Deep Blue's FIDE rating
Geoffrey P. Wyatt
Kasparov match, with the score tied at 4 all, IM and GM commentators were
speculating that DB is 2750 FIDE. Yet after just two additional games,
featuring some awful positional howlers on the part of DB, those same
commentators were talking 2550 FIDE, tops. Surely this demonstrates the
potential danger in trying to estimate the "true" strength of a computer
on the basis of a very short match.
Imagine if they had agreed on only 4 games this time instead of 6?!!
We would all think that the IBM team had taken an *amazing* leap forward
and that DB is now equal to the best human chessplayer on the planet.
Instead it is abundantly clear that DB can be crushed positionally, and
most certainly by other strong players besides Kasparov.
Hsu, Tan and company deserve credit but have many years of work ahead of
them if they hope to compete with a Kasparov who now understands very well
how to exploit the gaping silicon holes.
Comments?
--
geo...@teleport.COM Public Access User --- Not affiliated with TECHbooks
Public Access UNIX and Internet at (503) 220-1016 (2400-14400, N81)
Thomas Kettle
geo...@teleport.com (Geoffrey P. Wyatt) wrote:
>There is a very interesting irony in that after four games of the DB-
>Kasparov match, with the score tied at 4 all, IM and GM commentators were
>speculating that DB is 2750 FIDE. Yet after just two additional games,
>featuring some awful positional howlers on the part of DB, those same
>commentators were talking 2550 FIDE, tops. Surely this demonstrates the
>potential danger in trying to estimate the "true" strength of a computer
>on the basis of a very short match.
>
>Imagine if they had agreed on only 4 games this time instead of 6?!!
>We would all think that the IBM team had taken an *amazing* leap forward
>and that DB is now equal to the best human chessplayer on the planet.
>Instead it is abundantly clear that DB can be crushed positionally, and
>most certainly by other strong players besides Kasparov.
But don't forget -- if it were a four game match, Kasparov might have
played the first game differently -- it seems like he was a little
exploratory in the first game, and was willing to take a loss to learn more
about how DB "thinks." He would have had to start more cautiously from the
beginning with only four games, which could have resulted in more draws.
He wouldn't have had the chance to find the machine's weakness.
Tom
J. Szamosfalvi
: [...]
:
: Hsu, Tan and company deserve credit but have many years of work ahead of
: them if they hope to compete with a Kasparov who now understands very well
: how to exploit the gaping silicon holes.
: Comments?
But now the programmers also know the holes that should be fixed.
I'd bet at least $10 (!!:-) that Deep Blue will trounce the best
human within 5 years.
Don Fong
Geoffrey P. Wyatt <geo...@teleport.com> wrote:
>There is a very interesting irony in that after four games of the DB-
>Kasparov match, with the score tied at 4 all, IM and GM commentators were
>speculating that DB is 2750 FIDE. Yet after just two additional games,
>featuring some awful positional howlers on the part of DB, those same
>commentators were talking 2550 FIDE, tops. Surely this demonstrates the
>potential danger in trying to estimate the "true" strength of a computer
>on the basis of a very short match.
>
>Imagine if they had agreed on only 4 games this time instead of 6?!!
>We would all think that the IBM team had taken an *amazing* leap forward
>and that DB is now equal to the best human chessplayer on the planet.
>Instead it is abundantly clear that DB can be crushed positionally, and
>most certainly by other strong players besides Kasparov.
event. IMHO even 6 games is too short to draw conclusions. although
people will anyway. had DB's handlers accepted the draw offer in game 5,
Kasparov would have won by only 3.5-2.5, the slimmest possible margin of
victory. (that's assuming he still would have won game 6.)
--
--- don fong ``i still want the peace dividend''
--
Mader
G.P. Wyatt wrote:
>There is a very interesting irony in that after four games of the DB-
>Kasparov match, with the score tied at 4 all, IM and GM commentators were
>speculating that DB is 2750 FIDE. Yet after just two additional games,
>featuring some awful positional howlers on the part of DB, those same
>commentators were talking 2550 FIDE, tops. Surely this demonstrates the
>potential danger in trying to estimate the "true" strength of a computer
>on the basis of a very short match.
>Imagine if they had agreed on only 4 games this time instead of 6?!!
>We would all think that the IBM team had taken an *amazing* leap forward
>and that DB is now equal to the best human chessplayer on the planet.
