The American Contract Bridge League’s 8th
annual World Computer-Bridge Championship was held in
History
At the suggestion of New York Times bridge editor Alan Truscott, I petitioned the ACBL
in 1996 to hold a computer-bridge event…and thus was born the World
Computer-Bridge Championship.
The ACBL inaugurated this annual
computer-bridge championship in 1997 as a way of encouraging computer-bridge
software developers to accelerate their robots’ development to expert
class. Since then progress has
accelerated and the top robots have advanced significantly, but not yet to the
ranks of expert. The top robots often
exhibit expert play but not consistently enough to be considered expert.
This championship has been held
every year at an important human international bridge event, with me as the
organizer. The preceding seven
championships were held three times at ACBL NABCs, three times at World Bridge
Federation (WBF) World Championships and last year at the European Bridge
League’s (EBL) first European Open Championship. When hosting this championship, the ACBL, WBF and EBL have given
both financial and organizational support.
For a complete history and details of previous championships go to
ny-bridge.com/allevy/computerbridge.
Technical remarks
A bridge “table” consists of a central server, or Table Manager
(TM), and four connecting computers, which “seat” the robots. The TM distributes the four hands of each
board to the robots. Play proceeds
automatically with the TM receiving and passing information to the robots and
recording the play. This year P4 2.4 GHz/512
MB PCs were used, running under Windows XP.
The speed of play was set at 2 minutes per pair per deal, approximately
half that of human play.
Without getting into the
details, most of the robots are programmed with a combination of
knowledge-based AI, or sets of rules, and search-based AI, or simulations.
Preparation
This is very much a group effort between the organizer, the
computer-bridge software developers and the ACBL. An administrative committee sets the Conditions of Contest with
input from the software developers and bridge experts.
It is important that the software developers understand their opponent
robots’ methods in advance of the competition so that they can prepare
defenses. In human play, advance notice
is not necessary when unusual systems are not allowed, as is the case in this
competition. However, in
computer-bridge play the robot developers need time to program defenses and store information about the opponent robots’
methods. Contestants are required to
submit a Convention Card (CC) one month before the competition. This is in the form of a spreadsheet, with
over 100 questions. Once the CCs are
published contestants ask each other (through email) even more detailed
questions about their methods and conventions.
Since this information is stored in the robots’ memories before the
competition begins, few alerts are
necessary during play. In the few
instances where alerts are required the pertinent information is input into
memory and play continues.
Play format
The competition is in the form of team matches, with a team’s
robots seated N-S at one table and E-W at the other table. The boards in a match, or part of a match
such as a 16-board set, are played sequentially, first at one table (closed
room) and then at the other table.
The five-day event starts with a
24-board round-robin with the top four robots advancing to a 64-board knockout
(KO) semifinal with carryover. The
round-robin is scored on an international 30-VP scale, where the winning team
receives a maximum of 25 VPs for a 52 or more IMP victory. To earn a carryover in a semifinal or final
KO match, a team must both win its head-to-head round-robin match against its
KO opponent and finish higher in the overall round-robin standing. The carryover is the lesser of these two VP
differences
The Bridge World
Editorial on the use of a qualifying round-robin preliminary to a knockout
phase
In the August 2004 issue of the Bridge World,
Jeff Rubens comments that while a round-robin preliminary segment may provide a
substantial amount of guaranteed play for entrants who may have endured
considerable expense to play, there are so many flaws that its use may
sacrifice tournament quality.
Some of the flaws mentioned for human-play are:
the scoring system, where the closer it is to win-loss, the greater the danger
that small differences will have enormous consequences, and the closer it is to
total score, the greater the danger that winning margins of strong teams
against weak teams will be relevant; sportsmanlike dumping is a theoretical
threat in late matches that will determine who are matched in KO round, or who
qualifies for the KO round; and preparation against many different systems is
required which severely increases the preparation needed.
In computer-play some of these flaws do not
exist and others are minimized. The robots
aren’t allowed to dump and their brains can be checked if there is an
accusation. The robots aren’t
influenced by the state of affairs so the timing of the matches is irrelevant,
except for the suspense of the spectators.
[In fact, this was the case this
year. Viewing Table 2 we see that the
stronger robots played each other in early matches therefore greatly reducing
spectator suspense at the end. In the
future the highest ranked robots will be matched near the end of the
round-robin rather than near the beginning.] In our competition the preparation needed to study and defend
against all the opponents is limited.
This is due, in part, to only allowing systems and conventions that are
common international methods so that the software developers can concentrate on
improving their robots’ bridge skills and spend less time on understanding and
defending against complicated methods.
In our competition Convention Cards are due one month before the
competition. While the list of
allowable conventions is large (standard methods vary around the world) the
robots’ photographic memories easily digest the methods of their
opponents. For certain classified
uncommon treatments the robots explain the meaning to the opponent robots
through electronic or manuals means. We
use a 25-VP scale. It is not close to
win-loss and the winning margin against weak teams is softened as the VP award
is topped at 25, not 30. As can be seen
in this year’s competition, beating-up on the weaker teams did not affect the
round-robin standing. With one
exception, the top five round-robin finishers scored 25 VPs against the weakest
three teams. The one exception was
The
competition
The round-robin ended with Bridge Baron topping
all competitors with 147 VPs. Closely
behind were Wbridge5, with 145 VPs, Jack, with 138 VPs and
The
robots, their developers and the final round-robin standing are shown in Table
1. Table 2 shows the IMP/VP results of
each round-robin match and the round that each match was played.
