NN. The article belongs to
a PhD or MSc student (I don't know his name, unfortunately) but
October 1987
issue includes that article. The article also includes many aspects about that.
By the way, I am also dealing with that area.
Is your area about VLSI?
Jay Scott
In article <01bc1929$2ea74940$513f...@dsp.ee.vill.edu>, "Kumar Chellapilla" (kum...@ece.vill.edu) wrote:
> have been reading this newsgroup for a couple of weeks and would like to
>know if any Neural Network and/or Fuzzy Logic based approaches have been
>tried for computer chess.
I don't know of any fuzzy chess programs. Neural net chess programs
exist. They're research programs, quite weak. See SAL and NeuroChess
on my Machine Learning in Games web site, address below.
The simple idea of replacing a traditional evaluator with a trained
backprop net is not likely to produce a high-performance program. Hand-
written evaluators are faster, so even if you can train up a net that's
more accurate, you'll likely pay too much for it. But there are many
other ideas. See, for example, "Efficient neural net alpha-beta-
evaluators" by Alois Heinz, under "other papers" on the web site.
Jay Scott <j...@forum.swarthmore.edu>
Machine Learning in Games:
http://forum.swarthmore.edu/~jay/learn-game/index.html
LATENDRESSE Simon
Kumar Chellapilla" <kum...@ece.vill.edu> writes:
I have been reading this newsgroup for a couple of weeks and would like to
know if any Neural Network and/or Fuzzy Logic based approaches have been
tried for computer chess. Any good references or landmark publications
would be appreciated. Any thoughts on their success and future would also
be appreciated.
Ok, I never tought I'd say this, especially in this group, but... me
too!
Simon
Utku Ozcan
Hi!
I wanted to put more information about my reply.
IEEE-JSSC (Journal of Solid State Circuits)
October 1987 (Special Issue for Memory Circuits)
The article I comment on belongs to an MSc or PhD student from
Carnegie Mellon University.
For Fuzzy Logic there are many sites available, you easily find
many sites by using standard search engines.
When dealing with my diploma thesis, I just only could find this
article. Sir, if you have been able to find other documents about
Fuzz Logic or Neural Networks implementation chess VLSI chips,
would you please inform me? Also about Internet addresses.
I wish all the best with your works.
Hoping to hear from you.
Utku OZCAN
MSc Student of Technical University of Istanbul
Robert Hyatt
Utku Ozcan (oz...@netas.com.tr) wrote:
: Hi!
: Utku OZCAN
I think using them *in* a chess program while searching, is likely hopeless
because speed is important. However I once had a student here that thought
that using a NN to "train" the evaluation might be worthwhile... that is,
feed in lots of positions and let it learn what the various eval term
weights ought to be, then transcribe those values back into the actual
C evaluation code. Would be interesting to see how that might work, because
tuning is certainly an ad hoc thing at present for most programs...
Jay Scott
In article <5e9ran$h...@juniper.cis.uab.edu>, Robert Hyatt (hy...@crafty.cis.uab.edu) wrote:
>Utku Ozcan (oz...@netas.com.tr) wrote:
>: When dealing with my diploma thesis, I just only could find this
>: article. Sir, if you have been able to find other documents about
>: Fuzz Logic or Neural Networks implementation chess VLSI chips,
>: would you please inform me? Also about Internet addresses.
>I think using them *in* a chess program while searching, is likely hopeless
>because speed is important. However I once had a student here that thought
>that using a NN to "train" the evaluation might be worthwhile... that is,
>feed in lots of positions and let it learn what the various eval term
>weights ought to be, then transcribe those values back into the actual
>C evaluation code. Would be interesting to see how that might work, because
>tuning is certainly an ad hoc thing at present for most programs...
A garden variety neural net is certainly too slow to be a good
evaluator. Amazing accuracy might compensate for the slow speed,
but I don't think they have amazing accuracy either.
I want to re-mention a neural net idea that's not widely known.
Hold on, this takes a few steps.
- Train up a decision tree as an evaluation function. An
algorithm similar to ID3 will do.
- For a chess program, the evaluator doesn't need to return
values outside the alpha-beta window. You can pass alpha and beta
to the decision tree evaluator, and have it stop early if the
score falls outside the bounds. This is a generalization of the
"lazy evaluation" idea used in many chess programs. It treats the
decision tree as an extension of the search tree, and actually
performs normal alpha-beta cutoffs inside the decision tree,
while calculating the evaluation, thus saving time.