>Instead it is abundantly clear that DB can be crushed positionally, and
>most certainly by other strong players besides Kasparov.
Noone can speculate about an ELO rating for DB only by taking the pure results of a 4 game or 6
game match. I think the commentators looked at the gameplay of DB, they were impressed by the
victory of DB in the first game. When GK found the weak parts of the program he changed his
style and made the computer look like a beginner.
Imagine you are a tennis player with a extremly strong forehand but a weak backhand. You play
against a very strong opponent. Due to your forehand you win the first set and everbody is
impressed. But your opponent has recognized that your backhand is weak and for the rest of
the game he plays the ball only in this direction. You look like a beginner. Now, what is your
real strength?
What I want to say is that GK has checked the weak points of DB and consequently started to
play on the computers backhand. And what is the result of this: Everybody is complaining about
the Patzer-computer: Muscles but no brain. I don't think that many chess grand masters are able
to do such a good job as Kasparov did. If you play the ball on the computers forehand, it would
be possible that you are confronted with an announcement of a "Mate in 20", and wouldn't that
be very impressive, too?
Many people think that DP can beat GK by making his forehand even stronger - two plies more
search depth = 150 ELO more. I think this is the wrong way. The backhand - the positional play
and the chess knowledge - of the computer has to be improved!
Robert Hyatt
No evidence to support this *yet*. I, too, believe that it's going to
happen. However, as game 6 showed, there are "light-years" yet to be
travelled by chess algorithms before they catch the best humans. And
since the best chess machine around is still light-years behind the best
human, and since the human is travelling at near light-speed. The chase
is a long one. Of course, if machines are going even 1mm/year faster
(one millimeter per year) the end result is inevitable, but the chase
might be a *long* one. There are simply so many holes in the machine's
knowledge that are left to fill, it's going to take a long time.
--
Robert Hyatt Computer and Information Sciences
hy...@cis.uab.edu University of Alabama at Birmingham
(205) 934-2213 115A Campbell Hall, UAB Station
(205) 934-5473 FAX Birmingham, AL 35294-1170
halibut
: >But now the programmers also know the holes that should be fixed.
: >I'd bet at least $10 (!!:-) that Deep Blue will trounce the best
: >human within 5 years.
:
: No evidence to support this *yet*. I, too, believe that it's going to
: happen. However, as game 6 showed, there are "light-years" yet to be
: travelled by chess algorithms before they catch the best humans. And
Except for many "great" moves made by Kasparov (as well as many
"safe" moves) the game played by Kasparov in #6 looks like it could have
been played between a USCF master and a Sensory Chess Challenger.
Imre Olajos, Jr.
>But now the programmers also know the holes that should be fixed.
>I'd bet at least $10 (!!:-) that Deep Blue will trounce the best
>human within 5 years.
I think, one of the biggest problems that Deep Blue has currently
is its inability to learn from mistakes. Kasparov beat the machine
because he learned of the mistakes it made and managed to make good
use of this knowledge (see Game 6...). Until AI research gets this
problem solved, no algorithm will be able to beat a human chess
champion.
Oh, yeah, and there is that little problem of human feelings... If
Deep Blue finds that a position can go two ways (both ways having
the same advantages/disadvantages), it tosses a coin. A good
human chessplayer would certainly decide based on inspiration in
a similar situation. (Hope I'm clear enuff.)
------------------------------------------------------------------------
LaLa (Imre Olajos, Jr.) Magyar vagyok. _/ean /\/\ichel _/arre and
WWW : http://homepage.interaccess.com/~lala -=\/=- angelis fan
Home: la...@interaccess.com (USA, Chicago suburbs -> GO BULLS!) __o
Work: Imre....@att.com (Lucent Technologies, Bell Labs) _-\<,_
This email consists of 100% recycled electrons. (_)/ (_)
~~~~~~~~~~ "Music should be free." /Vangelis/ ~~~~~~~~~~~~~~~~~~~~~~~~~~
J. Szamosfalvi
: >But now the programmers also know the holes that should be fixed.
: >I'd bet at least $10 (!!:-) that Deep Blue will trounce the best
: >human within 5 years.
: No evidence to support this *yet*.
Obviously, providing evidence for something that will happen in
the future is not an easy task.... ;-)
: I, too, believe that it's going to
: happen. However, as game 6 showed, there are "light-years" yet to be
: travelled by chess algorithms before they catch the best humans. And
: since the best chess machine around is still light-years behind the best
: human, and since the human is travelling at near light-speed.
