Did Cray Blitz really have a win?
Sam Sloan
This question comes from the 1986 World Computer Chess Championship.
In the following position, with black to move, Cray Blitz said that it
had run out of computer time on the host computer and therefore
requested an adjudication.
In the final position is: 8/2p5/2B2k2/p4bp1/3P4/6K1/1P6/8
Black, Cray Blitz, had a pawn more, but it is almost impossible to win
this position, even for a strong human, much less for a computer.
Grandmaster Hort and International Master Valvo, the arbiters for the
event, proceeded to spend two hours analyzing this position and
finally declared it a win for black. Cray Blitz won the next game as
well and was declared world computer champion on tie-breaks.
As I recall, the winning method found by Hort and Valvo involved
bringing the king all the way around with Ke7, Kd8, Kc8, Kb8, Ka7,
Kb6, Ka5 (after the pawn has moved), trading pawns and then bringing
the king around behind White.
Because of the horizon effect, no computer of that era could have
found such a solution.
I have never been convinced that black has a win here with best play.
Can somebody run this through their computer and see what the result
is?
Remember, if it turns out that this position with best play is a draw
and not a win for black, then Cray Blitz did not win the 1986 World
Computer Chess Championship. Cray Blitz had already lost in round two.
As this was a five round Swiss, a loss here would have eliminated Cray
Blitz from any contention for a top prize.
I have posted this game at http://www.anusha.com/awit-rex.htm
Although this concerns events of 13 years ago, there were 59 postings
about this game on the newsgroups yesterday. Certain persons have
based their reputations on the result of this game.
Sam Sloan
[Event "World Computer Chess Championship"]
[Site "Koln, Germany"]
[Date "1986.??.??"]
[Round "4"]
[White "Awit "]
[Black "Rex"]
[Result "1-0"]
[ECO "A01"]
1. b3 e5 2. Bb2 Nc6 3. e3 d5 4. Bb5 Qg5 5. Nf3
Qxg2 6. Rg1 Qh3 7. Bxe5 Bg4 8. Rg3 Bxf3 9. Qxf3
Qh6 10. Qxd5 Nge7 11. Bxc6+ bxc6 12. Qf3 Nd5 13. Qg4
Qxh2 14. Bxg7 Bxg7 15. Qxg7 Ke7 16. Qe5+ Kd7 17. Na3
Ne7 18. Qd4+ Ke6 19. Qe4+ Kd7 20. Rg7 Qh5 21. Nc4
Ke8 22. Ne5 f5 23. Qg2 Qh6 24. Nxc6 Nxc6 25. Rxc7
Kf8 26. Rxc6 Qg7 27. Rf6+ Qxf6 28. Qxa8+ Kf7 29. Qxa7+
Kf8 30. d4 Ke8 31. Ke2 Kf8 32. c4 h6 33. a4
Rg8 34. a5 Rg7 35. Qc5+ Re7 36. a6 f4 37. a7
f3+ 38. Kf1 Kg8 39. a8=Q+ Kh7 40. Qh5 Rf7 41. Ra6
Kg7 42. Rxf6 Rxf6 43. Qe5 h5 44. Qxf3 1-0
Sam Sloan
Sam Sloan
[Event "WCCC"]
[Site "Cologne"]
[Date "1986.06.??"]
[Round "4"]
[White "Schach 2.7"]
[Black "Cray Blitz"]
[Result "0-1" adjudicated after Cray Blitz ran out of Cray]
1. d4 Nf6 2. c4 e6 3. Nf3 d5 4. Nc3 Be7 5. Bg5 O-O 6. e3 Ne4 7. Bxe7
Qxe7 8. Nxe4 dxe4 9. Nd2 e5 10. dxe5 Qxe5 11. Qc2 Na6 12. O-O-O Bg4
13. f3
exf3 14. gxf3 Bh5 15. Bd3 f5 16. Rhe1 Rad8 17. Bf1 Qxh2 18. c5 Qe5 19.
c6 Kh8
20. cxb7 Nc5 21. Nc4 Qg3 22. Qxf5 Rxd1+ 23. Rxd1 Qxf3 24. Qxf3 Bxf3
25. Rd4
Bxb7 26. Be2 g6 27. Ne5 Re8 28. Ng4 Re4 29. Kd2 h5 30. Nf6 Rxd4+ 31.
exd4 Ne4+
32. Nxe4 Bxe4 33. Ke3 Bd5 34. Kf4 Bxa2 35. Bd3 Kg7 36. Kg5 Bf7 37. Be4
a5
38. Bc6 Bg8 39. Be4 Bh7 40. Bd3 h4 41. Bf1 Bg8 42. Bh3 Bd5 43. Bg4 Be4
44. Kxh4 Kf6 45. Bd7 Bf5 46. Bc6 g5+ 47. Kg3 0-1 {Adjudicated a win
for Black}
Chris Whittington
Sam Sloan wrote in message <36e8f8b3...@nntp.mindspring.com>...
>Did Cray Blitz Really have a win?
>
>This question comes from the 1986 World Computer Chess Championship.
>
>In the following position, with black to move, Cray Blitz said that it
>had run out of computer time on the host computer and therefore
>requested an adjudication.
>
>In the final position is: 8/2p5/2B2k2/p4bp1/3P4/6K1/1P6/8
>
>Black, Cray Blitz, had a pawn more, but it is almost impossible to win
>this position, even for a strong human, much less for a computer.
>
>Grandmaster Hort and International Master Valvo, the arbiters for the
>event, proceeded to spend two hours analyzing this position and
>finally declared it a win for black. Cray Blitz won the next game as
>well and was declared world computer champion on tie-breaks.
>
>As I recall, the winning method found by Hort and Valvo involved
>bringing the king all the way around with Ke7, Kd8, Kc8, Kb8, Ka7,
>Kb6, Ka5 (after the pawn has moved), trading pawns and then bringing
>the king around behind White.
Computer chess is mostly about that indeterminate gap between coming out of
a stored opening library and going into a known win/loss/draw, either
through egtbs or mate or whatever.
In this gap, programs rely on heuristics. They think that maybe a double
pawn is pawn, or a knight in the centre is good, so, in effect they stagger
around, outputting moves that 'increase' their score.
In this context, HOW they score and evaluate becomes critical. How they
score and evaluate when there is nothing obvious to do to win material.
I don't know how far CrayBlitz was searching in this position. Probably
8,9,10 plies. There is no way it was seeing as far as passed pawn
conversion. It would possibly have had square of the pawn code (freepawn
able to run, and the opponent king not able to catch it), but this would
have been irrelevant to the evaluation with bishops on the board.
For several moves prior to the adjudication position, CrayBlitz was going
backwards, ending up defending its g6 pawn with its bishop on h7. I thought,
positionally, its opponent was playing quite nicely. CrayBlitz however, was
concentrating on one thing and one thing only, hanging on to its extra
pawns. And tying itself up in knots in the process.
I think, had the square i8 been available to it, CrayBlitz would have moved
its bishop there as well, in an attempt to cover the g6 pawn; and we'ld have
seen some more mindless repetitions of moves until CrayBlitz realised it was
only heading for a draw.
As it was, without square i8, CrayBlitz was forced to give up one of the
pawns it was defending (to prevent draw), and this freeing manoeuvre led to
the position where it was at least arguable that it could have been a
best-play win.
The truth is that the beast (at near GM level, allegedly) had no idea at all
what it was doing and merely blundered into the final position by luck and
no good judgement. The game was pitiful to watch and one can imagine
spectators and involved competitors feeling and saying, no doubt, that this
was (a) no world champion and (b) the process was a joke and (c) the
adjudication time point was critical to the result. Done earlier and it
could well have been a best play draw.
That it ended with adjudication merely compounded the nonsense.
I guess it was the same as now for Hyatt. Still being attacked for
bean-counting and relying on massive hardware. Losing to 400 nps program in
round 2. And then, suddenly, the rise of the micro-computers. Whilst the
main-frame manufacturors realised they'ld been conned and pulled out of
giving freebie time away. (Cologne 1986 had 2 Crays, 1 VLSI system, 20 Suns
in one, 2 Amdahls - whatever they are, 1 Gould - whatever that is, 1
Microvax, and a smattering of 68000's and 6502's).
This was Hyatt's finest hour. But it was spoilt. Was he really world
champion with his Cray Blitz ? And what happened when the micro programs
started to get competitive ?
So when Bruce says, in another post, that events 13 years ago are a long way
from now, he's wrong. History kind of repeats itself. Hyatt's revenge was by
proxy on Kasparov. All the same people involved. But proxy revenge is never
good enough. Hyatt is a driven man.
This one will run and run.
Chris Whittington
>
>Because of the horizon effect, no computer of that era could have
>found such a solution.
>
>I have never been convinced that black has a win here with best play.
>Can somebody run this through their computer and see what the result
>is?
>
>Remember, if it turns out that this position with best play is a draw
>and not a win for black, then Cray Blitz did not win the 1986 World
>Computer Chess Championship. Cray Blitz had already lost in round two.
>As this was a five round Swiss, a loss here would have eliminated Cray
>Blitz from any contention for a top prize.
>
>I have posted this game at http://www.anusha.com/awit-rex.htm
>
>Although this concerns events of 13 years ago, there were 59 postings
>about this game on the newsgroups yesterday. Certain persons have
>based their reputations on the result of this game.
>
>Sam Sloan
>
Henri H. Arsenault
(Sam Sloan) wrote:
> Remember, if it turns out that this position with best play is a draw
> and not a win for black, then Cray Blitz did not win the 1986 World
> Computer Chess Championship. Cray Blitz had already lost in round two.
> As this was a five round Swiss, a loss here would have eliminated Cray
> Blitz from any contention for a top prize.
>
Even if the adjudicators made a mistake in evaluation the position (which
is not at all established), it does not prove that Cray Blitz did not win
the tournament: the procedure was accompli9shed according to the
agreed-upon rules, and therefore the adjudication result stands.
