A diary on 3-Hirn chess
Ingo Althoefer
3-Hirn is the german name for a man-machine symbiosis playing chess.
3-Hirn consists of two different chess computers (or different chess programs)
and one human chess player, the coordinator.
When 3-Hirn has to make a move the coordinator starts both computers.
He watches at their search processes. In an appropriate moment he stops
them again:
If now both computers propose the same move this move has to be executed.
But if Computer I proposes move x and Computer II proposes another move y
the coordinator has the final choice between x and y. However, the
coordinator is not allowed to execute a third move z.
The coordinator is also allowed to organize the permanent brain of both
computers.
I started my 3-Hirn experiments early in 1985:
I had always been interested in computer chess but never found the time to
write a program of my own. In 85 my naive but simple idea was to take two good
chess programs and to write a coordinator p r o g r a m which "simply"
makes the final decision between the proposals of these two good programs.
But before writing this coordinating program I wanted to make a first test
with a human coordinator to see if it is in general possible to increase
the playing strength by using a coordinator ...
In the meantime ( 12 years ! ) I have learned a lot. Especially I am no longer
sure if there is a way to improve two different chess computers by help of an
automatic coordinator. But another thing I have learned as well:
3-Hirn with two commercial chess computers and a human coordinator is able
to play chess on a very high level and the games are completely different
from typical computer vs human games ... and it is absolutely fun to be the
coordinator in a strong 3-Hirn.
In earlier years I, as a human, had also played normal tournament chess. My
rating never was above 2000 ( the top was in the year 1980 ) and never went
below 1900 until I stopped tournament play completely in 1993. But my
knowledge about chess computers and their operating modes helped me to
achieve play on Grandmaster level in the "3-Hirn mode".
In the 12 years from 1985 to now I was the 3-Hirn coordinator in about 130
chess games with human players or computers under tournament conditions. Here
"tournament conditions" mean that each side had at least 120 minutes for the
whole game. I used products of 8 programmers and programming teams:
Thomas Nitsche , Dave Kittinger , Ed Schroeder ,
Dan & Kate Spracklen , Richard Lang , Johan de
Koning , Frans Morsch & Mathias Feist , Marty Hirsch ;
Thanks to all of them for their nice products !
The weakest computer in my 3-Hirn experiments had a rating of about 1500 in
the SSDF-list ( Mephisto III ), and the best one a value clearly above 2400
( Genius 3 on a Pentium 120 ).
In 1997 I intend to write a book on 3-Hirn. Part of this book shall be a
diary on my 3-Hirn experiences: on the nicest games, the greatest victories,
the most terrible losses, the most funny anecdotes, the psychological
processes, and on many little details I learned during the games.
I want to put about 30 of the games in this diary - and I want to present
these 30 games together with some annotations and comments in rgcc, maybe
one per week or so. Comments are very welcome, especially also analyses of
the games by/with current programs.
The most helpful commentators ( taken the average over the 30 games ) will
receive as a little present free copies of the book when it will have
appeared.
Let us start immediately with the very first 3-Hirn game I played. Computers
were Mephisto II and Mephisto III, both programmed by Thomas Nitsche. They
were not "top of the pop" in March 1985 when the game was played. But in
those days my intention was simply to show that a strong coordinator might
improve the strength of two different ( weak ) chess computers. One more reason
for using exactly these machines was that they were very handy in comparison
with other products. ( Remember: I had to put always two chess computers
in my bag, when I travelled to the playing places. ) Although
both programs were written by the same author their ways of "thinking" and
also their styles of play were very different: Mephisto II was rather
solid whereas Mephisto III played a very speculative chess ( perhaps you
may call it the "grandmodule" of Chess System Tal ... ). The opponent was a
young clubmate with a rating of about 1850 when the game was played.
Both Mephisto II and III had ratings about 1500 in the SSDF lists.
25.03.1985
White: 3-Hirn ( Mephisto II, Mephisto III, Ingo Althoefer )
Black: Thomas Hanf
Times: 120 minutes for 40 moves, afterwards 60 minutes for every 20 moves;
1. e4 e6 2. d4 d5 3. Nc3 Bb4 4. e5 c5 5. a3 cxd4 6. axb4 dxc3 7. Qg4 Ne7
8. Qxg7 Rg8 9. Qxh7 Qc7 10. f4 Nbc6 11. bxc3 Nxe5 12. Ne2 N5g6 13. Bd2 Bd7
14. h3 Bb5 15. Nd4 Bxf1 16. Rxf1 Nf8 17. Nb5 Qd7 18. Qd3 a6 19. g3 Rc8 20.