Up to certain degree "intelligence" can be substituted with raw
processing power which is `fairly easy' to obtain.
: The chase
: is a long one. Of course, if machines are going even 1mm/year faster
: (one millimeter per year) the end result is inevitable, but the chase
: might be a *long* one. There are simply so many holes in the machine's
: knowledge that are left to fill, it's going to take a long time.
I think `machines' advance a lot faster than humans. And don't
forget, whatever intelligence Deep Blue has also comes from humans.
Robert Hyatt
-->Robert Hyatt (hy...@willis.cis.uab.edu) wrote:
-->
-->: >But now the programmers also know the holes that should be fixed.
-->: >I'd bet at least $10 (!!:-) that Deep Blue will trounce the best
-->: >human within 5 years.
-->:
-->: No evidence to support this *yet*. I, too, believe that it's going to
-->: happen. However, as game 6 showed, there are "light-years" yet to be
-->: travelled by chess algorithms before they catch the best humans. And
-->
--> Except for many "great" moves made by Kasparov (as well as many
-->"safe" moves) the game played by Kasparov in #6 looks like it could have
-->been played between a USCF master and a Sensory Chess Challenger.
I agree. (note, the 10$ part of the above was not posted by me...)
Game 6 could have been kasparov against *any* machine, of any generation,
it looked so overpowered. Masterful play. (by the human... :) )
Robert Hyatt
J. Szamosfalvi <sza...@saul6.u.washington.edu> wrote:
-->Robert Hyatt (hy...@willis.cis.uab.edu) wrote:
-->
-->: >But now the programmers also know the holes that should be fixed.
-->: >I'd bet at least $10 (!!:-) that Deep Blue will trounce the best
-->: >human within 5 years.
-->
-->: No evidence to support this *yet*.
-->Obviously, providing evidence for something that will happen in
-->the future is not an easy task.... ;-)
-->
-->: I, too, believe that it's going to
-->: happen. However, as game 6 showed, there are "light-years" yet to be
-->: travelled by chess algorithms before they catch the best humans. And
-->: human, and since the human is travelling at near light-speed.
-->
-->Up to certain degree "intelligence" can be substituted with raw
-->processing power which is `fairly easy' to obtain.
-->
-->: The chase
-->: is a long one. Of course, if machines are going even 1mm/year faster
-->: (one millimeter per year) the end result is inevitable, but the chase
-->: might be a *long* one. There are simply so many holes in the machine's
-->: knowledge that are left to fill, it's going to take a long time.
-->
-->I think `machines' advance a lot faster than humans. And don't
-->forget, whatever intelligence Deep Blue has also comes from humans.
-->
As people often point out about Cray Blitz, don't forget where the
"stupidity" comes from too. :) In any case, yes programs are moving
forward. Just the rate of improvement is not quite as fast as many
would like, and is too fast for others. I'm a believer in evolution,
and we are seeing it slowly happen. many thought that the deep blue
supercharged, fuel injected, nitromethane powered, high-compression
engine would be the next revolution. In reality, it's a step forward,
but maybe not a giant step (yet.) Time will probably show that they
have a lot of (as yet) untapped potential. As I've said before, they
look frightening to me, particularly given a couple of years to massage
and tune the monster...
George D. Phillies
One proposes that the next two World Champions (FIDE) will be chosen
before a computer can beat the world champion. The PCA list shows 8
people (Kasparov, Karpov, Kramnik, Anand, Kamsky, Ivanchuk, Topalov, Judy
Polgar, Ehlvest) within 100 points of the top, which is a reasonable
estimate of who is likely to win the next (FIDE) World Championship,
assuming that FIDE can bring off a next championship after the current (WC
Karpov against Gata Kamsky) contest is set up and completed. With respect
to the complainer who maintained that only the top 5 on this list have a
chance of winning, while (e.g.) Polgar does not, observe that an 86
point difference in ELO ranking (GK at 2780, JP at 2694) implies
something like an expected 6-4 GK victory, close enough to be winnable by
the other side. Also (I've said this before) JP's tournament results
last year were typically in the 2720 range, so the PCA rating perhaps
understates her current strength.
Congratulations to the programmers of Deep Blue for persuading their
machine to go 2-4 rather than 0-6.