Do you think that in the recent football game where one team was elminated
when the other team scored a touchdown that video later showed as invalid
changes the result of the Super Bowl? The team that won was the team that
was declared the victor according to the rules, and that result stands
whether or not it can be later proven that a mistake was made. The same
goes for chess. The only exception is when it can be proven without any
doubt that willful cheating was involved, but no one has claimed that the
adjudicators were bought off...
Henri
Robert Hyatt
: Did Cray Blitz Really have a win?
: This question comes from the 1986 World Computer Chess Championship.
: In the following position, with black to move, Cray Blitz said that it
: had run out of computer time on the host computer and therefore
: requested an adjudication.
: In the final position is: 8/2p5/2B2k2/p4bp1/3P4/6K1/1P6/8
: Black, Cray Blitz, had a pawn more, but it is almost impossible to win
: this position, even for a strong human, much less for a computer.
: Grandmaster Hort and International Master Valvo, the arbiters for the
: event, proceeded to spend two hours analyzing this position and
: finally declared it a win for black. Cray Blitz won the next game as
: well and was declared world computer champion on tie-breaks.
: As I recall, the winning method found by Hort and Valvo involved
: bringing the king all the way around with Ke7, Kd8, Kc8, Kb8, Ka7,
: Kb6, Ka5 (after the pawn has moved), trading pawns and then bringing
: the king around behind White.
: Because of the horizon effect, no computer of that era could have
: found such a solution.
Just goes to show what you know about computer chess. The Cray Blitz
of that era was about as fast as Crafty was 18 months ago running on a
P6/200 type machine. We were searching about 80K nodes per second in
the middlegame, popping to 100-150K in endgames like the one in question.
Cray Blitz's search was reasonably similar to what many programs (not
so much crafty) do today. Full-width to some depth, then 4 plies of
selectivity where it could include moves it thought 'interesting' and
not include everything, and then that followed by a traditional q-search
that included getting out of check (recognizing mate in the q-search) as
well as including some non-capturing checks into the q-search. It did
similar extensions to what I do today, out of check, one-legal-move,
recaptures, captures around the king, passed pawn pushes.
IE your statement is uninformed. It had no more serious 'horizon effect'
problems than a program like Rebel does today. Crafty/Ferret probably suffer
more from horizon problems with the null-move R=2 we use. Cray Blitz used
R=1 and only allowed one null move in any complete path (non-recursive null
move).
In light of that, your statement is wrong. As I said, our PV sparked the
original idea. And I don't recall waiting 2 hours for the result. It was
closer to one.
: I have never been convinced that black has a win here with best play.
: Can somebody run this through their computer and see what the result
: is?
: Remember, if it turns out that this position with best play is a draw
: and not a win for black, then Cray Blitz did not win the 1986 World
: Computer Chess Championship. Cray Blitz had already lost in round two.
: As this was a five round Swiss, a loss here would have eliminated Cray
: Blitz from any contention for a top prize.
Another false statement. The tournament is over. The result recorded.
There is no facility to 'undo' an adjudication, whether it was right or
wrong. At the time, everyone involved agreed it was won.
: I have posted this game at http://www.anusha.com/awit-rex.htm
: Although this concerns events of 13 years ago, there were 59 postings
: about this game on the newsgroups yesterday. Certain persons have
: based their reputations on the result of this game.
I assume you mean me for one. Which would mean that finishing first or
tying for first at the ACM events from 1981-1984 didn't mean anything.
And you do know which program won the 1983 WCCC event, correct? No
adjudications. Direct communication to the computer with a terminal.
I hope you are in the wine-making business, because you got a lot of
sour grapes to work with, it seems.
--
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
ChessAngel
Sam Sloan wrote in message <36e8fb17...@nntp.mindspring.com>...
>Did Cray Blitz Really have a win?
>
>This question comes from the 1986 World Computer Chess Championship.
>
>In the following position, with black to move, Cray Blitz said that it
>had run out of computer time on the host computer and therefore
>requested an adjudication.
>
>In the final position is: 8/2p5/2B2k2/p4bp1/3P4/6K1/1P6/8
>
>Black, Cray Blitz, had a pawn more, but it is almost impossible to win
>this position, even for a strong human, much less for a computer.
>
>Grandmaster Hort and International Master Valvo, the arbiters for the
>event, proceeded to spend two hours analyzing this position and
>finally declared it a win for black. Cray Blitz won the next game as
>well and was declared world computer champion on tie-breaks.
>
>As I recall, the winning method found by Hort and Valvo involved
>bringing the king all the way around with Ke7, Kd8, Kc8, Kb8, Ka7,
>Kb6, Ka5 (after the pawn has moved), trading pawns and then bringing
>the king around behind White.
>
>Because of the horizon effect, no computer of that era could have
>found such a solution.
>
>Can somebody run this through their computer and see what the result
>is?
>
>Remember, if it turns out that this position with best play is a draw
>and not a win for black, then Cray Blitz did not win the 1986 World
>Computer Chess Championship. Cray Blitz had already lost in round two.
>As this was a five round Swiss, a loss here would have eliminated Cray
>Blitz from any contention for a top prize.
>
>
>Although this concerns events of 13 years ago, there were 59 postings
>about this game on the newsgroups yesterday. Certain persons have
>based their reputations on the result of this game.
>
>
>[Event "WCCC"]
>[Site "Cologne"]
>[Date "1986.06.??"]
>[Round "4"]
>[White "Schach 2.7"]
>[Black "Cray Blitz"]
>[Result "0-1" adjudicated after Cray Blitz ran out of Cray]
>
>1. d4 Nf6 2. c4 e6 3. Nf3 d5 4. Nc3 Be7 5. Bg5 O-O 6. e3 Ne4 7. Bxe7
>Qxe7 8. Nxe4 dxe4 9. Nd2 e5 10. dxe5 Qxe5 11. Qc2 Na6 12. O-O-O Bg4
>13. f3
>exf3 14. gxf3 Bh5 15. Bd3 f5 16. Rhe1 Rad8 17. Bf1 Qxh2 18. c5 Qe5 19.
>c6 Kh8
>20. cxb7 Nc5 21. Nc4 Qg3 22. Qxf5 Rxd1+ 23. Rxd1 Qxf3 24. Qxf3 Bxf3
>25. Rd4
>Bxb7 26. Be2 g6 27. Ne5 Re8 28. Ng4 Re4 29. Kd2 h5 30. Nf6 Rxd4+ 31.
>exd4 Ne4+
>32. Nxe4 Bxe4 33. Ke3 Bd5 34. Kf4 Bxa2 35. Bd3 Kg7 36. Kg5 Bf7 37. Be4
>a5
>38. Bc6 Bg8 39. Be4 Bh7 40. Bd3 h4 41. Bf1 Bg8 42. Bh3 Bd5 43. Bg4 Be4
>
>44. Kxh4 Kf6 45. Bd7 Bf5 46. Bc6 g5+ 47. Kg3 0-1 {Adjudicated a win
>for Black}
>
I have been looking at this position since yesterday. At home I have Fritz
5.32 and Rebel 10 both running it. Fritz has KE6 as next at 26 ply. I will
post more details later.
Personally I don't think a computer could have won from this position in 86.
However that's only my HO, but face facts even if it were now determined
that the position is a draw, Cray Blitz still won the event and it was won
without cheating. If the Schach people had objected to the adjudication then
was the time to do so, the officials at the event called it a win for black
and that's the way it is.
What would you do back in the earlier days when computers couldn't mate
without tablebases KBN V k adjudicate that as a draw? the point being that
adjudication takes into account best play for both sides not the level of
the competitors and GM Hort is far better IMHO (or was then) than Fritz is
today.
(\ /)
( \ _ _ / )
( \ ( ) / )
( /< >\ )
( / \/ \/ \ )
/ \
( )
Angel of Chess
Robert Hyatt
: Sam Sloan wrote in message <36e8f8b3...@nntp.mindspring.com>...
15-17 typically. Simple endgame. Huge hash table. 80-150K nodes per
sec in that particular simple ending.
: . There is no way it was seeing as far as passed pawn
: conversion. It would possibly have had square of the pawn code (freepawn
: able to run, and the opponent king not able to catch it), but this would
: have been irrelevant to the evaluation with bishops on the board.
It didn't see any promotion.
: For several moves prior to the adjudication position, CrayBlitz was going
: backwards, ending up defending its g6 pawn with its bishop on h7. I thought,
: positionally, its opponent was playing quite nicely. CrayBlitz however, was
: concentrating on one thing and one thing only, hanging on to its extra
: pawns. And tying itself up in knots in the process.
: I think, had the square i8 been available to it, CrayBlitz would have moved
: its bishop there as well, in an attempt to cover the g6 pawn; and we'ld have
: seen some more mindless repetitions of moves until CrayBlitz realised it was
: only heading for a draw.
That is ok. Because we had an interesting 'draw score' algorithm that was
published. Because for us, 'all draws were not created equal' and we always
took the draw that occurred farthest into the future, which generally would
pull us down a path that either let the opponent make a mistake, or which
would eventually let us see far enough to see how to win.
Nothing says you have to win the game as quickly as possible. And in those
positions where you can't see the win, if you avoid the draw, you can
occasionally stumble into the win. And the algorithm worked pretty well in
several games over the years.
: As it was, without square i8, CrayBlitz was forced to give up one of the
: pawns it was defending (to prevent draw), and this freeing manoeuvre led to
: the position where it was at least arguable that it could have been a
: best-play win.
: The truth is that the beast (at near GM level, allegedly) had no idea at all
: what it was doing and merely blundered into the final position by luck and
: no good judgement. The game was pitiful to watch and one can imagine
: spectators and involved competitors feeling and saying, no doubt, that this
: was (a) no world champion and (b) the process was a joke and (c) the
: adjudication time point was critical to the result. Done earlier and it
: could well have been a best play draw.
