FOST Cup 1995 - Tokyo
Report on the First FOST cup World Open Computer Go Championship
by David Fotland
The competition was held at the Nihon Ki-in, in Tokyo Japan, on
September 29 and 30, 1995.
The Foundation for the Fusion of Science and Technology (FOST) is
a non-profit organization, founded in June, 1994, to sponsor research
on gaming and simulation as well as projects that bring together
basic science and technology.
The computer go competition is sponsored by FOST, in cooperation with
the Japan Computer Go Society (JCGA), with the support of the
Nihon Ki-in, Apple Computer, Japan IBM, NEC, Fujitsu, and the Yomiuri
newspaper.
FOST plans to have this competition every year. For information
contact:
Japan Computer Go Association
Ishigami Kugenuma 1-5-17-702,
Fujisawa, Kanagawa, Japan 251
Tel: +81-466-22-7901
Fax: +81-467-83-4710
e-mail: MHB00531@niftyserve.or.jp
This year the prizes were 2,000,000 yen for first, 500,000 yen for
second, and 200,000 yen for third place. The top 3 finishers also
each won a computer.
Games are played using Japanese rules, with 5.5 point komi. Each
program is allowed 1 hour for 125 moves, based on the time kept by
the computer's own clock. There is a 10 minute time penalty for
programs that can not communicate using the Standard Computer Go
Modem Protocol.
The specification and source code for the Modem protocol can be found
via anonymous FTP at bsdserver.ucsf.edu. e-mail fotland@cup.hp.com
with any questions.
Most programs implemented the modem protocol. The evening before the
tournament started was set aside for protocol testing.
Using the protocol made the tournament run very quickly. We played
four rounds the first day, between 11:00 AM and 6:00 PM, and 3 more
rounds the second day between 9:00 AM and 3:00 PM. Logistically, the
computers were arranged on tables arranged in a U, with the computers
facing out. Between rounds, only the rs-232 cables were reconnected.
Computers were not moved for those who had the protocol. Initially
the organizers thought that there would only be time for 4 rounds,
but before the tounament started, all agreed to play as many rounds
as possible before the 3:00 deadline.
A large number of computers were provided by the organizers, including
about 6 Fujitsu 90 Mhz Pentium machines. The organizers provided a
program for entering and displaying moves that implemented the modem
protocol. When a program with the protocol played one without it,
The program without it used two computers, one for his program, and one
running the modem protocol. His opponent, with the modem protocol, did
not have to operate his program since it was connected by RS-232 to
the JCGA modem protocol/display program. There were several problems
getting the modem protocol working between machines requiring games to
be restarted after a few moves. Igo Meijin crashed every game, but
there were no other problems with crashes.
RESULTS:
Place
WINS SDS SOS
1 7 Chen ZhiXing Handtalk China Pentium 90
2 6 Michael Reiss Go4++ England Pentium 100
3 5 David Fotland Many Faces of Go USA Pentium 90
4 4 13 Ken Chen Go Intellect USA Macintosh
5 4 9 Satou Tosikazu TY '96 Japan Nintendo
6 4 8 Shimada Kou Takuchan Japan
7 3 9 Oishi Yasuo Goro Japan
8 3 7 27 Yoshida Takenao Taikyoku Igo Mituki Japan
9 3 7 23 Yan Shi-Jin Jimmy Taiwan IBM-PC
10 3 6 Honi Tuneo Utoro Japan
11 2 2 18 Yoshikawa Takeshiro Dai Hininbo Japan IBM-PC
12 2 2 17 Mikakana Masumi Ijoh Kankaku Japan
13 2 2 14 Gary Boos Teamgo USA IBM-PC
14 0 Saito Yoichi Igo Meijin Japan
After the competition, the top four progams played exhibition games
against amateurs at 9 stone handicap. Handtalk played a 6 Kyu girl
about 7 years old, and won. Many Faces played a 1 Dan teenager and
won by resignation at move 136. A 9-dan professional took over black
at that point, and in 30 moves had made the game even, but then
made some overplays trying to kill a big group and died himself, so
he also resigned. Go Intellect played a teenager and won. Go4++ lost.