George Phillies
Howell Tumlin
>
> >There is a very interesting irony in that after four games of the DB-
> >Kasparov match, with the score tied at 4 all, IM and GM commentators were
> >speculating that DB is 2750 FIDE. Yet after just two additional games,
> >featuring some awful positional howlers on the part of DB, those same
> >commentators were talking 2550 FIDE, tops. Surely this demonstrates the
> >potential danger in trying to estimate the "true" strength of a computer
> >on the basis of a very short match.
> >
> >Imagine if they had agreed on only 4 games this time instead of 6?!!
> >We would all think that the IBM team had taken an *amazing* leap forward
> >and that DB is now equal to the best human chessplayer on the planet.
> >Instead it is abundantly clear that DB can be crushed positionally, and
> >most certainly by other strong players besides Kasparov.
>
> played the first game differently -- it seems like he was a little
> exploratory in the first game, and was willing to take a loss to learn more
> about how DB "thinks." He would have had to start more cautiously from the
> beginning with only four games, which could have resulted in more draws.
> He wouldn't have had the chance to find the machine's weakness.
>
would ever be willing to accept a loss. I think that he underestimated
Deep Blue, and learned quickly not to slug it out in complex, tactical
positions.
Howell
J. Szamosfalvi
: I think, one of the biggest problems that Deep Blue has currently
: is its inability to learn from mistakes.
I think it depends how you view this. The real brains behind
Deep Blue are the programmers. But even if we restrict learning
to DB, I see no reason why it can't "learn". Two quarters ago
I wrote a Lisp program in CS473 (artificial intelligence) that
could be trained on a set of data to distinquish between edible
and poisonous mushrooms.
: Kasparov beat the machine because he learned of the mistakes it
: made and managed to make good use of this knowledge (see Game 6...).
: Until AI research gets this problem solved, no algorithm will be able
: to beat a human chess champion.
There are a couple of things wrong with this assertion. You're
assuming that the machine will make a mistake. This may very
well be the case, but I wouldn't base my presumption on someone
else's mistake. Also, what if IBM just throw in some hardware
and fine tune the eval function to double the ply depth? Deep
Blue doesn't need to be perfect, or better than a human champ
in every respect to beat him.
: Oh, yeah, and there is that little problem of human feelings... If
: Deep Blue finds that a position can go two ways (both ways having
: the same advantages/disadvantages), it tosses a coin. A good
: human chessplayer would certainly decide based on inspiration in
: a similar situation. (Hope I'm clear enuff.)
Obviously, there are certain things like intuition, feelings that
Deep Blue will probably never have. But I think its raw computing
power should be able to make up for the differences in these areas
in the not very distant future.
D Kirkland
: J. Szamosfalvi <sza...@saul6.u.washington.edu> wrote:
: >
: >But now the programmers also know the holes that should be fixed.
:
: No evidence to support this *yet*. I, too, believe that it's going to
: happen. However, as game 6 showed, there are "light-years" yet to be
: is a long one. Of course, if machines are going even 1mm/year faster
This is a great comment! I like the part:
"and since the human is travelling at near light-speed"
Many have assumed that the strength of a chess program is a
linear function of the search depth. But this has proven to
be *very* wrong in the last few years. Just as it would be
*very* difficult to gain speed at "near light-speed", it is
also proving to be *very* difficult for chess programs to
gain strength as they get up there.
This is also why Bob used the terms he did. I just wanted
to point this out so that *maybe* some of those who don't
realize this may start to catch on!
dan
D Kirkland
: able to beat a human chess champion.
I can't agree with you on this. I think the human chess champion
will be beat by a program long before AI gets good enough to do
it.
dan
Robert Hyatt
D Kirkland <dbk...@cc.utah.edu> wrote:
-->Imre Olajos, Jr. (im...@ihades.ih.att.com) wrote:
-->:
-->: Until AI research gets this problem solved, no algorithm will be
-->: able to beat a human chess champion.
-->
-->I can't agree with you on this. I think the human chess champion
-->will be beat by a program long before AI gets good enough to do
-->it.
-->
-->dan
Stickey Wicket <again>... two points of view: (1) the machine can't
beat the man until the machine works like the man. If this view is
correct, it might be a hopeless problem. (2) the machine can't beat
the man until the machine does everything the man does, albiet maybe
not in the same way. An example is the man understands an isolated
pawn on an open file, then the machine had better understand this.