- A decision tree may be a poor evaluator because it has "sharp
edges". As Berliner has pointed out, smoothness in the evaluator
prevents some weird artifacts that cause bad play. The decision tree
can be smoothed by converting it to an equivalent neural network
and training the network on more data. This type of neural network
retains the ability to make use of alpha-beta cutoffs in the
evaluation of the network! It may well be fast enough for a
performance program.
All the details have been worked out by Christoph Hense and
Alois Heinz. See some of the papers mentioned on this web page:
http://www.informatik.uni-freiburg.de/~heinz/
graham_douglass
In article <5e9ran$h...@juniper.cis.uab.edu>, hy...@crafty.cis.uab.edu says...
{snip}
>: When dealing with my diploma thesis, I just only could find this
>: article. Sir, if you have been able to find other documents about
>: Fuzz Logic or Neural Networks implementation chess VLSI chips,
>: would you please inform me? Also about Internet addresses.
>
>: Hoping to hear from you.
>
>: Utku OZCAN
>
>: MSc Student of Technical University of Istanbul
>
>: http://www.ehb.itu.edu.tr/
>
>because speed is important. However I once had a student here that thought
>that using a NN to "train" the evaluation might be worthwhile... that is,
>feed in lots of positions and let it learn what the various eval term
>weights ought to be, then transcribe those values back into the actual
>C evaluation code. Would be interesting to see how that might work, because
>tuning is certainly an ad hoc thing at present for most programs...
I have to register my disagreement with Bob about this.
Does he really believe that "training evaluation weights" is it? He already
admitted (in a posting a couple of months ago) that appropriate evaluation
weights should be different in different positions.
About a year ago, I remember CW writing that, while programmers create the
illusion that they have created an elegant game tree searching machine, the
reality is that most programs are full of "fixes" to make them play properly in
particular types of position.
Unless there is a simple underlying mathematics to chess that we have all missed,
I would say that the "fixes" are the real thing, and the game tree is the "fix"
for weaknesses in the knowledge.
The reason why I am excited that neural networks have entered the discussions is
that they may be able to overcome the problems of getting a good enough level
of knowledge into a computer to be able to play grandmaster level chess.
Merely tinkering with the evaluation weights in a game tree searching machine,
no matter how well it is done, will just make the computer play better in some
positions, and worse in others. Maybe the neural network can do the entire job -
and use the game tree just to give the computer an extra competitive advantage
over the old humans.
In the future, the efforts to achieve grandmaster play by game tree generation
may be seen as being like trying to get to the moon by building a super tall
ladder - doomed to failure, but the best approach available with the technology
of the day.
Graham
Robert Hyatt
GrahamDouglass wrote:
: In article <5e9ran$h...@juniper.cis.uab.edu>, hy...@crafty.cis.uab.edu says...
: {snip}
: >: When dealing with my diploma thesis, I just only could find this
: >: article. Sir, if you have been able to find other documents about
: >: Fuzz Logic or Neural Networks implementation chess VLSI chips,
: >: would you please inform me? Also about Internet addresses.
: >
: >: I wish all the best with your works.
: >: Hoping to hear from you.
: >
: >: Utku OZCAN
: >
: >: MSc Student of Technical University of Istanbul
: >
: >: http://www.ehb.itu.edu.tr/
: >
: >I think using them *in* a chess program while searching, is likely hopeless
: >because speed is important. However I once had a student here that thought
: >that using a NN to "train" the evaluation might be worthwhile... that is,
: >feed in lots of positions and let it learn what the various eval term
: >weights ought to be, then transcribe those values back into the actual
: >C evaluation code. Would be interesting to see how that might work, because
: >tuning is certainly an ad hoc thing at present for most programs...
: I have to register my disagreement with Bob about this.
: Does he really believe that "training evaluation weights" is it? He already
: admitted (in a posting a couple of months ago) that appropriate evaluation
: weights should be different in different positions.