: That it ended with adjudication merely compounded the nonsense.
: I guess it was the same as now for Hyatt. Still being attacked for
: bean-counting and relying on massive hardware. Losing to 400 nps program in
: round 2. And then, suddenly, the rise of the micro-computers. Whilst the
: main-frame manufacturors realised they'ld been conned and pulled out of
: giving freebie time away. (Cologne 1986 had 2 Crays, 1 VLSI system, 20 Suns
: in one, 2 Amdahls - whatever they are, 1 Gould - whatever that is, 1
: Microvax, and a smattering of 68000's and 6502's).
The main-frame manufacturers had not been 'conned'. Thru the 1989
WCCC event, I don't recall of anything winning one of the ACM or WCCC
events except for a 'mainframe' machine. In 1992 the two 'big' mainframes
weren't there. And in 1995 yes a micro beat deep thought. But the micro
'victories' are few and far between, so I doubt the folks at Cray, IBM,
Amdahl, CDC, Fujitsu, Hitechi, Honeywell, etc felt 'scammed' as you would
imply.
: This was Hyatt's finest hour. But it was spoilt. Was he really world
: champion with his Cray Blitz ? And what happened when the micro programs
: started to get competitive ?
Not a thing. Except that I became more interested in working on a machine
where I could play tens of thousands of games a year, rather than playing
in one tournament (at best) because of scheduling issues.
: So when Bruce says, in another post, that events 13 years ago are a long way
: from now, he's wrong. History kind of repeats itself. Hyatt's revenge was by
: proxy on Kasparov. All the same people involved. But proxy revenge is never
: good enough. Hyatt is a driven man.
I don't have a clue what you mean there. 'proxy over kasparov'? You'll find
out that Hsu and group have little to do with what I do, and what I do has
little to do with what they do. We just 'communicate' from time to time.
I don't get my thrills vicariously through them, however.
: This one will run and run.
Nope... it was over in 1986. Still is 'over'.
It doesn't really matter to me what you think about the game. The
original thread by Sam Sloan talked about 'cheating on the top boards'.
Now we have moved to attacking the TD and a GM for not adjudicating a
game properly without any notion that the game isn't won as they said.
And there has been no connection established between the adjudication
and the notion of 'cheating'.
There is some other 'motive' here. I assume it will become apparent as
time wears on.
Steffen A. Jakob
[schnipp]
> What would you do back in the earlier days when computers couldn't mate
> without tablebases KBN V k adjudicate that as a draw? the point being that
> adjudication takes into account best play for both sides not the level of
> the competitors and GM Hort is far better IMHO (or was then) than Fritz is
> today.
And most of all Hort's jokes are much better than those of Fritz.
Greetings,
Steffen, who claims that Germany is the true soccer world champion
of 1966.
--
Steffen A. Jakob | "Victory goes to the player who makes the
ste...@jakob.at | second-to-last mistake."
http://www.jakob.at/ | (Savielly Grigorievitch Tartakower)
Robert Hyatt
: I have been looking at this position since yesterday. At home I have Fritz
: 5.32 and Rebel 10 both running it. Fritz has KE6 as next at 26 ply. I will
: post more details later.
: Personally I don't think a computer could have won from this position in 86.
Here I disagree... because in 1986, the Cray was not exactly 'slow'. IE
as I said, in this event we ran on two different machines because of
scheduling problems... and on the faster of the two, we were searching
around 200K nodes per sec in the middlegame and 500K in the endgame. On
the slower machine, we were searching around 80K in the middlegame and
150K in the endgame (we were on this machine for round 4). That speed
is somewhat faster than Crafty on a P6/200 from 18 months ago. I don't know
whether Crafty or CB could have won that game. CB thought it could. Crafty
today thinks it can. But one thing is important... what we were doing in 1986
is not a lot different from what we are doing today. Similar extensions.
Similar evaluations. Similar search (CB's may have been better, as I am
trying 'simple things' today before I consider trying checks in the q-search
and selective things between the basic and q-searches.
: However that's only my HO, but face facts even if it were now determined
: that the position is a draw, Cray Blitz still won the event and it was won
: without cheating. If the Schach people had objected to the adjudication then
: was the time to do so, the officials at the event called it a win for black
: and that's the way it is.
: What would you do back in the earlier days when computers couldn't mate
: without tablebases KBN V k adjudicate that as a draw? the point being that
: adjudication takes into account best play for both sides not the level of
: the competitors and GM Hort is far better IMHO (or was then) than Fritz is
: today.
I agree. Fortunately there were very few adjudications back in those days,
probably no more than 2-3 per tournament, out of (say) 32+ games (usually
we had 16-24 participants, and played 4-5 rounds).
however there was one direct accusation of cheating with no supporting
evidence of any kind. Directed toward Tony Marsland (Awit). The only
comment from someone 'right there' was from Don Daily, who said "absolutely
not' when I asked him if he thought there was any cheating in that game.
That's quite good enough for me, since it is obvious that the originator of
this thread has a highly warped definition of the word 'cheating' since we are
now off into ' was this adjudication correct' and 'Cray Blitz didn't deserve
to win'. Completely overlooking the fact that we beat the program that he
thought 'should have won'.
Chris Whittington
Robert Hyatt wrote in message <7cbcit$5no$1...@juniper.cis.uab.edu>...
Cray Blitz and Hitech were vying for 'best' status in 1985-86. In June 1986
ICCAJ, p 111, Berliner tried to suggest Hitech was 2352 USCF (maybe 2100
FIDE). Presumably the sort of statistics he was coming up with were the
'scientific' basis of the icca president's claims that the programs were
'near GM' strength.
So, mapping to CrayBlitz, this would suggest that, at the time, the claims
were for 2100-2200 FIDE ELO.
"what we were doing in 1986 is not a lot different from what we are doing
today" and "we were searching around 200K nodes per sec in the middlegame"
indicates you believe Crafty and CrayBlitz to be very similar. Does that
stack up ?
Or that CrayBlitz lost to a 400 nps program in round 2 ?
Doesn't add up, does it ?
>
>: However that's only my HO, but face facts even if it were now determined
>: that the position is a draw, Cray Blitz still won the event and it was
won
>: without cheating. If the Schach people had objected to the adjudication
then
>: was the time to do so, the officials at the event called it a win for
black
>: and that's the way it is.
>: What would you do back in the earlier days when computers couldn't mate
>: without tablebases KBN V k adjudicate that as a draw? the point being
that
>: adjudication takes into account best play for both sides not the level of
>: the competitors and GM Hort is far better IMHO (or was then) than Fritz
is
>: today.
>
>I agree. Fortunately there were very few adjudications back in those days,
>probably no more than 2-3 per tournament, out of (say) 32+ games (usually
>we had 16-24 participants, and played 4-5 rounds).
By the interpretation of the rules, CrayBlitz won. I wasn't invloved, and I
wasn't there, but reading the etxts from that time, I'm forced to the
conclusion that the result and the championship title were very
controversial indeed. This isn't just about the title, it's about peer-group
recognition. And also the decline of the main-frames from 1986 on.
Remembering the battles between Ed and Bob last year and before, and viewing
Bob's repeated accusations that Ed's operator Louwmann cheated regularly,
methinks there is more to all this than meets the eye.
>
>however there was one direct accusation of cheating with no supporting
>evidence of any kind. Directed toward Tony Marsland (Awit). The only
>comment from someone 'right there' was from Don Daily, who said "absolutely
>not' when I asked him if he thought there was any cheating in that game.
Sloan can't do much more than generate suspicion on the Awit game. It's
obviously not very satisfactory if the Arbiter comes along, mentions an
obscure(?) move to win the game, the opponent program crashes, is rebooted
off-line and then returns with this move. Such an event is likely to annoy
anybody in Sloan's position. But he can't use it to accuse of cheating - no
proof. He can point to sloppy conditions, he can point to general sloppy
conditions, and point to the boundless cheating opportunities that arose.
Then he can point to other things, like 8 world titles available at one
tournament and so on.
For me, the conclusion would not be that Marsland cheated, nor that
CrayBlitz shouldn't be World Champion, but that the whole operation was a
kind of lie. That 1985-86 represented a kind of high-water mark for the
people involved. On one side they were playing a disorganised game, on the
other they were making money, fame, career, on the other it was pretend
science, on the other it was allegedly beneficial to mankind (AI). In fact,
this was cash-in time, as eight world titles from one tournament proved.
>That's quite good enough for me, since it is obvious that the originator of
>this thread has a highly warped definition of the word 'cheating' since we
are
>now off into ' was this adjudication correct' and 'Cray Blitz didn't
deserve
>to win'. Completely overlooking the fact that we beat the program that he
>thought 'should have won'.
Quite possibly. And quite possibly he's into sour grapes as you say. I think
that also, as with many others too, he experienced almost instantaneously,
on entering the room of computer chess, a really bad smell. From the smell
came the feeling of cheating and lies. Still lingers.
Chris Whittington
bruce moreland
wrote:
>Did Cray Blitz Really have a win?
>
>This question comes from the 1986 World Computer Chess Championship.
>
>In the following position, with black to move, Cray Blitz said that it
>had run out of computer time on the host computer and therefore
>requested an adjudication.
>
>In the final position is: 8/2p5/2B2k2/p4bp1/3P4/6K1/1P6/8
>
>Black, Cray Blitz, had a pawn more, but it is almost impossible to win
>this position, even for a strong human, much less for a computer.
>
>Grandmaster Hort and International Master Valvo, the arbiters for the
>event, proceeded to spend two hours analyzing this position and
>finally declared it a win for black. Cray Blitz won the next game as
>well and was declared world computer champion on tie-breaks.
>
>As I recall, the winning method found by Hort and Valvo involved
>bringing the king all the way around with Ke7, Kd8, Kc8, Kb8, Ka7,
>Kb6, Ka5 (after the pawn has moved), trading pawns and then bringing
>the king around behind White.