The man understands that a queen-side majority (kings on king-side)
is an advantage because of the potential distant passed pawn, then the
machine had better understand this. <etc> If this will "do" then a
computer world-champion is not only possible, but probable.
Many of us don't have the luxury of speed that DB has, so that we have
to carefully weigh knowledge vs speed, because a ply is a big problem,
if your opponent is the one with the extra ply. DB is still 1000 times
faster than the rest of us, so if they gave up a factor of ten for a
whomper of an eval function, they'd out-smart us with knowledge, and
*still* out-search us with speed.
They already out-search the human, but it's not enough, they now have
to start out-smarting him too.
Ray Helie
--
-- Noone can speculate about an ELO rating for DB only by taking the pure results of a 4 game or 6
-- game match. I think the commentators looked at the gameplay of DB, they were impressed by the
-- victory of DB in the first game. When GK found the weak parts of the program he changed his
-- style and made the computer look like a beginner.
--
-- Imagine you are a tennis player with a extremly strong forehand but a weak backhand. You play
-- against a very strong opponent. Due to your forehand you win the first set and everbody is
-- impressed. But your opponent has recognized that your backhand is weak and for the rest of
-- the game he plays the ball only in this direction. You look like a beginner. Now, what is your
-- real strength?
--
-- What I want to say is that GK has checked the weak points of DB and consequently started to
-- play on the computers backhand. And what is the result of this: Everybody is complaining about
-- the Patzer-computer: Muscles but no brain. I don't think that many chess grand masters are able
-- to do such a good job as Kasparov did. If you play the ball on the computers forehand, it would
-- be possible that you are confronted with an announcement of a "Mate in 20", and wouldn't that
-- be very impressive, too?
--
-- Many people think that DP can beat GK by making his forehand even stronger - two plies more
-- search depth = 150 ELO more. I think this is the wrong way. The backhand - the positional play
-- and the chess knowledge - of the computer has to be improved!
Thank you for writing that! I'm glad to see there are others that see the
same thing I see. You would think the DB team would wonder why such an
incredible increase in search speed has left them with a program that
still continues to make incredibly bad moves. The question is, are they
going to learn the lesson others (like yourself) already know and move on
to positional/knowledge improvements, or are they going to blow more time
and money going for more useless speed?
Food for thought: how many micro programs out there predicted 95% of the
moves DB made in the match? Kindof makes you wonder..
Ray Helie
-- >Kasparov match, with the score tied at 4 all, IM and GM commentators were
-- >speculating that DB is 2750 FIDE. Yet after just two additional games,
-- >featuring some awful positional howlers on the part of DB, those same
-- >commentators were talking 2550 FIDE, tops. Surely this demonstrates the
-- >potential danger in trying to estimate the "true" strength of a computer
-- >on the basis of a very short match.
-- >
-- >Imagine if they had agreed on only 4 games this time instead of 6?!!
-- >We would all think that the IBM team had taken an *amazing* leap forward
-- >and that DB is now equal to the best human chessplayer on the planet.
-- >Instead it is abundantly clear that DB can be crushed positionally, and
-- >most certainly by other strong players besides Kasparov.
Excellent point! And imagine if they had played 10 games, and Kasparov
continued to win? DB's performance rating would have been even lower!
As has been said, short matches are not very meaningful indicators of a
player's strength.
Robert Hyatt
->Mader <andrea...@siemens.at> wrote:
->--
->-- Noone can speculate about an ELO rating for DB only by taking the pure results of a 4 game or 6
->-- game match. I think the commentators looked at the gameplay of DB, they were impressed by the
->-- victory of DB in the first game. When GK found the weak parts of the program he changed his
->-- style and made the computer look like a beginner.
->--
->-- Imagine you are a tennis player with a extremly strong forehand but a weak backhand. You play
->-- against a very strong opponent. Due to your forehand you win the first set and everbody is
->-- impressed. But your opponent has recognized that your backhand is weak and for the rest of
->-- the game he plays the ball only in this direction. You look like a beginner. Now, what is your
->-- real strength?
->--
->-- What I want to say is that GK has checked the weak points of DB and consequently started to
->-- play on the computers backhand. And what is the result of this: Everybody is complaining about
->-- the Patzer-computer: Muscles but no brain. I don't think that many chess grand masters are able
->-- to do such a good job as Kasparov did. If you play the ball on the computers forehand, it would
->-- be possible that you are confronted with an announcement of a "Mate in 20", and wouldn't that
->-- be very impressive, too?