No I don't agree that training is "it" for the very reason you state. However,
there are many weights that remain constant throughout the course of the game,
and those might work... and you could train middlegame weights in middlegame
positions, and endgame weights in endgame positions. Whether this is good or
not is unknown because it was never tried here... perhaps one day someone will
voice an interest again and I'll snag 'im... :)
: About a year ago, I remember CW writing that, while programmers create the
: illusion that they have created an elegant game tree searching machine, the
: reality is that most programs are full of "fixes" to make them play properly in
: particular types of position.
: Unless there is a simple underlying mathematics to chess that we have all missed,
: I would say that the "fixes" are the real thing, and the game tree is the "fix"
: for weaknesses in the knowledge.
good analogy. That's been the "Bob" approach for a long time. Evaluate what
I can, search what I can, and ignore the rest. :)
: The reason why I am excited that neural networks have entered the discussions is
: that they may be able to overcome the problems of getting a good enough level
: of knowledge into a computer to be able to play grandmaster level chess.
: Merely tinkering with the evaluation weights in a game tree searching machine,
: no matter how well it is done, will just make the computer play better in some
: positions, and worse in others. Maybe the neural network can do the entire job -
: and use the game tree just to give the computer an extra competitive advantage
: over the old humans.
: In the future, the efforts to achieve grandmaster play by game tree generation
: may be seen as being like trying to get to the moon by building a super tall
: ladder - doomed to failure, but the best approach available with the technology
: of the day.
It might, but the problems are difficult. A group here was working with the
army missle command to try to use NN's to do target tracking. IE, take two
successive radar "pictures" and if you know in picture A you are looking at
a Friendly F15 and two enemy Migs... can you figure out in image B what is
what? It never really worked, but it was interesting...
: Graham
Jean-Peter Fendrich
Graham, Douglass wrote:
>
> In the future, the efforts to achieve grandmaster play by game tree generation
> may be seen as being like trying to get to the moon by building a super tall
> ladder - doomed to failure, but the best approach available with the technology
> of the day.
>
Well, I don't say that NN can't contribute to better chess programs but
the way you describe it will probably not happen during our life time
anyway.
I have been working with Neural Nets and Genetic Algorithms for about
4 years, more or less full time in my business. I can't see how any of
theese techniques in any way can substitute the search algorithms in a
chess program. At least not by what was state of the art 3 years ago.
I think that NN might be useful to guide the searching itself like
guiding the decision of doing extensions or not in runtime. There are
lot's of parameters that normally are set to fixed values during the
whole game or at least while the tree search is running. Theese
parameters could be more adaptable during the game or during the search
with the help of NN and/or GA.
Finally it might be possible to enhance the evaluation itself.
One big problem with the evaluation is the huge magnitude of test
positions that is probably needed to cover at least a small set of all
possible and sometimes very distinct types of knowledge we are talking
about.
IMHO the best evaluation function is not necessarily the one which
in average produces the best value for a bunch of positions. The
best evaluation function is the one which is best guiding the search
(assuming search is used...). A lot of weird positions should all be
regarded as bad. It is not enough to solve hard positional issues.
This makes the training or learning a real challenge.
Another challenge is of course the phase of defining the NN or GA
with input nodes, parameters etc. But this is the fun part ...
Good luck - you will need it ... :-)
--
J-P Fendrich
brucemo
Graham, Douglass wrote:
> Merely tinkering with the evaluation weights in a game tree searching machine,
> no matter how well it is done, will just make the computer play better in some
> positions, and worse in others. Maybe the neural network can do the entire job -
> and use the game tree just to give the computer an extra competitive advantage
> over the old humans.
>
> In the future, the efforts to achieve grandmaster play by game tree generation
> may be seen as being like trying to get to the moon by building a super tall
> ladder - doomed to failure, but the best approach available with the technology
> of the day.
Occasionally you hear someone suggest that you can solve chess by applying a tree
search all the way to mate or forced draw. This is an impractical line of research
because the time it would take to run this program is for all intents infinite.
Whenever someone says "neural net", there is a flurry of activity here. I am
wondering, though if it's the same sort of thing as when someone suggests tree
searching to the end of the game.
Will it be practical, within foreseeable time limits, to make a chess program that
plays well with a neural net? Is there research being actively pursued in this
area, or is this similar to the argument someone presented here last year, that a
half-way decent mathematician could produce something that would destroy the game
tree approach if they were to bother to spend a little time on it (but that none
has bothered)?