Table 1: Entries along with the 24-board round-robin final standing.
|
Robot |
Developers |
Country |
VPs |
|
Bridge Baron |
Stephen Smith, George Yanakiev, Jason Rosenfeld and
Tom Throop |
|
147 |
|
Wbridge5 |
Yves Costel |
|
145 |
|
Jack |
Hans Kuijf, Wim Heemskerk and Martin Pattenier |
The |
138 |
|
|
Tomio and Yumiko
Uchida |
|
131 |
|
Q-Plus Bridge |
Hans Leber |
|
108 |
|
|
Ian Trackman and
Mike Whittaker |
|
63 |
|
|
Rodney Ludwig |
|
37 |
|
Sabrina |
Pierre Cormault and Gérard Joyez |
|
3 |
Table
2. Round-robin match results
|
|
Jack |
Bridge
Baron |
Wbridge5 |
|
Q-Plus Bridge |
|
|
Sabrina |
Position VPs |
|
Jack IMPs Round VPs |
|
72-15 1 25-5 |
47-53 2 14-16 |
27-58 3 9-21 |
44-45 4 15-15 |
74-6 5 25-3 |
152-1 6 25-0 |
124-8 7 25-0 |
3 138 |
|
Bridge
Baron |
15-72 1 5-25 |
|
62-36 3 20-10 |
56-21 4 22-8 |
81-19 5 25-4 |
95-35 6 25-4 |
114-33 7 25-1 |
121-12 2 25-0 |
1 147 |
|
Wbridge5 |
53-47 2 16-14 |
36-62 3 10-20 |
|
60-31 5 20-10 |
80-30 6 24-6 |
94-6 7 25-0 |
102-10 1 25-0 |
151-5 4 25-0 |
2 145 |
|
|
58-27 3 21-9 |
21-56 4 8-22 |
31-60 5 10-20 |
|
77-40 7 22-8 |
70-44 1 20-10 |
136-2 2 25-0 |
141-1 6 25-0 |
4 131 |
|
Q-Plus
Bridge |
45-44 4 15-15 |
19-81 5 4-25 |
30-80 6 6-24 |
40-77 7 8-22 |
|
101-22 2 25-2 |
152-18 3 25-0 |
184-4 1 25-0 |
5 108 |
|
|
6-74 5 3-25 |
35-95 6 4-25 |
6-94 7 0-25 |
44-70 1 10-20 |
22-101 2 2-25 |
|
55-31 4 19-11 |
97-32 3 25-3 |
6 63 |
|
|
1-152 6 0-25 |
33-114 7 1-25 |
10-102 1 0-25 |
2-136 2 0-25 |
18-152 3 0-25 |
31-55 4 11-19 |
|
112-19 5 25-0 |
7 37 |
|
Sabrina |
8-124 7 0-25 |
12-121 2 0-25 |
5-151 4 0-25 |
1-141 6 0-25 |
4-184 1 0-25 |
32-92 3 3-25 |
19-112 5 0-25 |
|
8 3 |
Q-Plus Bridge did not make the semifinals but
it did win the round-robin “best-played hand” award for its play on this deal
which occurred in the forth round against Jack. 26 IMPs were at stake. 13
IMPs to Q-Plus Bridge if it makes 6© and 13 IMPs to Jack if it fails.
|
||||||||||||||||||||
|
West |
North |
East |
South |
|
Q-Plus |
Jack |
Q-Plus |
Jack |
|
1§ |
Pass |
1© |
1ª |
|
2ª |
Pass |
4ª |
Pass |
|
5§ |
Pass |
6© |
All Pass |
With the 1ª bid by South,
West’s hand became bigger, and the Q-Plus Bridge robots bid to an overly
aggressive heart slam. South led a low
spade, and declarer finessed, drew trumps, eliminated diamonds and spades,
cashed the §A and followed with a low club. The defense was helpless and Q-Plus Bridge
scored +1430.
Semifinal
In the semifinals, round-robin
first place finisher, Bridge Baron, had a 14 IMP carryover against fourth place
finisher,
The last time Jack played Wbridge5 in a
championship KO stage was the 2002 finals.
That final was the closest KO final in computer-bridge history with Jack
winning by 1 IMP. This year Jack
continued its winning ways, this time defeating Wbridge5 more comfortably,
157-118. In the other semifinal match
Bridge Baron defeated Micro Bridge 166-126.
Board 64 of the semifinals earned Jack 17 of its 39 IMP
margin of victory.
Board 64 North
Dlr: West ª 6 5 3
Vul: E-W © Q 10 4
¨ 10 8 7 2
§ 10 8 5