Nd4 e5 21. fxe5 Qxh3 22. Bf4 Ne6 23. Nxe6 fxe6 24. b5 Ra8 25. bxa6 bxa6
26. Rxa6 Rb8 27. Qd1 Rxg3 28. Bxg3 Qxg3+ 29. Kd2 Qxe5 30. Qg4 Rc8 31. Qa4+
Kd8 32. Rf8+ Kc7 33. Ra7+ Kb8 34. Ra8+ 1 - 0 ( Black resigned )
Times used: White 88 minutes Black 42 minutes
Comments are welcome.
Ingo Althoefer.
Cpsoft
Ingo Althoefer <alth...@pdec01.uucp> wrote in article <5a145t$a...@fsuj19.rz.uni-jena.de>...
I have to admit to being utterly fascinated by this 3-hirn concept, and have
10 million questions to ask you. In no particular order:
1. Have you ever considered the possibility of taking on Kasparov in this
mode, or has anyone suggested it to you ? Because I think this mode, maybe
with 3 machines, would give very interesting games. Maybe with (and no
disrespect to you) with a strong GM coordinator ..... ?
2. What is your favourite/best combination of programs ?
3. I note you are from old DDR-Jena. Did the 1991 events make things different
for you ?
4. Have you found anything about the combination program types ?
Best with two strongest programs, or with a maniac and a stable program, or
with two brute forcers/materialists or what ?
5. I gave you an old CSTal to try, but you said "although it plays some
occasional stunning moves, it was not consistent enough". Can you elaborate ?
6. And, and, and ....... :)
Chris Whittington
2.
>
> Ingo Althoefer.
>
brucemo
Ingo Althoefer wrote:
> 3-Hirn with two commercial chess computers and a human coordinator is able
> to play chess on a very high level and the games are completely different
> from typical computer vs human games ... and it is absolutely fun to be the
> coordinator in a strong 3-Hirn.
What happens when you use a coin-flip as your coordinator?
bruce
Komputer Korner
Ingo Althoefer wrote:
>
> 3-Hirn is the german name for a man-machine symbiosis playing chess.
> 3-Hirn consists of two different chess computers (or different chess programs)
> and one human chess player, the coordinator.
>
> When 3-Hirn has to make a move the coordinator starts both computers.
> He watches at their search processes. In an appropriate moment he stops
> them again:
>
> If now both computers propose the same move this move has to be executed.
> But if Computer I proposes move x and Computer II proposes another move y
> the coordinator has the final choice between x and y. However, the
> coordinator is not allowed to execute a third move z.
>
snipped
>
> Comments are welcome.
>
> Ingo Althoefer.
Finally people will begin to understand how the level of correspondence
chess has improved immensely during the last couple of years. Imagine
a correspondence player who "extends" the 3 HIRN concept to 10
machines all running on Pentium 200's or Pentium Pro's!!!! It doesn't
seem fair does it? I quit correspondence when I realized that in order
to compete you had to spend 20 hours a day analyzing.
--
Komputer Korner
The komputer that kouldn't keep a password safe from
prying eyes, kouldn't kompute the square root of 36^n,
kouldn't find the real motive in ChessBase and missed
the real learning feature of Nimzo.
Ingo Althoefer
In article <32C471...@nwlink.com>, brucemo <bru...@nwlink.com> wrote:
>Ingo Althoefer wrote:
>
>> 3-Hirn with two commercial chess computers and a human coordinator is able
>> to play chess on a very high level and the games are completely different
>> from typical computer vs human games...and it is absolutely fun to be the
>> coordinator in a strong 3-Hirn.
>
>What happens when you use a coin-flip as your coordinator?
>
>bruce
I did such a coin flipping experiment mannualy in 1988 with Rebel 5
and Novag Forte A ( both on C6502 processors ). Switching of permanent brains
the flipping side played 16 games against Rebel 5 ( score 8:8 ) and another
16 games against Forte A ( score 8.5 : 7.5 for flipper ).
Before this experiment I had assumed that coinflipping looses performance.