Your ladder comments may be true, but progress made by the game-tree crowd has been
obvious within the last ten years. This approach could eventually beat Kasparov.
Is any other research avenue proving to be as promising?
bruce
Jay Scott
In article <330C18...@nwlink.com>, brucemo (bru...@nwlink.com) wrote:
>Occasionally you hear someone suggest that you can solve chess by applying a tree
>search all the way to mate or forced draw. This is an impractical line of research
>because the time it would take to run this program is for all intents infinite.
>
>Whenever someone says "neural net", there is a flurry of activity here. I am
>wondering, though if it's the same sort of thing as when someone suggests tree
>searching to the end of the game.
>
>Will it be practical, within foreseeable time limits, to make a chess program that
>plays well with a neural net? Is there research being actively pursued in this
>area, or is this similar to the argument someone presented here last year, that a
>half-way decent mathematician could produce something that would destroy the game
>tree approach if they were to bother to spend a little time on it (but that none
>has bothered)?
There's a high level of crackpot activity in neural networks, yes.
Fortunately there's also a high level of serious activity. For
chess specifically, the situation is a little complicated. The
most popular forms of neural network are not suitable for direct
use in a performance program--too slow. Specialized networks have
been invented which may or may not be useful--the question is open.
Future networks may be invented which are wonderful, but I can't
guess whether that will happen. The field is wide open, so it's
possible, but it also seems difficult to do.
>Your ladder comments may be true, but progress made by the game-tree crowd has been
>obvious within the last ten years. This approach could eventually beat Kasparov.
>
>Is any other research avenue proving to be as promising?
Rational search is a more promising line, in my view. If you analyze
depth-limited alpha-beta at a fundamental level, it turns out to be
based on simplifying assumptions which are not true, and which may
not be the best simplifying assumptions to make in practice. For
example, the minimax idea that the value of your move is the value
of your opponent's best reply is an approximation which does not take
into account your uncertainty about which of your opponent's moves
really is the best one. Of course, this approximation is what allows
the alpha-beta algorithm to work, so it's a very practical one.
Rational search is about designing search algorithms which work
"rationally", which means, they work according to mathematical
decision theory. No rational search algorithm has been shown to
perform better than alpha-beta in practice, but I suspect it's only
a matter of time and hard thinking.
Info about these things can be found on my web site, Machine
Learning in Games.
Jay Scott
In article <5eic89$3...@larch.cc.swarthmore.edu>, Jay Scott
Admittedly, in the short run the current tradition of alpha-beta
search with various pruning and extensions and human-tuned
evaluators will continue to produce the best results. From my
artificial intelligence point of view, that's more like "development"
than "research", so I didn't count it. :-)
It will be difficult to overturn such a dominant paradigm, and
for good reason. It's dominant because it works. And because it's
dominant it has been highly developed, which makes it hard to
surpass: even if you have a better idea, you have to work like hell
to catch up with the sophisticated optimizations that have been
applied to the old idea.
It's rather like replacing CRT's with flat panel displays. Someday
the flat panels will win, but in the meantime CRT's are dominant,
and dominance has a self-perpetuating aspect. And we can't guess
what flat panel technology will be the big winner, or even if there
will be only one winner.
Anders Thulin
In article <5eei0d$c...@lana.zippo.com>, <Graham Douglass> wrote:
>Does he really believe that "training evaluation weights" is it? He already
>admitted (in a posting a couple of months ago) that appropriate evaluation
>weights should be different in different positions.
If that difference could be recognized and encoded in a feature
vector, is there any reason why a NN wouldn't be able to use it as
input for training?
The problem appears rather to be how to come up with the evaluation
scores necessary for input to the training process.
--
Anders Thulin Anders...@lejonet.se 013 - 23 55 32
Telia Research AB, Teknikringen 2B, S-583 30 Linkoping, Sweden
graham_douglass
In article <330C18...@nwlink.com>, brucemo says...
>
>Graham, Douglass wrote:
{snip}
>> In the future, the efforts to achieve grandmaster play by game tree generation
>> may be seen as being like trying to get to the moon by building a super tall
>> ladder - doomed to failure, but the best approach available with the technology
>> of the day.