>
>Because of the horizon effect, no computer of that era could have
>found such a solution.
>
>I have never been convinced that black has a win here with best play.
>Can somebody run this through their computer and see what the result
>is?
>
>Remember, if it turns out that this position with best play is a draw
>and not a win for black, then Cray Blitz did not win the 1986 World
>Computer Chess Championship. Cray Blitz had already lost in round two.
>As this was a five round Swiss, a loss here would have eliminated Cray
>Blitz from any contention for a top prize.
>
>I have posted this game at http://www.anusha.com/awit-rex.htm
Here is an excerpt from the tournament rules, published in the
December '85 ICCAJ, page 268.
"7. The Tournament Director has the right to adjudicate a game after
five hours of total clock time. The adjudication will be made on the
premise othat perfect chess will be played by both sides from the
final position."
So first of all it doesn't matter whether the computer can find it in
a hundred years. There's the rule, there are problems with the rule,
but it was published well in advance of the tournament, and both teams
entered knowing the rule.
As to whether the game was adjudicated properly under the rule, who
knows? An IM and a GM looked at the game for a long time and did the
best they could. What else should have been done? Why are you
bringing up this difficult adjudication question 13 years later? Do
you have a refutation of the winning line seen by Valvo and Hort, and
if so, let's see it.
bruce
Robert Hyatt
: Cray Blitz and Hitech were vying for 'best' status in 1985-86. In June 1986
: ICCAJ, p 111, Berliner tried to suggest Hitech was 2352 USCF (maybe 2100
: FIDE). Presumably the sort of statistics he was coming up with were the
: 'scientific' basis of the icca president's claims that the programs were
: 'near GM' strength.
No Idea. HiTech was playing in USCF events and did have a valid USCF rating.
(Although FIDE is not USCF-250 in any category at all, more like USCF-100 or
so, at the 2000-up level)
: So, mapping to CrayBlitz, this would suggest that, at the time, the claims
: were for 2100-2200 FIDE ELO.
I didn't have a feel for "FIDE" ratings at the time, but we did play in
several USCF events and I knew how we did against 2350-type USCF players.
We even played a demo game (40/2hr) at the 1984 US Open against a 2390 USCF
player and won.
: "what we were doing in 1986 is not a lot different from what we are doing
: today" and "we were searching around 200K nodes per sec in the middle game"
: indicates you believe Crafty and CrayBlitz to be very similar. Does that
: stack up ?
: Or that CrayBlitz lost to a 400 nps program in round 2 ?
For a non-bean counter, you sure do 'count beans' don't you? IE suddenly
one game is the defining game for a program? If that is the case, I'm going
to go dig up a under 25 move loss vs Crafty you played (lonnie) at a long
time control. That would then define "your program"??? Or could it be that
maybe you had a bug since you sent him many versions? Or could it be just
that 'occasional bad book move choice'? From _MY_ perspective, I'd suspect
any of the latter _first_ and not say "your program is crap, just look at
this game. How could _any_ decent program play like this?" And in my
case, I had explained that gross evaluation change already, several times.
: Doesn't add up, does it ?
Only when you can't add at all.
Please read this carefully. I get tired of having to repeat it over and
over.
In the summer of 1985, as I was getting ready to move to Birmingham to work
on my PhD. Bert Gower and I played a lot of games at night, between Cray
Blitz (running on a vax at 100 nps) and his Super Constellation chess machine.
We won most games tactically. But we occasionally got into an endgame with
our pawns weakened enough that we would draw or even lose. I made a change
to the evaluation that seriously penalized pawn structure 'holes' and after
playing a hundred games or so we became convinced that we had it working
well as we lost no more games to that supercon.
However, we were testing at 100nps, but we played chess at 50K-200K nodes
per second. At the Denver ACM event in October of 1985 we lost to Bebe,
something we had _never_ done before. And then we lost to Berliner's
new chess machine "HiTech (that was its first ACM event). I wrote it
off to 'folks were catching up with us' and returned to Birmingham and
continued the Ph.D. process. As we prepared for the 1986 WCCC event (about
6 months after the ACM event) we made few changes, because we ran more
tests on the vax and were still happy with the way it played there.
In 1986 at the WCCC event, in round 1 we won easily and were feeling
satisfied. But in round 2, we lost. And I mean we lost, we didn't get
beat. CB refused to push any pawns, and if you look at the game, you will
see just how passive it played as a result. After that loss we _knew_
something was wrong. I was thinking 'some sort of eval bug, or some sort
of serious parallel search bug' and Gower and I went to work testing (we had
18 hours before the next round. We did go back to try an older version of
CB that Harry had on the Livermore machine and it played _far_ differently,
because it was the 1984 version, prior to the 'pawn hole' changes. When we
noticed the difference, the question was 'what is different' and my careful
notes pinpointed that instantly. I commented the 4 lines of pawn hole
evaluation (some clever vector stuff, inside the main loop for pawn evaluation
so it was executed at least once for every pawn on the board) and the program
played like it used to.
We decided to leave everything else alone, but remove those lines, and continue
the tournament. In round 3 it attacked and blew Bebe out. In round 4 it
reached a +2 endgame that it probably would have won had we not run out of
machine time. In round 5 it simply out-played HiTech and won that game as
well.
So yes, it played badly in round 2. The pawn hole was interacting badly with
some other code that understood that holes were bad when they were exploited.
The vax couldn't see deep enough to see the holes being exploited, so our
change prevented the holes and the supercon didn't know how to take advantage
of our passivity. A serious testing bug in using one opponent, I know. But
in 1985 there were few good choices for sparring opponents as all were so much
weaker than CB that the comparison wasn't interesting except for debugging.
On the Cray, it got a penalty for making the hole, _and_ a penalty when it
saw the hole exploited, and it concluded 'ok, this is gross and I can't make
_any_ pawn holes of any kind. If you look at the Bobby game (if I remember
that game properly after 13 years) I think that we reached a position where
we had pawns on the queen-side on their original and Bobby pushed his as
close as possible. And _we_ thought we were winning because he had a
_huge_ number of pawn holes behind his advanced pawns and we had _none_.
So every pawn advance by "you" was good for us, even if you wound up with
3/4 of the board as 'your space' because we just chortled and said 'lookit
all them holes you have' without a clue of how to exploit them.
Now I hope I can avoid typing that again. We had a serious eval problem
(I won't call it a bug, because it did what it was written to do, but on
a machine that could search 5-6 plies deeper the evaluation was simply
wrong). However, rather than picking game 2 for your cute remarks, what
about game 3? Or game 5? Or telling us what _your_ program plays instead
of Bh7 in game 4?
IE it isn't a good idea to throw rocks in a glass house. If you can't do
better, then it doesn't make a lot of sense to criticize another program.
However, to say that CB was "not a good program" is a stupid suggestion,
because anybody can have a lucky game and win one tournament. We won
several, and that pushes the concept of 'luck' way too far.
Unless there is another motive involved...
:>
:>: However that's only my HO, but face facts even if it were now determined
:>: that the position is a draw, Cray Blitz still won the event and it was
: won
:>: without cheating. If the Schach people had objected to the adjudication
: then
:>: was the time to do so, the officials at the event called it a win for
: black
:>: and that's the way it is.
:>: What would you do back in the earlier days when computers couldn't mate
:>: without tablebases KBN V k adjudicate that as a draw? the point being
: that
:>: adjudication takes into account best play for both sides not the level of
:>: the competitors and GM Hort is far better IMHO (or was then) than Fritz
: is
:>: today.
:>
:>I agree. Fortunately there were very few adjudications back in those days,
:>probably no more than 2-3 per tournament, out of (say) 32+ games (usually
:>we had 16-24 participants, and played 4-5 rounds).
: By the interpretation of the rules, CrayBlitz won. I wasn't invloved, and I
: wasn't there, but reading the etxts from that time, I'm forced to the
: conclusion that the result and the championship title were very
: controversial indeed. This isn't just about the title, it's about peer-group
: recognition. And also the decline of the main-frames from 1986 on.
Main-frames didn't 'decline'? Where were you at these events? Deep Thought
became the best program starting in 1987, and was never replaced as the best,
although it lost one game to fritz in 1995 in Hong Kong. But your recollection
about 'mainframes declining' is a tad exaggerated. If I could get a block of
time, I'd be happy to play you a couple of games with that 'declined mainframe'
because if you can't beat Crafty, you certainly don't stand a chance against
'that declined mainframe'. I've played a few games vs 'it' over the past
few years, and Crafty never does very well. Still has a factor of 10+
computing speed advantage over Crafty, unfortunately, even when using the
current quad xeon.
: Remembering the battles between Ed and Bob last year and before, and viewing
: Bob's repeated accusations that Ed's operator Louwmann cheated regularly,
: methinks there is more to all this than meets the eye.
All you have to do is ask around. Tony Scherzer first brought this up to
me. Then others (I won't mention names unless it is necessary.) But the
'move now' thing _absolutely_ happened. Too many of _US_ caught him doing
that. If you like, I can give you a name or two to contact via private
email.
:>
:>however there was one direct accusation of cheating with no supporting
:>evidence of any kind. Directed toward Tony Marsland (Awit). The only
:>comment from someone 'right there' was from Don Daily, who said "absolutely
:>not' when I asked him if he thought there was any cheating in that game.
: Sloan can't do much more than generate suspicion on the Awit game. It's
: obviously not very satisfactory if the Arbiter comes along, mentions an
: obscure(?) move to win the game, the opponent program crashes, is rebooted
: off-line and then returns with this move. Such an event is likely to annoy
: anybody in Sloan's position. But he can't use it to accuse of cheating - no
: proof. He can point to sloppy conditions, he can point to general sloppy
: conditions, and point to the boundless cheating opportunities that arose.
: Then he can point to other things, like 8 world titles available at one
: tournament and so on.