->--
->-- Many people think that DP can beat GK by making his forehand even stronger - two plies more
->-- search depth = 150 ELO more. I think this is the wrong way. The backhand - the positional play
->-- and the chess knowledge - of the computer has to be improved!
->
->Thank you for writing that! I'm glad to see there are others that see the
->same thing I see. You would think the DB team would wonder why such an
->incredible increase in search speed has left them with a program that
->still continues to make incredibly bad moves. The question is, are they
->going to learn the lesson others (like yourself) already know and move on
->to positional/knowledge improvements, or are they going to blow more time
->and money going for more useless speed?
->
->
->Food for thought: how many micro programs out there predicted 95% of the
->moves DB made in the match? Kindof makes you wonder..
Hard to say. Someone reported that in game 1, Crafty made something
like 31 out of the 36 moves DB made, running on a 486 or something.
That's not unexpected. The problem is the 5 moves it *didn't* make.
Those were probably critical points where DB's speed made a big
difference. Quite probably, 95% of the moves in a game are not so
critical and one of several maintains status quo, but at a few critical
points, there is only one move (sometimes) that must be played. I
would not draw the conclusion that Crafty is almost as good as DB
just because it got 31/36 right. Those other 5 might well have lost
the game quickly.
Marcel van Kervinck
: Kasparov match, with the score tied at 4 all, IM and GM commentators were
: speculating that DB is 2750 FIDE. Yet after just two additional games,
: featuring some awful positional howlers on the part of DB, those same
: commentators were talking 2550 FIDE, tops. Surely this demonstrates the
: potential danger in trying to estimate the "true" strength of a computer
: on the basis of a very short match.
When asked after 3 games, Kasparov estimated DB's strength to be
at ~2700. With the usual reservations, ofcourse.
Marcel
-- _ _
_| |_|_|
|_ |_ Marcel van Kervinck
|_| bue...@urc.tue.nl
John Brock
Geoffrey P. Wyatt <geo...@teleport.com> wrote:
>There is a very interesting irony in that after four games of the DB-
>Kasparov match, with the score tied at 4 all, IM and GM commentators were
>speculating that DB is 2750 FIDE. Yet after just two additional games,
>featuring some awful positional howlers on the part of DB, those same
>commentators were talking 2550 FIDE, tops. Surely this demonstrates the
>potential danger in trying to estimate the "true" strength of a computer
>on the basis of a very short match.
The 2/26/96 issue of Time magazine has an article on the match (a
reasonably good one actually, considering their readership). It
includes a "tale of the tape" type table comparing Kasparov and Deep
Blue, in particular their ratings: Kasparov 2775, Deep Blue 2450. Any
ideas where this might come from? It occurs to me that it ought to be
possible to calculate a provisional rating from a six game match, and
maybe 2450 is the number that pops out of a 4-2 score against a 2775
player. I don't know how this is calculated though: could anyone post
the formula and the result? Just intuitively 2450 seems a bit low --
perhaps it should have been 2550 like you said, and Time just got it
wrong.
--
John Brock
jbr...@panix.com
David Franklin
[in reply to someone else, whose name I appear to have misplaced].
: -- Many people think that DP can beat GK by making his forehand even stronger - two plies more
: -- search depth = 150 ELO more. I think this is the wrong way. The backhand - the positional play
: -- and the chess knowledge - of the computer has to be improved!
:
: Thank you for writing that! I'm glad to see there are others that see the
: same thing I see. You would think the DB team would wonder why such an
: incredible increase in search speed has left them with a program that
: still continues to make incredibly bad moves. The question is, are they
: going to learn the lesson others (like yourself) already know and move on
: to positional/knowledge improvements, or are they going to blow more time
: and money going for more useless speed?
:
I think DB are entirely aware of the shortcomings in their strategy. But
as they claim to get +90% wins against any PC program, and would probably
beat 99.99% of humans, one has to wonder just how "incredibly bad" the
moves were.
I don't know what you "knowledge junkies" think is stopping you from
building a high-knowledge program that blows DB out of the water. OK,
you ain't got the hardware, but (a) you seem to think extra speed is
useless, and (b) once you get a really impressive PC version, I'm sure
you could get Intel to sponser a massively parallel version. If the
reason is just that you don't know how to write such a program (which
is fine, cos I don't know anyone who does), then perhaps you shouldn't
be so critical of the DB team for not knowing how to either.
