ACBL/WBF 17^th World Computer-Bridge Championship

by Alvin Levy, World Computer-Bridge Championship organizer

Bali, Indonesia, 
September 23-28, 2013
Held at the 41^st World Teams Championships
See the on-site coverage at the World Bridge Federation (WBF) website, photos and complete results at www.computerbridge.com

The World Computer-Bridge Championship is held annually at a major human championship.  It has been held at the WBF World Championships, the American Contract Bridge League (ACBL) North American Bridge Championships (NABC), and at the first European Bridge Federation (EBL) Open Championship (2003).  Six of the best robots were entered in this year’s championship, including: the two top past winners, defending champion Jack (The Netherlands) and WBridge5 (France); past champions Shark Bridge (Denmark) and Bridge Baron (USA); and many time runner-ups Q-Plus Bridge (Germany) and Micro Bridge (Japan).

The format is a 48-board round robin with the two top finishers playing for the Gold medal in a 64-board KO match, with possibly carryover from the round robin stage.  The Conditions of Contest call for a semifinal stage when there are seven or more entries.  Twice before, in 2001 and 2005, were there as few as six robot teams entered.  The greatest number of entries was ten, in 2009.

The contestants all used the same computers, 2.9 GHz Intel Core i5 desktop PCs under Windows 7 OS.

With one round to go it was possible for WBridge5, Q-Plus Bridge, Micro Bridge and Jack to take the two final spots.  The standing after four rounds: WBridge5 (53.26); Q-Plus Bridge (46.51); Jack (45.77); Micro Bridge (42.37); Shark Bridge (27.43); Bridge Baron (24.66).  Interestingly, the new VP scale should eliminate ties. However, if Micro Bridge beat Jack by 10 IMPs they would tie with 54.07 VPs, but out of the money as Q-Plus Bridge would have finished second.  A three way tie was almost possible.  If Micro Bridge won by 10 IMPs and Q-Plus Bridge lost by 15 IMPs, Q-Plus Bridge would have finished with 54.05 VPs.  This might show the merits of the new scale.

Board 5 from the last round robin had the theme...bid one more!

¹ diamonds and hearts;
² See Below

The play started ♦K, ♦A, ♦x, ruff and over-ruff and now declarer had no trouble in holding the trump losers to one for down one in 5♠x. E/W -100.

At the other table,

¹ Hearts and a minor

The play started ♦K, ♦A, ♦x, ruff and over-ruff and now declarer had no trouble making 4♠. E/W +420.

11 IMPs to Jack.

In another match,

¹ Diamonds and hearts
2 See Below

Making 5, N/S +850

The play started ♦K, ♦A, ♦x, ruff and over-ruff and now declarer had no trouble making 4♠, E/W +420.

15 IMPs to Q-Plus Bridge.

In the other match,

¹  diamonds and hearts
² See Below

Made 5, N/S +850

The play started ♦K, ♦A and ♥4.  Now declarer led a spade to the jack, and had to loss two trumps for down  E/W -100.

13 IMPs to WBridge5

² interesting comments by the developers Hans Kuijf (Jack), Yves Costel (Wbridge5) and Hans Leber (Q-Plus Bridge)

Hans Kuijf ..."Robots are certainly less partner oriented than humans. There are many reasons for that. Most human partnerships demand that. In fact humans tend to invite rather than bid game themselves just because of that. Even when in fact they have a hand across which partner cannot possibly make the correct decision. The truth is somewhere in the middle: Humans include partner too much and robots not that much. The reason in Jack [5♥ bid) is this: he anticipates a bid, not based on simulations for which we lack the CPU time, but on a rule base. If partner is able to make the right decision in a number of samples then Jack includes him in the bidding (either by inviting or making some other descriptive bid).  In this case: 5♦ tends to show longer diamonds than hearts.  Humans [South] however will certainly bid 5♣ or 4N with equal length)."

Yves Costel..."WBridge5 has a rule to add one to the length of a long suit with AKQ or AQJT or AKJT.  In that case diamonds are considered longer than hearts".