Maybe this is still true when programs are more specialized.
Ingo Althoefer.
Ingo Althoefer
In article <32C4B7...@netcom.ca>,
Komputer Korner <kor...@netcom.ca> wrote:
>
>Finally people will begin to understand how the level of correspondence
>chess has improved immensely during the last couple of years. Imagine
>a correspondence player who "extends" the 3 HIRN concept to 10
>machines all running on Pentium 200's or Pentium Pro's!!!! It doesn't
>seem fair does it? I quit correspondence when I realized that in order
>to compete you had to spend 20 hours a day analyzing.
If you play correspondence chess with help of computers you can do much
better than using the 3-Hirn approach. 3-Hirn gives you "only" about 200 Elo
points compared with the computers playing alone.
Using computers in less restrictive ways allows for much more.
Ingo Althoefer.
PS: In the German correspondence chess organisation ( BDF ) it is explicitely
stated that players are allowed to use computers.
Michel van der List
In article <5a145t$a...@fsuj19.rz.uni-jena.de>,
Ingo Althoefer <alth...@pdec01.uucp> wrote:
>[...]
>Although both programs were written by the same author their
>ways of "thinking" and also their styles of play were very
>different: Mephisto II was rather solid whereas Mephisto III
>played a very speculative chess
>
>[...]
>
>25.03.1985
>White: 3-Hirn ( Mephisto II, Mephisto III, Ingo Althoefer )
>Black: Thomas Hanf
>Times: 120 minutes for 40 moves, afterwards 60 minutes for every 20 moves;
>
>[...]
>
>Comments are welcome.
I think it would be really nice to know which program
recommended what move if there was a difference. Especially as
your knowledge of this process improved...
Michel.
--
-------------------------------------------------------------------
Michel van der List SmithKline Beecham Pharmaceuticals R & D
vander...@sbphrd.com UW2230, 709 Swedeland Road, P.O. Box 1539
(610) 270-4525 King of Prussia, PA 19406-0939
chessman
On 28 Dec 1996 10:59:15 GMT, alth...@pdec01.uucp (Ingo Althoefer)
wrote:
snip
>
>I did such a coin flipping experiment mannualy in 1988 with Rebel 5
>and Novag Forte A ( both on C6502 processors ). Switching of permanent brains
>the flipping side played 16 games against Rebel 5 ( score 8:8 ) and another
>16 games against Forte A ( score 8.5 : 7.5 for flipper ).
>
>Before this experiment I had assumed that coinflipping looses performance.
>Maybe this is still true when programs are more specialized.
>
>Ingo Althoefer.
Rebel 5 ??....in 1988...I doubt it..Ed could you confirm if Rebel 5
was out in 1988...I don't remember Rebel 5...but I may be wrong..
Ingo Althoefer
In article <32c55fef....@netnews.voicenet.com>,
chessman <ches...@voicenet.com> wrote:
>Rebel 5 ??....in 1988...I doubt it..Ed could you confirm if Rebel 5
>was out in 1988...I don't remember Rebel 5...but I may be wrong..
Mephisto Rebel 5 was the first commercial chess computer of Ed Schroeder.
Ed Schroeders program Rebel had been playing in the World Computer Chess
Championship 1986 in Cologne ( almost winning the title already that year
and not only 6 years later in Madrid ). Later in 1986 Rebel 5, based on the
Cologne program, appeared on the market.
Afterwards Ed Schroeder used other names for his programs: MM 4 and Gideon ..
Finally, in 92 or 93 he returned to the name Rebel, "starting" with version 6.
Ingo Althoefer.
Ingo Althoefer
Michel van der List <vander...@sbphrd.com> wrote:
>I think it would be really nice to know which program
>recommended what move if there was a difference. Especially as
>your knowledge of this process improved...
>
>Michel.
For many of the 3-Hirn games I did not record which computer proposed which
move: The coordinator has to operate 2 computers manualy ( especially also
taking back those moves which were not selected ), and this makes already
some stress; so often I was not in the mode to protocol all these data.
A nice thing would be some auto-protocol device for the coordinator but I
do / did not have one.
One thing I learned over the years is that it is not a good idea for the
coordinator to try to be balanced: There have been indeed games where I took
only proposals of program I and none of program II ( except the moves with
identical proposals ).