: For me, the conclusion would not be that Marsland cheated, nor that
: CrayBlitz shouldn't be World Champion, but that the whole operation was a
: kind of lie. That 1985-86 represented a kind of high-water mark for the
: people involved. On one side they were playing a disorganised game, on the
: other they were making money, fame, career, on the other it was pretend
: science, on the other it was allegedly beneficial to mankind (AI). In fact,
: this was cash-in time, as eight world titles from one tournament proved.
There were not 8 titles at stake in 1986. There was one. "The 1986 World
Computer Chess Champion". The micros still had their own annual events
that were _thick_ with cheating, collusion, misbehavior, etc. IE you might
compare the WCCC events to the WMCCC events and you come away with the
idea that maybe "we academics" knew how to do it right, while the commercial
micro interests led to fixed games, cheating with hardware, entering multiple
copies of the same program and then resigning the lower-rated program when
the two met. We _never_ had that nonsense going on at a WCCC. We left that
to the 'micro guys'. And at every ACM event Mike would come back thoroughly
disgusted with the behavior at the just finished WCCC. And would coin
nicknames like "terror in trevemunde' (if that is spelled correctly).
:>That's quite good enough for me, since it is obvious that the originator of
:>this thread has a highly warped definition of the word 'cheating' since we
: are
:>now off into ' was this adjudication correct' and 'Cray Blitz didn't
: deserve
:>to win'. Completely overlooking the fact that we beat the program that he
:>thought 'should have won'.
: Quite possibly. And quite possibly he's into sour grapes as you say. I think
: that also, as with many others too, he experienced almost instantaneously,
: on entering the room of computer chess, a really bad smell. From the smell
: came the feeling of cheating and lies. Still lingers.
Perhaps from your perspective. However, in all the events I have attended,
which started with ACM 1975 and excluded only 2 of those, plus all of the
WCCC events held in North America (1977, 1983, 1989) I never saw you at a
one of those. So I'm not sure we have the same 'perspective'. You have
been in the 'micro-debacles' which I have had no experience with until the
last two. I have been involved in the 'big-iron' tournaments where there
were few problems of any kind, and except when be beat HiTech in 1986 there
has never been any accusations of cheating of any kind that I recall.
Maybe the 'big iron' "circle" is a bit cleaner, at that.
Henri H. Arsenault
wrote:
> I have been looking at this position since yesterday. At home I have Fritz
> 5.32 and Rebel 10 both running it. Fritz has KE6 as next at 26 ply. I will
> post more details later.
Are you sure? I have never seen Fritz 5.32 reaching a full-width 26-ply
depth even after running all day. I have never even seen 20.
Henri
Robert Hyatt
: In rec.games.chess.computer Chris Whittington <ch...@cpsoft.demon.co.uk> wrote:
: : Cray Blitz and Hitech were vying for 'best' status in 1985-86. In June 1986
: : ICCAJ, p 111, Berliner tried to suggest Hitech was 2352 USCF (maybe 2100
: : FIDE). Presumably the sort of statistics he was coming up with were the
: : 'scientific' basis of the icca president's claims that the programs were
: : 'near GM' strength.
: No Idea. HiTech was playing in USCF events and did have a valid USCF rating.
: (Although FIDE is not USCF-250 in any category at all, more like USCF-100 or
: so, at the 2000-up level)
Bert Gower called and reminded me of one important piece of data about 'rating'
that I had forgotten. Somewhere in the 1986-1989 time-frame (our machine was
pretty constant during that time so there were no huge jumps in performance for
Cray Blitz) we played a game vs a GM. I don't remember the specifics, but
do remember that 4 GM players (I think they were all GMs) played the 'top 4'
computer programs of the time in a 30/60min type time control. The games
were broadcast in Europe on TV on a Sunday Afternoon. We had an early
deep thought machine, HiTech, Cray Blitz, and I don't remember the last
program but probably NuChess (Slate). We played (and I am not going to get
this spelled correctly, my apologies for doing so but the FIDE web site seems
to continuously be broken so I'm not even going to try it to check) GM Paul
Van den Sterren. If I recall the outcomes, the chiptest/deepthought machine
drew their game, HiTech lost, Cray Blitz won, and I don't recall the 4th
result.
But that was significant, in that it proved that the top programs might have
been 'near' GM level in some ways. Clearly the top 3 programs 'broke even'
with 1 win, 1 loss and 1 draw.
Sorry I don't remember the exact year, but I did so many such demos over the
years that I simply don't recall. I do remember it being on a Sunday afternoon,
I played from here at UAB so it was after 1985 when I came here. It was done
via voice communication (each of the computers had a 'proxy human' at the
match site relaying our moves to the board and relaying our opponent's moves
back to us.
In thinking about that, I supposed we can discount that result too, as
'cheating' was possible. :)
But that 'program' took several GM scalps. In Blitz, at the supercomputer
conference in Albuquerque (where the ACM event was held in 1991 I think) we
had a multi-night battle-royal with GM Ivanov. The first night we shredded
him pretty well. He came back two nights later primed to play and ran into
the same buzz-saw.
For the record, this blitz match was played with 5 minutes on each clock,
with me pressing the clock and moving piece for Cray Blitz. No 'special'
treatment like 5 minutes of cpu time or anything. It looked just like a
normal blitz game. Only funny thing about it was that I drew quite a crowd
that night, and instead of watching the game, they were watching me type, as
I had to enter moves in algebraic (ne4, etc) and take moves from the program
and make them, all the while my clock was running. And I didn't lose a single
game on time. Was a lot of fun. And it showed 'computers are here to stay
and have to be reckoned with, particularly at 5 minute blitz.'
For those curious, I had Cray Blitz set to use 3 secs/move until it had
used one minute of its time, then it dropped to 2 secs/move until it had
used another minute, and then it used 1 second per move for the remainder
of the game. Typical search depths were 8-9 plies in the middlegame.
Bob
Robert Hyatt
: In article <36e92...@nemo.idirect.com>, "ChessAngel" <An...@OfChess.com>
: wrote:
: Henri
If you look at this position, there is little material left. 3-4 pawns and
a bishop for each side. 26 plies is not hard at all if it has been running
for a day. I can get to 20 plies pretty quickly myself, even with
full tablebase probes going on. 11 plies takes 4.5 secs, 12 takes 8.1, 13
takes 13.8, 14 takes 32.1, 15 takes 1:10. Extrapolating, 20 would take
32 times longer (2^5) or about 30 minutes. 26 would take 64 times longer
or just over a day or so total. Sounds about right. And this is with
tablebase probes frying the disk. Turn them off and this drops way down.
(fritz doesn't probe in the search of course, but crafty does).
ChessAngel
Robert Hyatt wrote in message <7cbpn6$97m$2...@juniper.cis.uab.edu>...
>In rec.games.chess.computer Henri H. Arsenault <ars...@phy.ulaval.ca>
wrote:
>: In article <36e92...@nemo.idirect.com>, "ChessAngel"
<An...@OfChess.com>
>: wrote:
>
>:> I have been looking at this position since yesterday. At home I have
Fritz
>:> 5.32 and Rebel 10 both running it. Fritz has KE6 as next at 26 ply. I
will
>:> post more details later.
>
>: Are you sure? I have never seen Fritz 5.32 reaching a full-width 26-ply
>: depth even after running all day. I have never even seen 20.
>
>: Henri
>
>If you look at this position, there is little material left. 3-4 pawns and
>a bishop for each side. 26 plies is not hard at all if it has been running
>for a day. I can get to 20 plies pretty quickly myself, even with
>full tablebase probes going on. 11 plies takes 4.5 secs, 12 takes 8.1, 13
>takes 13.8, 14 takes 32.1, 15 takes 1:10. Extrapolating, 20 would take
>32 times longer (2^5) or about 30 minutes. 26 would take 64 times longer
>or just over a day or so total. Sounds about right. And this is with
>tablebase probes frying the disk. Turn them off and this drops way down.
>(fritz doesn't probe in the search of course, but crafty does).
>
Thanks however Henry is correct. Though your post here Bob on Crafty was
interesting. I do have a crafty by the way, as we probably all do :-)
I had briefly looked prior to leaving and thought 26 he was in fact totally
correct, 20 plies. Morning eyes :-)
I get home, same 20 plies??. But only can give the following displayed info.
1...........Ke6
2. Kf3... Kd6
3. Be8...c5
4. dxc5...Kxc5
5.Bf7.....Kd4
6. Ba2 (-1.25) depth 19/19 12:52:42
So, it hasn't changed it's eval for about 12 hours, I originally set-up
about 24 hours ago 6:00 pm EST.
Robert Hyatt
: Robert Hyatt wrote in message <7cbpn6$97m$2...@juniper.cis.uab.edu>...
Do you have a reasonable sized hash table for that big a search? IE
fritz seems to die when it overruns available memory.
anti...@spam.demon.co.uk
> is somewhat faster than Crafty on a P6/200 from 18 months ago. I don't know
> whether Crafty or CB could have won that game. CB thought it could. Crafty
> today thinks it can.
I haven't got extensive evidence as yet, it will take time to replicate these
results, naturally.
Last night I decided to run the "questionable position" on Fritz using my
Pii-400. To spice things up, I had Black using the Fritz 5.32 engine, and
Crafty 16.1 using the White engine.
The position was left running in TOURNAMENT mode, so the adjudicated position
was the start point for the tournament so to speak. The results indicate that
it was a draw, funnily enough fritz still thinks it is winning (but doesn't
make progress) whilst Crafy has the final position as 0.00 :-)
I plan to run the test a number of times, but I am only doing so because I
think the position looks drawn OTB (IMHO).
Personally, I also think the concept of best-play for the adjudication is fine
for humans but is flawed for Computers. The competition was about which
computer was best, to introduce a factor based on human decisions for the
adjudication does not make sense. For example, we know that two novices won't
play best chess and one or the other may get an unexpected result - the same
applies to "prototype computer programs"; to have someone else look at a
position from a different perspective entirely and then adjudicate on it seems
daft (IMHO).