A pro observed the games of the top programs and played against them
briefly. Based on this, Handtalk was awarded a 5 Kyu diploma, Go4++
a 7 Kyu diploma, and Many Faces of Go an 8 Kyu diploma.
The top 4 programs were clearly superior to the others. Games between
the top programs were close, with the lead changing place several times.
Games between the top 4 and other programs were usually won by large
margins.
Handtalk has been enhanced since last year with full board lookahead,
accounting for a significant increase in strength. It is slower than
before but still faster than the other top programs. Mr Chen works on
it full time, and is assisted by a Chinese Pro. Handtalk beat the
other top programs convincingly. Many Faces was never in the game.
Go4++ had some chances to win, but Handtalk was a better fighter.
Even if Handtalk had lost its last round game against Go Intellect
it would have still won the tounament. Ken Chen forgot to load his
joseki library in the last round game against Handtalk, and went wrong
early in one corner, giving Handtalk the lead.
Go4++ is an old program that hasn't been in an Asian world computer
go championship for about 7 years. It is the slowest of the programs,
and only became practical at tournament time levels with the Pentium
processor. Its strength is very accurate assessment of territory. It
was able to make an unusually large center territory in most of its games.
Many Faces of Go was slightly enhanced from the 9.0 version that is
shipping. It had a very close game with Go4++. The lead changed
several times, but Go4++ cut off some stones near then end and won
by 6.5 points. The game against Go Intellect was an upset. Go
Intellect was leading most of the game, but let Many Faces crawl along
the edge through its entire big territory in the endgame. Many Faces
had bad luck choosing colors. It tends to play a few slack moves in the
opening, so it does better against the top programs taking black. It
had White in all 3 games against the top programs.
Go Intellect did not play up to its expected strength. It played a
quiet, territory oriented game against Go4++. When it finally
invaded Go4++' center territory it was too late and the invasion
died. It had some chances to win the game against Handtalk, but in
the end lost by about 30 points. Go Intellect is still running on
a slow, 68040 based Macintosh. Ken plans to port to the Pentium for
next year.
There were some cognitive psychologists there who are studying Go.
A Japanese study shows that strong players only analyze on average
1.5 moves before selecting one. Go Intellect and Handtalk, written
Many Faces of Go, written by a 3 dan, examines about 10 moves before
selecting one. Go4++, written by a 2 Kyu, examines about 50 moves
before selecting one.
Complete tournament grid (in order by initial seeding)
Two Korean programs and one Japanes program dropped out at the last
minute, and Teamgo was added at the last minute. Teamgo was very
difficult to operate so it played one lessround than the other
programs. Saito's program crashed in the first round against
Many Faces of Go, and resigned all its other games without playing.
1 2 3 4 5 6 7 8 9 10 11 12 13 14
1 Ken Chen - 0 0 1 0 1 1 1
2 David Fotland 1 - 0 1 0 1 1 1
3 Michael Reiss 1 1 - 0 1 1 1 1
4 Yan Shi-Jin 0 0 - 0 1 1 0 1
5 Chen Zhixing 1 1 1 - 1 1 1 1
6 Oishi Yasuo 0 0 0 1 0 - 1 1
7 Yoshida Takenao 0 0 0 0 - 1 1 1
8 Saito Koichi 0 - 0 0 0 0 0
9 Honi Tuneo 0 0 - 1 0 1 0 1
10 Yoshikawa Takeshiro 0 0 0 1 0 - 1 0
11 Satou TosiKazu 0 0 0 1 1 - 1 1
12 Gary Boos 0 1 0 0 0 - 1
13 Shimada Kou 1 0 0 1 1 0 - 1
14 Mikakana Masami 0 0 1 0 1 0 0 -