In typical games programs of equal strength propose different moves in about
40 percent of the positions.
Ingo Althoefer.
Ingo Althoefer
Ingo Althoefer
In article <32C471...@nwlink.com>, brucemo <bru...@nwlink.com> wrote:
>Ingo Althoefer wrote:
>
>> 3-Hirn with two commercial chess computers and a human coordinator is able
>> to play chess on a very high level and the games are completely different
>> from typical computer vs human games...and it is absolutely fun to be the
>> coordinator in a strong 3-Hirn.
>
>What happens when you use a coin-flip as your coordinator?
>
>bruce
I did such a coin flipping experiment mannualy in 1988 with Rebel 5
chessman
On 28 Dec 1996 10:59:15 GMT, alth...@pdec01.uucp (Ingo Althoefer)
wrote:
snip
>
>I did such a coin flipping experiment mannualy in 1988 with Rebel 5
>and Novag Forte A ( both on C6502 processors ). Switching of permanent brains
>the flipping side played 16 games against Rebel 5 ( score 8:8 ) and another
>16 games against Forte A ( score 8.5 : 7.5 for flipper ).
>
>Before this experiment I had assumed that coinflipping looses performance.
>Maybe this is still true when programs are more specialized.
>
>Ingo Althoefer.
Ingo Althoefer
>was out in 1988...I don't remember Rebel 5...but I may be wrong..
Mephisto Rebel 5 was the first commercial chess computer of Ed Schroeder.
Ingo Althoefer
Komputer Korner
Was there any difference between the Hirn method and the method of
flipping when playing against a human?
Cpsoft
This is being reposted by special request.
Apologies to those who already read it.
Chris Whittington
Cpsoft <po...@cpsoft.demon.co.uk> wrote in article <01bbf42e$0ab61a40$c308...@cpsoft.demon.co.uk>...
Ingo Althoefer
In article <01bbf633$b41bb3e0$c308...@cpsoft.demon.co.uk>,
Cpsoft <po...@cpsoft.demon.co.uk> wrote:
>I have to admit to being utterly fascinated by this 3-hirn concept, and
>have 10 million questions to ask you. In no particular order:
>
>1. Have you ever considered the possibility of taking on Kasparov in this
>mode, or has anyone suggested it to you ? Because I think this mode, maybe
>with 3 machines, would give very interesting games. Maybe with (and no
>disrespect to you) with a strong GM coordinator ..... ?
In the year 1999 the city Weimar ( 20 km apart from Jena ) will be "Cultural
Capital of Europe". There are some diffuse plans to have a 3-Hirn match
with one of the world's top three GM's in this context. However, we are
still in search for a strong sponsor.
I think that stronger chess players which also know a lot about computer chess
should be able to lift 3-Hirn chess to a much higher level than I can do.
Look at the following "dream teams":
Deep Blue in the 2-variation mode and GM John Nunn
or
Deep Blue in the 2-variation mode and Hans Berliner
( unfortunately Hans Berliner does not like the 3-Hirn principle ... )
or
Deep Blue in the 2-variation mode and GM Christopher Lutz
( Lutz is the only GM who has written a chess program )
or
Deep Blue in the 2-variation mode and GM Joel Benjamin
( the advantage of Benjamin would be that he has been doing a lot of testing
with Deep Blue already )
But I think that Kasparov would not be willing to fight against such an
opponent.
By the way: 2-variation mode means that the machine computes not only the
best but also the second best proposal.
>2. What is your favourite/best combination of programs ?
I had my best experiences with programs which have solid evaluation functions
( like Genius or Fritz or Rebel ... ): when I am coordinating I also use the
information from the principal variations and from the evaluations and thus I
like reliable ( and comparable ) evaluations.
In AEGON 96 I tried 3-Hirn with Rebel 7 and M-Chess 5 (without prior training).
For me it was a terrible tournament because I was unable to interprete the
M-Chess values appropriately...
>3. I note you are from old DDR-Jena. Did the 1991 events make things
> different for you ?
You mean 1989 ( smooth revolution ) and 1990 ( reunification ) ? Of course
these made things different: however not in the way you may imagine now.
I have been studying and teaching in the University of Bielefeld ( western
Germany ) until I got my professorship in Jena in 1994. Without 1989 and 1990
there would have been no way for me to get a job in DDR :-) .