>
>Occasionally you hear someone suggest that you can solve chess by applying a tree
>search all the way to mate or forced draw. This is an impractical line of research
>because the time it would take to run this program is for all intents infinite.
>
>Whenever someone says "neural net", there is a flurry of activity here. I am
>wondering, though if it's the same sort of thing as when someone suggests tree
>searching to the end of the game.
>
>Will it be practical, within foreseeable time limits, to make a chess program that
>plays well with a neural net? Is there research being actively pursued in this
This is being discussed in another thread. I do not believe that anyone is
trying to make a neural net to play the whole game at the moment, but I believe
that is practical.
To summarize what has been said in another thread, we would need to create an
NN capable of recognising 50,000 patterns with fuzzy boundaries that typically
arise during chess games. People who know more about NNs than me have postulated
that if this really is the requirement, it would be possible to do it.
>area, or is this similar to the argument someone presented here last year, that a
>half-way decent mathematician could produce something that would destroy the game
>tree approach if they were to bother to spend a little time on it (but that none
>has bothered)?
Guilty as charged. I originally put it as a question, but no-one replied to it.
To get an answer, I reposted the same thread in controversial terms, which did
illicit much more information.
This time, there are significant differences:
1. I did not start the thread
2. I did not say anything controversial
3. The thread developed with people adding to each other's thoughts
4. The thoughts in the thread are new - wheras the mathematical solution thread
was a search to see what had been done over the last 100 years
5. The quality of discussion has been high (in my opinion)
>
>Your ladder comments may be true, but progress made by the game-tree crowd has been
>obvious within the last ten years. This approach could eventually beat Kasparov.
Good argument. I know this is a hotly disputed area, and I hope that no-one
chooses to discuss this point in this thread (because I'd much prefer to stick
to NNs, and there are loads of other threads to discuss this in), but over the
last few years I have switched sides, and I no longer believe that depth of
search alone will be enough to beat the top grand masters.
More significantly is the progress being made in other fields of AI. There have
been many occasions on which AI (artificial intelligence) has failed to live up
to expectations, and consequently, it has acquired a bad name. However, in
recent years, several AI technologies have been quietly making excellent
progress. The tools are getting better at quite a rate. I personally believe
that it is AI, much more than the Internet, which is going to profoundly change
our lives over the next 20 years.
And the computer technology is improving. I've seen forecasts that show that
in 10 years, computers, for the same price, will be 1000 times more powerful
than they are today. There is lots of supporting evidence that this will happen.
I think people are going to be surprised at the things that these machines will
be able to do. They will steadily climb the scale from unskilled work to
maximum skill work.
On Wednesday night, on Tomorrow's World on BBC1, they showed a system for
automatically analysing people's CVs (resumes to Americans) for job interview
shortlisting. The system was pitted against a recruitment consultant. The system
picked the same three candidates, and put them in the same order, as the
consultant. This is a package you can buy off the shelf today.
It seems short sighted to think that they will not soon be able to play top
level chess.
>
>Is any other research avenue proving to be as promising?
(see above)
>
>bruce
Thanks for providing some opposition!
Graham
Komputer Korner
Jay Scott wrote:
>
snipped
> It's rather like replacing CRT's with flat panel displays. Someday
> the flat panels will win, but in the meantime CRT's are dominant,
> and dominance has a self-perpetuating aspect. And we can't guess
> what flat panel technology will be the big winner, or even if there
> will be only one winner.
>
> Jay Scott <j...@forum.swarthmore.edu>
>
> Machine Learning in Games:
> http://forum.swarthmore.edu/~jay/learn-game/index.html
This whole concept requires an underlying assumption that chess can
be worked out to a mathematical formula or at least some
algorithms that will enable a program to find the optimal moves.
This is flawed for the following reason. Chess is not part of the
natural world and has no connection to it. The rules were developed
over a long period of time and have no relation to a logical whole.
AI people constantly try to connect the study of numbers with the study
of games. Game theory is good for games that depend on probability
theory but for games like chess that are search problems, I wouldn't
hold out much hope. If AI can't solve the travelling salesman problem
how can it hope to solve a 2 person search tree problem?
--
Komputer Korner
The inkompetent komputer.
graham_douglass
In article <33112A...@netcom.ca>, Komputer says...