Anyway, here are last nights score (not checked, so it could be a duff move
has been played - but then that would strengthen the argument that a human
should not adjudicate a computer position).
Who - Cray [D55]
47.Kg3 Ke6 48.Bf3 Kd6 49.Kf2 c5 (Be6) 50.Ke3 (dxc5+) 50...cxd4+ (Be6) 51.Kxd4
g4 (Be6) 52.Bxg4 (Bb7) 52...Bxg4 53.b4 a4 (axb4) 54.Kc3 Bc8 (a3) 55.Kb2 Kc6
(a3+) 56.Ka3 Kb5 (Bd7) 57.Kb2 Be6 (Kxb4) 58.Ka3 Bc4 (Ka6) 59.Kb2 Bf7 (Kxb4)
60.Ka3 Be6 (Ka6) 61.Kb2 Bc4 (Kxb4) 62.Ka3 Bf7 (Ka6) 63.Kb2 Bd5 (Kxb4) 64.Ka3
Be4 (Ka6) 65.Kb2 Bf3 (Kxb4) 66.Kc3 (Ka2) 66...Bd5 (Ka6) 67.Kb2 Bb3 (Kxb4)
68.Kc3 (Ka3) 68...Bg8 (Bd1) 69.Kb2 Bh7 (Kxb4) 70.Kc3 (Ka2) 70...Bg6 (a3)
71.Kb2 Bd3 (Kxb4) 72.Ka3 Bf5 (Be2) 73.Kb2 Be4 (Kxb4) 74.Kc3 (Ka3) 74...Bh7
(Bf3) 75.Kb2 Bf5 (Kxb4) 76.Kc3 (Ka3) 76...Bd7 (Bc8) 77.Kb2 Kb6 (Kxb4) 78.Ka3
(b5) 78...Bb5 (Kc7) 79.Ka2 (Kb2) 79...Kc6 (Kc7) 80.Ka3 (Kb2) 80...Kd5 (Kb7)
81.Ka2 (Kb2) 81...Kd4 (Be2) 82.Ka3 (Kb2) 82...Ke5 83.Kb2 Kf4 (Bc4) 84.Ka3 Ke4
(Bc4) 85.Kb2 Kd3 (Bc4) 86.Kb1 Ke3 (Bc4) 87.Kb2 Kd2 88.Ka2 (Kb1) 88...Kc3 (Bd7)
89.Ka3 Kd4 (Bd7) 90.Kb2 Ke5 (Bd7) 91.Ka3 (Ka2) 91...Kd6 (Bc4) 92.Kb2 (Ka2)
92...Kd7 (Bc4) 93.Ka3 Kc6 (Bc4) 94.Kb2 Kb7 (Bc4) 95.Ka3 (Ka2) 95...Ka6 (Bc4)
96.Kb2 Ka7 (Bc4) 97.Ka3 (Ka2) 97...Kb7 (Bc4) 98.Kb2 Kc8 (Bc4) 99.Ka3 (Ka2)
99...Kd7 (Bc4) 100.Kb2 Kd6 (Bc4) 101.Ka3 Ke6 (Bc4) 102.Kb2 Ke7 (Bc4) 103.Ka3
Kf7 ½-½
--
Adios Amigo
Carl Tillotson
Lancashire Chess Association
homepage: http://www.lancashirechess.demon.co.uk/
Virtual Access 4.50 build 266 (32-bit)
Using Win98
anti...@spam.demon.co.uk
> We decided to leave everything else alone, but remove those lines, and continue
> the tournament. In round 3 it attacked and blew Bebe out. In round 4 it
Isn't that unethical - although maybe in the rules.
What's the difference between me playing Rounds 1 & 2 in the OPEN and then
getting Fischer to play my final three rounds?
I guess you are all going to bleat about "changing parameters etc..." - Fact
remains the program in Round 1 and 2 DID NOT COMPETE in Rounds 3-5. In my book
that is the same as changing the player, although I suspect the rulebook for the
Computer Chess World is different for use mere mortals :-)
Robert Hyatt
: In article <7cbj6v$7n4$1...@juniper.cis.uab.edu>, Robert Hyatt wrote:
:> We decided to leave everything else alone, but remove those lines, and continue
:> the tournament. In round 3 it attacked and blew Bebe out. In round 4 it
: Isn't that unethical - although maybe in the rules.
Absolutely not. Generally 1/4 to 1/2 of the programs get modified between
rounds as problems are seen. Most commonly time allocation is adjusted but
changing things between rounds is perfectly normal.
As far as humans go, have you never studied some opening line between rounds
so you could play it in the next round? Was that unethical?
: What's the difference between me playing Rounds 1 & 2 in the OPEN and then
: getting Fischer to play my final three rounds?
No connection... we used the same computer/program (we changed computers
between rounds 3 and 4 due to our scheduling situation.)
: I guess you are all going to bleat about "changing parameters etc..." - Fact
: remains the program in Round 1 and 2 DID NOT COMPETE in Rounds 3-5. In my book
: that is the same as changing the player, although I suspect the rulebook for the
: Computer Chess World is different for use mere mortals :-)
Nope... _you_ can change between rounds. I have played in several USCF
events, and when I found out my opponent for the next round, I checked
previous round game scores to see what he tended to play, and I'd go back
to my opening books to make sure I was prepared. Or I might even decide to
play something different than normal, based on how he handled that opening
in a previous round. Nothing unusual there at all.
Robert Hyatt
: In article <7cbcit$5no$1...@juniper.cis.uab.edu>, Robert Hyatt wrote:
:> is somewhat faster than Crafty on a P6/200 from 18 months ago. I don't know
:> whether Crafty or CB could have won that game. CB thought it could. Crafty
:> today thinks it can.
: I haven't got extensive evidence as yet, it will take time to replicate these
: results, naturally.
: Last night I decided to run the "questionable position" on Fritz using my
: Pii-400. To spice things up, I had Black using the Fritz 5.32 engine, and
: Crafty 16.1 using the White engine.
: The position was left running in TOURNAMENT mode, so the adjudicated position
: was the start point for the tournament so to speak. The results indicate that
: it was a draw, funnily enough fritz still thinks it is winning (but doesn't
: make progress) whilst Crafy has the final position as 0.00 :-)
: I plan to run the test a number of times, but I am only doing so because I
: think the position looks drawn OTB (IMHO).
: Personally, I also think the concept of best-play for the adjudication is fine
: for humans but is flawed for Computers. The competition was about which
: computer was best, to introduce a factor based on human decisions for the
: adjudication does not make sense. For example, we know that two novices won't
: play best chess and one or the other may get an unexpected result - the same
: applies to "prototype computer programs"; to have someone else look at a
: position from a different perspective entirely and then adjudicate on it seems
: daft (IMHO).
I disagree... because _how_ can you estimate what a computer can do and what
it can't? We had an interesting algorithm for repetition that _always_ made
CB take the repetition or draw score that occurred deepest in the game tree,
with the idea that the longer we put it off, the more likely our opponent would
make a mistake or that we might find a way to a score _better_ than a draw
several moves into the future (this worked well in several games over the
years.)
But _how_ would you try to guess what the computer could/could not do?
that was why the adjudication process was changed many years ago to be based
on perfect play. Because that is a 'concrete' style of adjudication with no
subjectiveness at all.
: Anyway, here are last nights score (not checked, so it could be a duff move
: has been played - but then that would strengthen the argument that a human
: should not adjudicate a computer position).
: Who - Cray [D55]
: 47.Kg3 Ke6 48.Bf3 Kd6 49.Kf2 c5 (Be6) 50.Ke3 (dxc5+) 50...cxd4+ (Be6) 51.Kxd4
: g4 (Be6) 52.Bxg4 (Bb7) 52...Bxg4 53.b4 a4 (axb4) 54.Kc3 Bc8 (a3) 55.Kb2 Kc6
: (a3+) 56.Ka3 Kb5 (Bd7) 57.Kb2 Be6 (Kxb4) 58.Ka3 Bc4 (Ka6) 59.Kb2 Bf7 (Kxb4)
: 60.Ka3 Be6 (Ka6) 61.Kb2 Bc4 (Kxb4) 62.Ka3 Bf7 (Ka6) 63.Kb2 Bd5 (Kxb4) 64.Ka3
: Be4 (Ka6) 65.Kb2 Bf3 (Kxb4) 66.Kc3 (Ka2) 66...Bd5 (Ka6) 67.Kb2 Bb3 (Kxb4)
: 68.Kc3 (Ka3) 68...Bg8 (Bd1) 69.Kb2 Bh7 (Kxb4) 70.Kc3 (Ka2) 70...Bg6 (a3)
: 71.Kb2 Bd3 (Kxb4) 72.Ka3 Bf5 (Be2) 73.Kb2 Be4 (Kxb4) 74.Kc3 (Ka3) 74...Bh7
: (Bf3) 75.Kb2 Bf5 (Kxb4) 76.Kc3 (Ka3) 76...Bd7 (Bc8) 77.Kb2 Kb6 (Kxb4) 78.Ka3
: (b5) 78...Bb5 (Kc7) 79.Ka2 (Kb2) 79...Kc6 (Kc7) 80.Ka3 (Kb2) 80...Kd5 (Kb7)
: 81.Ka2 (Kb2) 81...Kd4 (Be2) 82.Ka3 (Kb2) 82...Ke5 83.Kb2 Kf4 (Bc4) 84.Ka3 Ke4
: (Bc4) 85.Kb2 Kd3 (Bc4) 86.Kb1 Ke3 (Bc4) 87.Kb2 Kd2 88.Ka2 (Kb1) 88...Kc3 (Bd7)
: 89.Ka3 Kd4 (Bd7) 90.Kb2 Ke5 (Bd7) 91.Ka3 (Ka2) 91...Kd6 (Bc4) 92.Kb2 (Ka2)
: 92...Kd7 (Bc4) 93.Ka3 Kc6 (Bc4) 94.Kb2 Kb7 (Bc4) 95.Ka3 (Ka2) 95...Ka6 (Bc4)
: 96.Kb2 Ka7 (Bc4) 97.Ka3 (Ka2) 97...Kb7 (Bc4) 98.Kb2 Kc8 (Bc4) 99.Ka3 (Ka2)
: 99...Kd7 (Bc4) 100.Kb2 Kd6 (Bc4) 101.Ka3 Ke6 (Bc4) 102.Kb2 Ke7 (Bc4) 103.Ka3
: Kf7 ½-½
: --
: Adios Amigo
: Carl Tillotson
: Lancashire Chess Association
: homepage: http://www.lancashirechess.demon.co.uk/
: Virtual Access 4.50 build 266 (32-bit)
: Using Win98
--
Chris Whittington
Robert Hyatt wrote in message <7cduh0$sg9$1...@juniper.cis.uab.edu>...