>4. Have you found anything about the combination program types ?
> Best with two strongest programs, or with a maniac and a stable program, or
> with two brute forcers/materialists or what ?
See my answer on 2. An important point is that I have to test programs inten-
sively before I use them successfully in a 3-Hirn setting.
>5. I gave you an old CSTal to try, but you said "although it plays some
> occasional stunning moves, it was not consistent enough". Can you
> elaborate ?
One of the reasons is that CSTs evaluation function is not too conservative.
Another one is that CST does not have the overall strength of say Genius or
Fritz. And if I use two programs of rather different strengths and always have
this in mind then I have psychological problems ( is the nice move of the
weaker program really okay with respect to tactics? ... ).
Ingo Althoefer.
Stefan Hahndel
In article <5a145t$a...@fsuj19.rz.uni-jena.de>, alth...@pdec01.uucp (Ingo Althoefer) writes:
|>
|> I had always been interested in computer chess but never found the time to
|> write a program of my own. In 85 my naive but simple idea was to take two good
|> chess programs and to write a coordinator p r o g r a m which "simply"
|> makes the final decision between the proposals of these two good programs.
I observed your interesting experiments with the Dreihirn concept over the
last few years. Unfortunately I also was always too sparse in time to write a
chess program of my own. I remember, when I first read about your experiments
in the German computer chess magazine CSS, I also thought, a programer should
try to write an artificial coordinator, some kind of expert system module
that uses a few other strong chess programs and decides which move to take
from it.
Now, several authors of chess programs try to take advantage by taking
parallel computers. But they all have the same problem: Most chess programs
of today use some kind of improved alpha-beta searchm (e.g.PAB,Scout,PVS, etc.)
All of these algorithms allows to prune big parts from the game tree
(to be more exact: from the game graph). BUT, in order to cut most of the
"normal" minimax game tree, they have to go sequentially throuh the game
tree AND have to sort the first moves on each node and ...
===> all those algorithms are optimized for a sequencial seach process.
If one tries to parallelize such an algoritm in order to run on several
processors, it has to search many, many more nodes in the tree and the
obtained speed off from the parallel search isn't so good, as one may have
expected.
Another point: Most of todays chess programs already search very deep and
1 or 2 plys more brings not so much improvements in playing strength as a
few years ago. I remember, a few years ago someone (I believe, it was
Berliner, but I am not sure about ...) published an estimation
of a 150-200 ELO point improvement per ply. I think, this rule is no longer
correct. It is correct if you go from 4 to 5 or from 5 to 6 ply ... but not
if your chess program searches 11 instead of 10 ply.
So, instead of trying to parallelize those algorithms in order to win 1 or
2 more ply in the search, it would be very interesting to think about a concept
similar to the Dreihirn concept of Althoefer.
I would propose the following concept: We use m different
chess programs/search algorithms. Some of those should be "traditional chess
engines" with different evaluation functions, as they are used today. Others,
should be search engines, specialized to search for special sub-goals in
chess. Suppose, we have n processors available. Then an "intelligent"
preprocessing algorithm should evaluate the current position and select a
chess program/search engine for each processor. It should ensure, that
always 2 or 3 "traditional chess" engines are selected in order to avoid
simple tactical blunders. On the other hand it should formulate possibles
plans/subgoals and select some of the search modules, optimized to search for
this special subgoals.
Then all those processors should search a time intervall t. After
time point t, another decision module shoud collect all results and select
one of the obtained moves as final move. Perhaps, one of the specialized
search modules has find a way to fulfil one of the formulated sub-goals.
If not, there is always a (good) move from the traditional chess engines, that
could be selected.
I believe, with such a concept a chess program (I should say: a collection
of chess programs and some very specialized problem solvers) could play better
chess, especially in closed positions, where long range planning would be
necessary. In such a closed position, chess programs of today have no chance
to find good tactical move. But, perhaps in the concept above, the preprocessing
module could formulate as important sub-goal, to eliminate that annoying
pair of pawns on c4 and d4 ... ;-) Another specialized module would search,
how to fulfill the sub-goal "open line e"., etc. ....
Of course, the main difficulty in such a concept would be to write a
good preprocessing module, that is able to formulate adequate sub-goals and
to play the decision module to select properly between the alternatives.