Aesthetically, everyone would like to see a non-brute force method
for micros, but it just hasn't happened. To use the forehand/backhand
analogy - the easiest way I can see to build a "tennis robot" would be
just to give it a 600 mph serve. I don't need to worry about how to make
it return serve, volley, etc., and it'll still do OK against human
opposition (heck, it'd probably win most matches by default if the
serve hit its opponent :-)
:
: Food for thought: how many micro programs out there predicted 95% of the
: moves DB made in the match? Kindof makes you wonder..
Food for thought: how many games would Kasparov win against decent opposition
if he made completly incorrect moves 5% of the time?
Dave
Paul Rubin
David Franklin <kao...@dircon.co.uk> wrote:
>I don't know what you "knowledge junkies" think is stopping you from
>building a high-knowledge program that blows DB out of the water. OK,
>you ain't got the hardware, but (a) you seem to think extra speed is
>useless, and (b) once you get a really impressive PC version, I'm sure
>you could get Intel to sponser a massively parallel version. If the
>reason is just that you don't know how to write such a program (which
>is fine, cos I don't know anyone who does), then perhaps you shouldn't
>be so critical of the DB team for not knowing how to either.
Has anyone seriously tried writing a high-knowledge program in recent
times? One that really tries to do good positional evaluation in the
top few search levels (maybe with a rule-based expert system) and ALSO
does a full-width search to lots of ply using normal (fast)
brute-force methods, using the fancy evaluator to choose between
variations along the principal lines. The programs that I know of
that tried to use lots of knowledge lost because they didn't do enough
searching and consequently missed some tactics. But even PC-class
hardware is fast enough now to do BOTH.
David Greene
knowledge-based approaches might help illicit cognitive shortcuts used
in lieu of computational speed.
Brute-force methods (with limited knowledge) create an opportunity to
see emergent behavior -- something like Deep Blue begins to look like
it's playing a strategy but is simply optimizing look ahead.
Extremes are usually more useful for science than confounding variables
with compromise. Besides the problem of how to compromise between
approaches is not trivial -- if two approaches disagree, when should
look-ahead prevail and when should the knoweldge base?
I'd rather see how far (how good a player) a program could be relying on
one extreme or the other. I suspect a hybrid system *might* have a
better chance at winning, but just winning isn't the point.
-David
David Franklin
: I'd rather see how far (how good a player) a program could be relying on
: one extreme or the other. I suspect a hybrid system *might* have a
: better chance at winning, but just winning isn't the point.
Given that the *extreme* end of the DB situation is an eval function
of basically +1 for mate, -1 for mated, 0 for stalemate,repetition etc.,
I don't think you'd get much useful chess for such a system until you got
to about 50-60 ply...
At the moment, I'd say winning *is* the point. But suppose a tech.
breakthrough (e.g. the mythical quantum computer) occurs, and DB2 runs
100000000 times faster, and convincingly beats Kasparov. I'd hope some
interest would then evolve in a system that could beat Kasparov without
having to outsearch him by a factor of about 10^16.
Dave
SheppardCo
> times? One that really tries to do good positional evaluation in the
> top few search levels (maybe with a rule-based expert system) and ALSO
> does a full-width search to lots of ply using normal (fast)
> brute-force methods, using the fancy evaluator to choose between
> variations along the principal lines.
Experience suggests that this approach, while seeming to make
excellent use of available computer resources, has hidden
difficulties. In particular, the expert-system has to integrate
its own knowledge with the information that comes from a
deep search, but the knowledge returned by a search is of
an essentially different nature that what might be gleaned
from positional analysis.
An example: during the deep search portion of the tree, the
strategic nature of the tree changes along some lines of play,
but not along others. The evaluations cannot be compared
because they have a different basis.
Warm Regards
Brian Sheppard
Thien D Tran
: In article <4g8333$i...@maureen.teleport.com>,
Hi all!
I think that the Time got it right!
Deep Blue was beaten by Frizt 4! So, it deserves that low points!
How is Frizt 4 or 5 rating now?
best regards,
Thien Tran
: John Brock
: jbr...@panix.com
--
Jim Ebright
<sza...@saul6.u.washington.edu> wrote:
>Obviously, there are certain things like intuition, feelings that
>Deep Blue will probably never have. But I think its raw computing
>power should be able to make up for the differences in these areas
>in the not very distant future.