{snip}
>
>This whole concept requires an underlying assumption that chess can
>be worked out to a mathematical formula or at least some
>algorithms that will enable a program to find the optimal moves.
The idea of turning to Neural Networks is to avoid the need for mathematical
formulae or rigid knowledge with rigid boundaries!
The benefits we seek are fuzzy knowledge with fuzzy boundaries that does not
have to be supplied by a person.
>This is flawed for the following reason. Chess is not part of the
>natural world and has no connection to it. The rules were developed
>over a long period of time and have no relation to a logical whole.
The human opposition is also faced with these difficulties!
>AI people constantly try to connect the study of numbers with the study
>of games. Game theory is good for games that depend on probability
>theory but for games like chess that are search problems, I wouldn't
>hold out much hope. If AI can't solve the travelling salesman problem
>how can it hope to solve a 2 person search tree problem?
Ah - but how do the ai travelling salesman solutions compare with the human
opposition?
To be classified as ai, it is not necessary to solve a problem perfectly - you
just have to do it better than the human opposition!
>
>--
>Komputer Korner
>
>The inkompetent komputer.
Ahhhh - there there... :-)
Jay Scott
In article <33112A...@netcom.ca>, Komputer Korner (kor...@netcom.ca) wrote:
>Jay Scott wrote:
>>
>snipped
>> It's rather like replacing CRT's with flat panel displays. Someday
>> the flat panels will win, but in the meantime CRT's are dominant,
>> and dominance has a self-perpetuating aspect. And we can't guess
>> what flat panel technology will be the big winner, or even if there
>> will be only one winner.
>>
>> Jay Scott
>>
>be worked out to a mathematical formula or at least some
>algorithms that will enable a program to find the optimal moves.
>natural world and has no connection to it. The rules were developed
>over a long period of time and have no relation to a logical whole.
>of games. Game theory is good for games that depend on probability
>theory but for games like chess that are search problems, I wouldn't
>hold out much hope. If AI can't solve the travelling salesman problem
>how can it hope to solve a 2 person search tree problem?
Who's talking about solving? Humans haven't solved chess, but that
doesn't stop us from playing well. AI hasn't solved chess, but we're
not asking it to. Optimal play is great, but great play needn't
be optimal, and who wouldn't be happy with that?
The bit about chess not being part of the natural world seems to
be from outer space. You wouldn't perhaps be a space alien
in disguise, would you? :-) Maybe you mean that the messy rules of
chess don't lend themselves to analysis. I'll just point out that
computers, unlike humans, are built from the transistors up to be
good at dealing with arbitrary rules.
brucemo
Graham, Douglass wrote:
> Thanks for providing some opposition!
>
> Graham
I'm happy to discuss this. I'd like to learn more, is there a good
book on neural nets?
If the technique were in line with my goals (strong player, near
term, good chance of success), I would adopt it. It doesn't seem
like it is, so I don't use it.
But that's not to say it's the wrong technique for someone else, nor
does this mean that I can't learn anything from it.
bruce
Jean-Peter Fendrich
There are more books in the subject than you will ever read...
A good starting point is
http://www.emsl.pnl.gov:2080/docs/cie/neural/neural.homepage.html
There you will find some useful descriptions and links.
--
J-P Fendrich
Tom C. Kerrigan
Komputer Korner (kor...@netcom.ca) wrote:
> This whole concept requires an underlying assumption that chess can
> be worked out to a mathematical formula or at least some
> algorithms that will enable a program to find the optimal moves.
> This is flawed for the following reason. Chess is not part of the
> natural world and has no connection to it. The rules were developed
My God, this is even worse than classic Star Trek logic...
I hope you were drinking hard right before you posted this.
Tic Tac Toe is something you don't see a lot in the natural world, but it
doesn't exactly evade solution.
Chess is Tic Tac Toe, only bigger.
Cheers,
Tom
Komputer Korner
And how did you solve TIC TAC TOE? Trial and error!!!!!
Games are artificially constructed. Subjects in the natural world are
all connected to each other so it is easier to construct the
solutions to the problems. Chess is just a game with artificial rules
and will only succumb to deep search.
brucemo
Komputer Korner wrote:
> Tom C. Kerrigan wrote:
> > My God, this is even worse than classic Star Trek logic...
> >
> > I hope you were drinking hard right before you posted this.