I am looking forward to hear some comments from authors of chess programs.
|> In 1997 I intend to write a book on 3-Hirn. Part of this book shall be a
|> diary on my 3-Hirn experiences: on the nicest games, the greatest victories,
|> the most terrible losses, the most funny anecdotes, the psychological
|> processes, and on many little details I learned during the games.
A good idea!
|>
|> 25.03.1985
|> White: 3-Hirn ( Mephisto II, Mephisto III, Ingo Althoefer )
|> Black: Thomas Hanf
|> Times: 120 minutes for 40 moves, afterwards 60 minutes for every 20 moves;
[...]
|>
|> Comments are welcome.
I assume, you have all your games in a modern chess database.
It would be helpful, if you could post the games in PGN, in order to save
possible commentators some work.
Stefan Hahndel
--
Technische Universitaet Muenchen Phone: +49-89-48095-257
Institut fuer Informatik Fax: +49-89-48095-250
Orleansstr.34, D-81667 Muenchen, Germany;
http://www6.informatik.tu-muenchen.de/~hahndel/
mclane
alth...@pdec01.uucp (Ingo Althoefer) wrote:
>One of the reasons is that CSTs evaluation function is not too conservative.
>Another one is that CST does not have the overall strength of say Genius or
>Fritz. And if I use two programs of rather different strengths and always have
>this in mind then I have psychological problems ( is the nice move of the
>weaker program really okay with respect to tactics? ... ).
>Ingo Althoefer.
If the programs you use are very THE SAME how can you be psychological
sure that they don't have the same LACK of knowledge or
search-algorithms in the same position ?!
I have seen fights between programs that were totally different , and
they evaluated totally different and searched different.
I you e.g. let Nimzo3.4 and Fritz4.01 play and you should decide
between these 2, your result could not increase very much.
Because the 2 programs are pretty the same.
In my opinon one should try with programs that evaluate different and
maybe work different. Playing strength is not that important.
This is my experience from mail-chess. Many programs have the same
evaluations after 12 or 13 plies, but some programs have ideas and
plans. Why should I trust programs that have the same main-lines and
the same moves and evals ?! Maybe both are wrong...
If you have 50 % this paradigm and 50 % this paradigm-program you can
have a 50% chance to find the better move.
Ingo Althoefer
In article <E39qv...@news.prima.ruhr.de>,
mclane <mcl...@prima.ruhr.de> wrote:
>If the programs you use are very THE SAME how can you be psychological
>sure that they don't have the same LACK of knowledge or
>search-algorithms in the same position ?!
In 3-Hirn it is not too important that the programs have knowledge. It is
much more important that the coordinator has strategic knowledge: He has
to filter out plans from the "strategically random" proposals of the two
programs. For example, this works very well in endgames with bishops on
squares of different colors.
>I you e.g. let Nimzo3.4 and Fritz4.01 play and you should decide
>between these 2, your result could not increase very much.
>Because the 2 programs are pretty the same.
I am not sure. In my eyes Nimzo 3 and Fritz 4 are rather different.
>In my opinon one should try with programs that evaluate different and
>maybe work different.
Make experiments and tell us the results. Especially you would a be a good
coordinator in a 3-Hirn with CST and some other program because you know so
well about the peculiarities of CST.
Ingo Althoefer.
Stefan Meyer-Kahlen
On 30 Dec 1996 09:57:28 GMT, hah...@informatik.tu-muenchen.de (Stefan
Hahndel) wrote:
>
>In article <5a145t$a...@fsuj19.rz.uni-jena.de>, alth...@pdec01.uucp (Ingo Althoefer) writes:
>|>
>|> I had always been interested in computer chess but never found the time to
>|> write a program of my own. In 85 my naive but simple idea was to take two good
>|> chess programs and to write a coordinator p r o g r a m which "simply"
>|> makes the final decision between the proposals of these two good programs.
>
>I observed your interesting experiments with the Dreihirn concept over the
>last few years. Unfortunately I also was always too sparse in time to write a
>chess program of my own. I remember, when I first read about your experiments
>in the German computer chess magazine CSS, I also thought, a programer should
>try to write an artificial coordinator, some kind of expert system module
>that uses a few other strong chess programs and decides which move to take
>from it.
Don't hesitate. Right now I am working on such a coordinator, and it
is already playing games.
Stefan
[snip]