>In rec.games.chess.computer anti...@spam.demon.co.uk wrote:
>: In article <7cbj6v$7n4$1...@juniper.cis.uab.edu>, Robert Hyatt wrote:
>
continue
>:> the tournament. In round 3 it attacked and blew Bebe out. In round 4
it
>
>
>Absolutely not. Generally 1/4 to 1/2 of the programs get modified between
>rounds as problems are seen. Most commonly time allocation is adjusted but
>changing things between rounds is perfectly normal.
>
>As far as humans go, have you never studied some opening line between
rounds
>so you could play it in the next round? Was that unethical?
>
>
>: What's the difference between me playing Rounds 1 & 2 in the OPEN and
then
>: getting Fischer to play my final three rounds?
>
>No connection... we used the same computer/program (we changed computers
>between rounds 3 and 4 due to our scheduling situation.)
What's wrong isn't all these special rules here there and everywhere, and
changing programs and this and that; it's the utter failure to make any of
it clear to newbies, or even to habitues. If you've been before to a comp
chess tournament, read the rules in the journals, listened to the arguments
about this and that, know that this program has some special dispensation to
do this, and another to do that, then fine. All is clear and obvious. If
sometimes dubious, but that's another matter.
But, if you're a newbie, or a spectator, you've not seen the rules (you
don't get a copy as a competitor - it is assumed you read the journal,
except you may not have done), you don't know that Hyatt can link by voice
(not in the rules), or that someone can adjust their clock in ways strange
(not in the rules), or any one of a zillion little things (not in the
rules), that DONT GET PUBLICISED - then people can get angry.
It's a sloppy, disorganised tournamant that they put on. Simple as that. And
the reason that so many people get seriously pissed off. Add to it a few
program crashes, and a few dubious personalities, and .......
Chris Whittington
>
>: I guess you are all going to bleat about "changing parameters etc..." -
Fact
>: remains the program in Round 1 and 2 DID NOT COMPETE in Rounds 3-5. In my
book
>: that is the same as changing the player, although I suspect the rulebook
for the
>: Computer Chess World is different for use mere mortals :-)
>
>Nope... _you_ can change between rounds. I have played in several USCF
>events, and when I found out my opponent for the next round, I checked
>previous round game scores to see what he tended to play, and I'd go back
>to my opening books to make sure I was prepared. Or I might even decide to
>play something different than normal, based on how he handled that opening
>in a previous round. Nothing unusual there at all.
>
>
>
>
Robert Hyatt
: Robert Hyatt wrote in message <7cduh0$sg9$1...@juniper.cis.uab.edu>...
:>In rec.games.chess.computer anti...@spam.demon.co.uk wrote:
:>: In article <7cbj6v$7n4$1...@juniper.cis.uab.edu>, Robert Hyatt wrote:
:>
:>:> We decided to leave everything else alone, but remove those lines, and
: continue
:>:> the tournament. In round 3 it attacked and blew Bebe out. In round 4
: it
:>
:>: Isn't that unethical - although maybe in the rules.
:>
:>Absolutely not. Generally 1/4 to 1/2 of the programs get modified between
:>rounds as problems are seen. Most commonly time allocation is adjusted but
:>changing things between rounds is perfectly normal.
:>
:>As far as humans go, have you never studied some opening line between
: rounds
:>so you could play it in the next round? Was that unethical?
:>
:>
:>: What's the difference between me playing Rounds 1 & 2 in the OPEN and
: then
:>: getting Fischer to play my final three rounds?
:>
:>No connection... we used the same computer/program (we changed computers
:>between rounds 3 and 4 due to our scheduling situation.)
: What's wrong isn't all these special rules here there and everywhere, and
: changing programs and this and that; it's the utter failure to make any of
: it clear to newbies, or even to habitues. If you've been before to a comp
: chess tournament, read the rules in the journals, listened to the arguments
: about this and that, know that this program has some special dispensation to
: do this, and another to do that, then fine. All is clear and obvious. If
: sometimes dubious, but that's another matter.
: But, if you're a newbie, or a spectator, you've not seen the rules (you
: don't get a copy as a competitor - it is assumed you read the journal,
: except you may not have done), you don't know that Hyatt can link by voice
: (not in the rules), or that someone can adjust their clock in ways strange
: (not in the rules), or any one of a zillion little things (not in the
: rules), that DONT GET PUBLICISED - then people can get angry.
This is wrong. At _every_ ACM event I have attended, and at _every_ WCCC
event I have attended, when you walk in the door, spectator, competitor,
or whatever, you find a stack of booklets waiting. These booklets have
everything you need to know. A copy of the rules. Information about each
participant. The complete crosstable from the last event. All the games
from the last event. Plus other interesting things like times for the
rounds, when the player's meeting happens (which spectators can attend, btw)
and also info about special events (like IM Mike Valvo's blindfold match
against 6 or 8 programs at one tournament, or a panel discussion on computer
chess, or a computer chess paper session, or whatever.)
Information has _always_ been available. And at the players meeting we
always discussed the special cases. IE Cray Blitz can't play past 1:30am
on Sunday, Monday and Tuesday, but we have the machine from the start of
round 1 until 1:30 (end of round 2) on Sunday so the first round has no
time limit other than that imposed by the start of round 2. Other people
would explain odd situations they had, such as "I have a machine from 7pm
to 10pm, but at 10pm I have to change to a second machine because the first
goes down for PM at 10. There were usually a fair number of 'special cases'
that were discussed and resolved at the players meetings, including any
questions/concerns about the wording of a recent rule change or whatever.
When I went to my first ACM event (I didn't know there were competitions
until I attended the 1975 ACM event with someone presenting a paper on a
project we worked on together) I found this sort of information, which is
how I found out where the next tournament was being held and so forth.
: It's a sloppy, disorganised tournamant that they put on. Simple as that. And
: the reason that so many people get seriously pissed off. Add to it a few
: program crashes, and a few dubious personalities, and .......
That's your 'anti-ICCA' side coming out. These tournaments have _never_ been
sloppy. They _have_ been treated as 'research events' and they made every
attempt to accomodate everyone's requirements for participation. The only
'sloppy' events were the WMCCC events where the commercial competitors were
so interested in winning that they cheated with hardware, intentionally threw
games to fix final standings, and so forth. And the ICCA found it quite
difficult to solve all the problems that 'commercial interests' created.
But for the 'non-commercial' events, everything went very smoothly, everyone
had a good time (except for a couple that were paranoid, and maybe one that
made statements that his program couldn't live up to). To paint these events
in any other light is to do them a grave injustice. If so many of the old
competitors had not been driven away from here in years past, you'd find out
that most thought these events were the highlight of each year in computer
chess. I know I did, whether I won or lost. I _never_ had a bad time.
bruce draney
crossposted to 4 chess newsgroups?
Could you please put it in one or at most two newsgroups?
Would everyone involved in the discussion trim the headers please?
Thanks, and best regards,
Bruce
Albert Silver
Robert Hyatt <hy...@crafty.cis.uab.edu> wrote:
> In rec.games.chess.computer Chris Whittington <ch...@cpsoft.demon.co.uk>
wrote:
>
<snip>
> : . There is no way it was seeing as far as passed pawn
> : conversion. It would possibly have had square of the pawn code (freepawn
> : able to run, and the opponent king not able to catch it), but this would
> : have been irrelevant to the evaluation with bishops on the board.
>
> It didn't see any promotion.
>
> : For several moves prior to the adjudication position, CrayBlitz was going
> : backwards, ending up defending its g6 pawn with its bishop on h7. I thought,
> : positionally, its opponent was playing quite nicely. CrayBlitz however, was
> : concentrating on one thing and one thing only, hanging on to its extra
> : pawns. And tying itself up in knots in the process.
>
> : I think, had the square i8 been available to it, CrayBlitz would have moved
> : its bishop there as well, in an attempt to cover the g6 pawn; and we'ld have
> : seen some more mindless repetitions of moves until CrayBlitz realised it was
> : only heading for a draw.
>
> That is ok. Because we had an interesting 'draw score' algorithm that was
> published. Because for us, 'all draws were not created equal' and we always
> took the draw that occurred farthest into the future, which generally would
> pull us down a path that either let the opponent make a mistake, or which
> would eventually let us see far enough to see how to win.
>
> Nothing says you have to win the game as quickly as possible. And in those
> positions where you can't see the win, if you avoid the draw, you can
> occasionally stumble into the win. And the algorithm worked pretty well in
> several games over the years.
Indeed, I would be surprised to find a tournament player to whom that has not
arrived. Examples abound at every level of chess. In Tim Krabbe's page of
extraordinary moves, one finds one game where prior to a fantastic sac in the
endgame, both players had been pretty much repeating moves. Just before being
confronted with the draw, White probably stumbled on the winning move.
Albert Silver
<snip>
> Nope... it was over in 1986. Still is 'over'.
>
> It doesn't really matter to me what you think about the game. The
> original thread by Sam Sloan talked about 'cheating on the top boards'.
> Now we have moved to attacking the TD and a GM for not adjudicating a
> game properly without any notion that the game isn't won as they said.