I played on some the best chess computers in the world 25 years ago.
Any reasonable player could beat them... even me! :) (...but not always ;)
Now it takes the world champion to bet the best... The only thing that
will make the above quote untrue is lack of further funding.
--
A/~~\A 'moo2u from osu' Jim Ebright e-mail: ebr...@bronze.coil.com
((0 0))_______ "A million dollars, ... I can get that. I can get that
\ / the \ in cash. That's no problem." -secret White House tape.
(--)\ OSU | "Always be your best. Never be petty." -resignation speech.
bob_ra...@mindlink.bc.ca
>At the moment, I'd say winning *is* the point. But suppose a tech.
>breakthrough (e.g. the mythical quantum computer) occurs, and DB2 runs
>100000000 times faster, and convincingly beats Kasparov. I'd hope some
>interest would then evolve in a system that could beat Kasparov without
>having to outsearch him by a factor of about 10^16.
At the time I was reading this thread, yours was the last posting in
it, and I was just about to launch into a tirade about how all you
guys are missing the point. But you at least seemed to have nailed
it.
What is ultimately so unsatisfying about IBM's (non)achievement is
their towering _laziness_! This isn't about knowledge-based (read
"bigger bag of tricks") or speed-based computing. What's missing from
the equation is the normal process of thesis/antithesis/synthesis
thinking that people do when they solve chess scenarios, either game
positions or composed problems.
Whenever a person loses a game of chess, they know not only _that_
they lost it but _how_ they lost it; or when they are looking for a
good move in a position and they see a move and eventually reject it
they know what they didn't like about it and only go looking for moves
that will address what it was about they previous move that they
didn't like. Of the next moves that they consider on that basis,
maybe some of them are no good for a different reason, so now they've
got two stars to navigate by, and on like that.
It's like the Newton/Raphson (sp?) method for finding local maxima &
minima of a curve by successive approximation. It zeroes in quickly
because in essence it lets the curve (which knows more about itself
than anyone) do all the work. Why isn't the work of computer chess
programming pioneers being put to the task of coding the equivalent of
symbolic logic and truth-tables to solve what's going on in a given
position by _building up_ one large live idea out of a number of small
dead ones, instead of the current "nope that one's no good, nope, that
one's no good, nope, that one's no good" method of "evaluation" and
"selection".
When the IBM guys report that some of them are losing interest in the
chess project from a computer science point of view, to me that smacks
of sour grapes. And when Kasparov sweet-talks what a swell job the
IBM team has done, it sounds a whole lot like he knows a profitable
venture when he sees one and wants to keep that particular
golden-egg-laying goose alive for another $$ infusion later. I bet he
talks way different about the match in private.
--
Bob Raisbeck
bob_ra...@mindlink.bc.ca
K. Kretschel
|> Whenever a person loses a game of chess, they know not only _that_
|> they lost it but _how_ they lost it; or when they are looking for a
|> good move in a position and they see a move and eventually reject it
|> they know what they didn't like about it and only go looking for moves
|> that will address what it was about they previous move that they
|> didn't like. Of the next moves that they consider on that basis,
|> maybe some of them are no good for a different reason, so now they've
|> got two stars to navigate by, and on like that.
I do not agree with that. Noone knows really why he lost unless he does
a lot of analysis which means nothing but that he is calculating farther
and more exactly than he could during the game. It would be easy to
program the computer to do the same, but that would not make sense unless
it is also programmed to be able to learn from it.
Now this is the big difference: The human learns from his mistakes, in
terms of computer chess he is able to change his evaluation function.
Of course a computer could be programmed to do the same. However it
seems to be a task about that programmers seem not to know enough to do
it efficiently until now. So they still stick to the old methods gaining
strength by speed, and as long as they are more successful than people
trying new methods they win the competition.
Klaus
--
E-Mail: k.kre...@dlr.de
CAPSA
How DARE you bring up human philosophica concepts concerning knowledge
and science (another word for "knowledge"?) when discussing the Deep
Blue vs Kasparov match and computer methods of attacking (and that IS
the correct word) chess? I admit, you did hit the nail on the head in
the last sentences. The whole thing concerned money, NOT scientific
investigation.
Fraternally,
Jerome Bibuld