In the game of chess, human players typically have good ideas in strategy
and long-range planing, whereas they often make tactical errors resulting in
the loss of a piece or a mate in two. With chess computers it is almost the
other way round: The machine will never loose a piece within its horizon of
computation, whereas strategic planing happens only by chance. Rather
naturally a question arises:
Is it possible to combine the strengths of human and machine in chess ?
Starting in 1985, I conducted several experiments to answer this question.
I was an amateur player with an Elo rating of 1900, when I stopped club
play in 1994. (The better you are, the higher your Elo rating will be.
International Masters (=IM) are typically between 2400 and 2500, "normal"
Grandmasters(=GM) are at 2500 or above. Currently the strongest human is
Garry Kasparov with about 2800 Elo points. A rule of thumb says that you
are expected to score 75 percent against an opponent who is 200 Elo
points below you.)
My basic man+machine approach was the 3-Hirn setting. "Hirn" is a
term in German language and means something like "brain" or "mind". (The
name "3-Hirn" was my own idea.) In chess a 3-Hirn consists of two different
chess computers and one human chess player, the controller. ( So 3 = 2
computers + 1 human. ) Both computers are started for the position on the
board. They are run in an "infinite mode" and compute and compute. The
controller inspects their displays all the time (in "infinite mode" a chess
computer always shows its currently best candidate move) and stops them in
an appropriate moment. Then he has the final choice among the proposals of
the two machines. If both computers prefer the same move this one has to be
executed. So the human is not allowed to overrule the machines. He is
"only" the referee in cases of disagreement. (But of course the power to stop
the machines in appropriate moments gives him a lot of influence.) The
computers propose only moves which are tactically sound, and so the human
controller can fully concentrate on strategic decisions. 3-Hirn plays chess
against other humans or computers.
Between 1985 and 1996 I played more than 100 games under tournament
conditions with such 3-Hirns. Typically I had some of the most modern
commercial chess computers in "my team". In all these games I was the
controller in the man+machine symbiosis. Of course 3-Hirn was in general
not allowed to enter normal chess tournaments, so most of these games were
played in privately organised events. Most remarkable were two matches: In
January 1992 IM Dr. Helmut Reefschläger ( Elo 2405 in that year ) lost an
8-games match by 3:5 against a 3-Hirn in which I used the "Mephisto Lyon
68030" (Elo about 2260) and the "Chess-Machine" (Elo about 2230) as
computers. In October 1995 GM Christopher Lutz (Elo 2570 in that year)
beat 3-Hirn by 4.5 : 3.5 . In that event I used the PC chess programs Genius3
and Fritz3, both running on Pentium computers with 120 MHz speed (Elo
ratings below 2450).
In 1996 the new chess program Fritz4 (Elo also below 2450) had a very nice
feature, the k-best mode. In this k-best mode not only the single
best but the k best moves are computed by the program, k being some natural
number like 2 or 3. I got the idea of Double - Fritz + Boss.
In this setting the program Fritz is running in its 2-best mode. The human
controller ( = the Boss ) stops in an appropriate moment and then has the
final choice among the two candidate moves of Fritz. Double-Fritz+Boss
played an 8-games match against GM Gennady Timoshchenko (Elo 2530 in 1996;
he was an assistent of Garry Kasparov from 1982 to 1986) and achieved a
narrow 4.5 : 3.5 victory. "Of course" I was the Boss of Double-Fritz.
For me it was a very natural idea to combine the two approaches of 3-Hirn
and Double-Computer+Boss. So in 1997 I introduced List - 3 - Hirn.
Two different chess computers are involved, and one human controller. The
chess computers are running in some k-best modes ( for appropriate values
Two different chess computers are involved, and one human controller. The
chess computers are running in some k-best modes ( for appropriate values
of k, typically k=3 ), and the human has the final choice among the
proposals from the two k-best lists. The name "List"-3-Hirn shall indicate
that the final move is selected from lists of proposals. With this
List-3-Hirn construction I played an 8-games match against GM Arthur Yusupov
(number 31 in the world with Elo 2640 in 1997), winning by 5:3 . However,
this was a match in "Shuffle Chess", a variant of chess propagated for
instance by former World Champion Bobby Fischer. In Shuffle Chess the
starting positions of the pieces on the back ranks are determined at random
to exclude the influence of opening theory. In the match with Yusupov I
used several chess computers in different combinations, all with ratings
about Elo 2500. (In all matches mentioned above, the thinking time was
40 moves in 2 hours, and afterwards 20 moves in 1 hour.) From all these
experiments I conclude that a human amateur player can improve the
strengths of chess computers considerably by appropriate controlling,
even if the computers are much stronger than this human.
Originally I had intended to play some match with List-3-Hirn against a
Top-10 GM in 1998 (and another match against the human World Champion in
1999 or 2000 ...), but the masters are no longer interested. One of them
was so honest to tell me that money was not the key point for his decision
to reject my match offer, but that he felt that the use of chess computers
changed the character of the game in a way he did not like. After this
reaction I decided to stop my experiments. Currently I am in the process
of writing a book on my experiences in man+machines symbioses in chess.
One of my Ph.D. students is Stefan Meyer-Kahlen, who is the "father" of
the chess program Shredder, the Micro-World-Computer-Chess-Champion in
1996. In his thesis project, Stefan is constructing an automatic 3-Hirn,
consisting of three different programs. Two of them are "normal" chess
computers, and the third one is a controlling program
which has the final choice among the proposals of the other two. This
controlling program is also doing the time management. Early in 1998,
this construction participated successfully in the Paderborn computer
chess tournament under the name "Clever & Smart", running on a Dual-PC
( a PC with two Pentium processors, one for each for the two normal
programs ).
In the future I will investigate the approach with one human and
several computers in some problem solving areas outside of chess.
Natural candidates for this are the fields of :
Computer-assisted analysis in Medicine,
Computational Biology,
Computer-assisted theorem proving in Mathematics .
Perhaps some readers are wondering why I write "computer-assisted" and
not "computer-aided" ( as for instance in CAD: Computer-Aided Design ).
It is my personal way to distinguish: In cases where the human uses one
computer I say "aided", and "assisted" when several machines are involved.