> >
> > Tic Tac Toe is something you don't see a lot in the natural world, but it
> > doesn't exactly evade solution.
> >
> > Chess is Tic Tac Toe, only bigger.
> And how did you solve TIC TAC TOE? Trial and error!!!!!
> Games are artificially constructed. Subjects in the natural world are
> all connected to each other so it is easier to construct the
> solutions to the problems. Chess is just a game with artificial rules
> and will only succumb to deep search.
I predict a gigantic semantic argument will ensue, the significance of which will
be zero. Please let my prediction be wrong.
bruce
Chris Whittington
--
http://www.demon.co.uk/oxford-soft
brucemo <bru...@nwlink.com> wrote in article <331A7A...@nwlink.com>...
No, no. Deep significance.
That is if you give 50 monkeys 50 typewriters and leave them for 50 years,
they will then produce the Complete Works of Shakespeare.
KK is in the process of disproving this. Tom is helping him..
Chris Whittington
>
> bruce
>
Komputer Korner
> > Games are artificially constructed. Subjects in the natural world are
> > all connected to each other so it is easier to construct the
> > solutions to the problems. Chess is just a game with artificial rules
> > and will only succumb to deep search.
>
> that I can't prove the solution without pencil, paper, and some coffee.
>
> Tell me that solving Tic Tac Toe (artificial rules) is harder than, say,
> mapping the human chromosomes (subject in the natural world).
>
> I wonder why Kirk never asked Spock to say things like this when Spock
> came up with such inane logic...
>
> Cheers,
> Tom
If you have a mathematical solution to TIC TAC TOE, I would like to
see it.
Komputer Korner
Chris Whittington wrote:
>
> --
> No, no. Deep significance.
>
> That is if you give 50 monkeys 50 typewriters and leave them for 50 years,
> they will then produce the Complete Works of Shakespeare.
>
> KK is in the process of disproving this. Tom is helping him..
>
> Chris Whittington
>
> >
> > bruce
> >
That was pretty subtle but isn't funny once it becomes that subtle.
Tom C. Kerrigan
Chris Whittington
--
http://www.demon.co.uk/oxford-soft
Komputer Korner <kor...@netcom.ca> wrote in article
<331B66...@netcom.ca>...
> Chris Whittington wrote:
> >
> > --
>
> > No, no. Deep significance.
> >
> > That is if you give 50 monkeys 50 typewriters and leave them for 50
years,
> > they will then produce the Complete Works of Shakespeare.
> >
> > KK is in the process of disproving this. Tom is helping him..
> >
> > Chris Whittington
> >
> > >
> > > bruce
> > >
>
> That was pretty subtle but isn't funny once it becomes that subtle.
Well I have to disagree.
I gave it the KK Kold Kedal for funny
and
the KK Kold Kedal for subtle :)
Can I have a KK post of the week prize now please ?
Chris Whittington
Tom C. Kerrigan
Komputer Korner (kor...@netcom.ca) wrote:
> Chris Whittington wrote:
> > That is if you give 50 monkeys 50 typewriters and leave them for 50 years,
> > they will then produce the Complete Works of Shakespeare.
> > KK is in the process of disproving this. Tom is helping him..
I thought it was straightforward and laughed out loud. Thanks, Chris! :)
Cheers,
Tom
Tom C. Kerrigan
Komputer Korner (kor...@netcom.ca) wrote:
> If you have a mathematical solution to TIC TAC TOE, I would like to
> see it.
Hum, this might be interesting. Maybe I'll work on it during my ski trip
(other stuff to do now). I'll report back, and then you can show me your
mathematical solution to determining roles of chromosomes.
Cheers,
Tom
Michael F Jorgensen
Komputer Korner wrote:
> If you have a mathematical solution to TIC TAC TOE, I would like to
> see it.
Just what do you mean by a 'mathematical solution'?
And just exactly what is wrong with 'trial-and-error'?
And where does TTT differ from chess, other than complexity?
-Michael.
--
WWW: http://fermi.phys.ualberta.ca/~mjorg/
Dept. of Phys., U of Alberta | TODAY is the
Edmonton, AB T6G 2J1, Canada | BEST DAY
e-mail: mj...@phys.ualberta.ca | of the WEEK
Komputer Korner
Exactly my point.