> And there has been no connection established between the adjudication
> and the notion of 'cheating'.
>
> There is some other 'motive' here. I assume it will become apparent as
> time wears on.
>
> --
> 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
>
-----------== Posted via Deja News, The Discussion Network ==----------
http://www.dejanews.com/ Search, Read, Discuss, or Start Your Own
anti...@spam.demon.co.uk
> I disagree... because _how_ can you estimate what a computer can do and what
> it can't? We had an interesting algorithm for repetition that _always_ made
Exactly my point. You can't determine what a computer can do or what it can't do.
Therefore in a "competition" where reputations are made/lost based on the results
of the tournament, it is plain that to introduce an element of "human
intervention" will serve only to muddy the waters.
The only way to determine whether a computer can win a position is to let it. Of
course, I accept your argument that this wasn't possible 13 years ago due to
"machine rental time" - did you never get together to settle the argument once
and for all and run the adjudicated position through the computers later?
> CB take the repetition or draw score that occurred deepest in the game tree,
> with the idea that the longer we put it off, the more likely our opponent would
> make a mistake or that we might find a way to a score _better_ than a draw
> several moves into the future (this worked well in several games over the
> years.)
This is a very sophisticated algorithm, is it well known? I accept it is a clever
way of handling the position - how did you avoid problems with the "horizon
effect" ?
> But _how_ would you try to guess what the computer could/could not do?
> that was why the adjudication process was changed many years ago to be based
> on perfect play. Because that is a 'concrete' style of adjudication with no
> subjectiveness at all.
"Perfect Play" is subjective - you are relying on the opinion of "experts" to
determine "best play". With due respect to the adjudicators at the time, how
could THEY be sure it was best play? Is it not possible for them to have missed
the correct continuation?
"Best Play" is a misnomer, since chess is about the "second to last mistake" :-)
anti...@spam.demon.co.uk
> Absolutely not. Generally 1/4 to 1/2 of the programs get modified between
> rounds as problems are seen. Most commonly time allocation is adjusted but
> changing things between rounds is perfectly normal.
Sorry I can't buy this, you are still changing the program to do something it
wasn't programmed to do in the first round or so. Surely to be fair to ALL
competitors the SAME PROGRAM should be used for all rounds?
> As far as humans go, have you never studied some opening line between rounds
> so you could play it in the next round? Was that unethical?
No, but this is akin to introducing a "learning factor" in the program. All
that a human is doing is "learning" from the experience - you are not
replacing the human with another.
Agreed, we can talk all day about semantics, but the fact remains the program
used in Rounds 1 and 2 were not the same as used in Rounds 3-5. You may say
that this is the same as a "human" player reading up latest theory in between
games, I would say it is the same as changing the player and putting a
different one in place. The fact that the "program" performed considerably
better adds to my argument (IMHO).
Of course, you will defend this practice because it has been around for many
years. I, as a mere observer looking in, believe that it is not quite right -
but still it is only my opinion.
Robert Hyatt
: In article <7cduh0$sg9$1...@juniper.cis.uab.edu>, Robert Hyatt wrote:
:> Absolutely not. Generally 1/4 to 1/2 of the programs get modified between
:> rounds as problems are seen. Most commonly time allocation is adjusted but
:> changing things between rounds is perfectly normal.
: Sorry I can't buy this, you are still changing the program to do something it
: wasn't programmed to do in the first round or so. Surely to be fair to ALL
: competitors the SAME PROGRAM should be used for all rounds?
:
As I said, "fair" means "by the rules". And the rules specifically forbade
changes _during a game_ but not _between_ games. So I fail to see the use
of the word "fair" here being correct.
:> As far as humans go, have you never studied some opening line between rounds
:> so you could play it in the next round? Was that unethical?
: No, but this is akin to introducing a "learning factor" in the program. All
: that a human is doing is "learning" from the experience - you are not
: replacing the human with another.
Neither is modifying a couple of lines of a program "replacing that program
with another". When the total is > 40,000 lines, then changing 1 line is
a tiny fraction.
: Agreed, we can talk all day about semantics, but the fact remains the program
: used in Rounds 1 and 2 were not the same as used in Rounds 3-5. You may say
: that this is the same as a "human" player reading up latest theory in between
: games, I would say it is the same as changing the player and putting a
: different one in place. The fact that the "program" performed considerably
: better adds to my argument (IMHO).
: Of course, you will defend this practice because it has been around for many
: years. I, as a mere observer looking in, believe that it is not quite right -
: but still it is only my opinion.
Totally your opinion... and nothing wrong with having one. However, a program
has _never_ been considered a 'static entity'. Those kinds of programs are
burned into proms or FPGA's where they can't be changed. The original idea
of a 'stored program computer' was to allow these 'changes'. And in the case
of chess, I don't see the issue, any more than I see the issue with two humans
discussing something about a game after the fact and one learning from the
other. one of the ways a program 'learns' is via modifications by the
programmer. Computers are not 'self-aware' yet, and are not going to
modify themselves when something goes wrong, since they don't 'know' something
went wrong.
: --
: Adios Amigo
: Carl Tillotson
: Lancashire Chess Association
: homepage: http://www.lancashirechess.demon.co.uk/
: Virtual Access 4.50 build 266 (32-bit)
: Using Win98
--
Robert Hyatt
: In article <7cduna$sg9$2...@juniper.cis.uab.edu>, Robert Hyatt wrote:
:> I disagree... because _how_ can you estimate what a computer can do and what
:> it can't? We had an interesting algorithm for repetition that _always_ made
: Exactly my point. You can't determine what a computer can do or what it can't do.
: Therefore in a "competition" where reputations are made/lost based on the results
: of the tournament, it is plain that to introduce an element of "human
: intervention" will serve only to muddy the waters.
: The only way to determine whether a computer can win a position is to let it. Of
: course, I accept your argument that this wasn't possible 13 years ago due to
: "machine rental time" - did you never get together to settle the argument once
: and for all and run the adjudicated position through the computers later?
No. The next time we would have access to a Cray would be the _next_ year
at the tournament. So this never happened. And it didn't matter anyway since
the result was recorded. We lost games by adjudication, we won games by
adjudication, and we drew games by adjudication. We never complained about
any result, since the rules were quite clear.
:> CB take the repetition or draw score that occurred deepest in the game tree,
:> with the idea that the longer we put it off, the more likely our opponent would
:> make a mistake or that we might find a way to a score _better_ than a draw
:> several moves into the future (this worked well in several games over the
:> years.)
: This is a very sophisticated algorithm, is it well known? I accept it is a clever
: way of handling the position - how did you avoid problems with the "horizon
: effect" ?
:
Best reference is "The Cray Blitz draw heuristic" published in the ICCA
journal. For CB, we didn't have a "horizon effect" to mention, any more than
I see this as a problem in Crafty today. It searched too deeply, with too
many search extensions. That went away when we moved to a super-fast
computer.
:> But _how_ would you try to guess what the computer could/could not do?
:> that was why the adjudication process was changed many years ago to be based
:> on perfect play. Because that is a 'concrete' style of adjudication with no
:> subjectiveness at all.
: "Perfect Play" is subjective - you are relying on the opinion of "experts" to
: determine "best play". With due respect to the adjudicators at the time, how
: could THEY be sure it was best play? Is it not possible for them to have missed
: the correct continuation?
: "Best Play" is a misnomer, since chess is about the "second to last mistake" :-)
Tournament games have been adjudicated since the first tournament, no doubt.
And mistakes can happen. But when you have more than one IM, plus at least
one GM (usually more are handy), plus computer programs handy, mistakes are
not common. And remember, _we the programmers_ agreed on those rules before
we started any tournament. Because there was no other viable solution since
some games can not be played to mate. What do you do when round N-1 has not
finished and it is time to start round N? Adjudication is an essential part
of any tournament, particularly when more than one round per day is played,
or in our case, when machine time was limited.
Henri H. Arsenault
<hy...@crafty.cis.uab.edu> wrote:
> Best reference is "The Cray Blitz draw heuristic" published in the ICCA
> journal. For CB, we didn't have a "horizon effect" to mention, any more than
> I see this as a problem in Crafty today. It searched too deeply, with too
> many search extensions. That went away when we moved to a super-fast
> computer.
>
Interesting, since this implies that if one looks deep enough, there is no
point in looking any deeper since there is no better move than what has
been found. This is opposite to what is suggested (but not proven) by the
published experiments by you and others regarding "new best moves", and it
again raises the question of how deep must one look in order to achieve
this (if there is such a thing, it is probably at least 18-ply).
Henri
Henri H. Arsenault
> Agreed, we can talk all day about semantics, but the fact remains the program
> used in Rounds 1 and 2 were not the same as used in Rounds 3-5. You may say
> that this is the same as a "human" player reading up latest theory in between
> games, I would say it is the same as changing the player and putting a
> different one in place. The fact that the "program" performed considerably
> better adds to my argument (IMHO).
>
> Of course, you will defend this practice because it has been around for many
> years. I, as a mere observer looking in, believe that it is not quite right -
> but still it is only my opinion.
>
The reason tht it has been around for many years is probably that it is
impossible to verify if a program has been changed between games;
personally I don't see any difference between this and "consultation"
between grandmasters between adjourned games. There is little point in
making rules that cannot be verified.
Henri
Robert Hyatt
: <hy...@crafty.cis.uab.edu> wrote:
: Henri
I'm not sure I would go that far. It just means that given two moves,
one that repeats the position in 3 moves, one that repeats in 10 moves,
we _always_ took the repeat-in-10 move, hoping that the opponent would
make a mistake, or we might find a way to avoid repeating and actually
win once we got deep enough.
anti...@spam.demon.co.uk
> Neither is modifying a couple of lines of a program "replacing that program
> with another". When the total is > 40,000 lines, then changing 1 line is
> a tiny fraction.
Yet those four lines were the difference between winning and losing :-)