Hans Leber ..." ... after 4♠ the program sees the additional length in both hearts and diamonds, and then prefers the major over the minor without "thinking", i.e., purely rule based. I would expect the program to have made a simulation, but it did not, because 5♦ did not come into the picture, which I consider an error.  If it had run a simulation (I started one manually in the situation), it would have selected 5♦.  So 5♥ is mainly an error which turned out lucky. Over South's 5♦ North would pass if East and West pass,  but run to 5H if 5D is doubled.)"

It is quite interesting to see how the developers can explore their robots' 'thinking' process!

Board 10 of the final round robin started the same way at all six tables, with East dealer, the bidding started 1♥ - 3 ♠ - Pass - Pass.  At five of the six tables East reopened with a Dbl. 

Opening lead ♠8, making E/W +600

 In one match Micro Bridge (West) bid 3NT, as shown above, and could not be stopped from making game, while at the other table, Jack (West) bid 4♥ and went down 3 for -300 and 11 IMPs to Micro Bridge.  In another match Shark Bridge (West) bid and made 4♣ for +130 while at the other table Q-Plus Bridge (West) passed.  In 3♠x Shark Bridge played correctly by setting up a diamond trick for a heart discard before playing trumps and was +730 (only double dummy defense can beat 3♠) and IMPs.  In the final match WBridge5 (East) doubled and WBridge5 (West) passed.  After a heart lead, East cashed the club ace and the diamond ace on which Bridge Baron correctly unblocked the king (not needed in this particular layout) and was +730.  At the other table, Bridge Baron did not balance with a double, and WBridge5 went down one in 3♠ when declarer played on trumps before diamonds.  13 IMPs to Bridge Baron.

A good save on board 15 of the final 16 boards from the last round robin was found three times.  In one match Jack picked up 11 IMPs against Micro Bridge.

Opening lead ♠4, down 3, E/W -800

Making six, N/S +1370

11 IMPs to Jack

Wbridge5 found the save against Bridge Baron for 17 IMPs (6♥ was not doubled for -200).  Q-Plus Bridge gained 13 IMPs against Shark Bridge when it was allowed to play 5♥x for -500, and made +1390 at the other table on the ♥A lead.

For a comparison to the human play, two 16-board sessions were taken from the championship round robin play (round robin sessions 1 and 14), and used in the final 16-board session of the 2^nd and 3^rd round robin.  One can compare the robot results to the human results as all the robot play records are shown and the links to the human records are given at the official site, www.computerbridge.com

Round 14 of the human round robin was used in the third round of the robot round robin.  On board 12, 7♥ was the final contract at three of the six robot tables and 6♥ was the contract at the other three tables.

In one match Jack picked up 13 IMPs against Shark Bridge.

13 IMPs to Jack

In another match Bridge Baron picked up 13 IMPs against Micro Bridge.

¹ 2NT or game force; ² 2 or 5 key cards; ³ no kings

13 IMPs to Bridge Baron

In the other match WBridge5 picked up 13 IMPs against Q-Plus Bridge.

¹ game force; ² no ace, 7+ hcp; ³ 5 key cards

¹ 2NT or game force; ²control; ³0 or 3 key cards

13 IMPs to WBridge5

In human competition, the deal was played at 22 tables in each category.  7♥ was reached 9 times in the Bermuda Bowl, 9 times in the Venice Cup (played once in 4♥), and 9 times in the d'Orsi Senior Trophy (played once in 5♥).  So the robots' percentage was slightly better than in the human competition.

Interestingly robot play avoids ruling based on hesitations.  In the human play there was at least one case in the Bermuda Bowl that required an Appeal Committee to rule on this deal.  See http://newinbridge.com/news/2013/oct/break-tempo-%E2%80%93-or-not  Apparently this is a bigger problem than generally known.  See Barry Rigal's comment at http://newinbridge.com/news/2013/oct/don%E2%80%99t-amend-rules#comment-5900

Board 15 of the round produced a large swing on an opening lead against 6♦.

In one match, both Q-Plus Bridge and WBridge5 bid and made 6♦.

¹ one key card
opening lead, ♠ 8, making 6, N/S +1370

opening lead, ♣ A, making 6, N/S +1370

The other time 6♦ was bid.

¹ one key card, ² not the diamond queen… at least N/S are on the same wave length….can't play in 5♦?

opening lead, ♣ A, making 6, N/S +1370

At the other table Shark Bridge made +660 in 3NT.  12 IMPs to Jack

At the final match, the contracts were 3NT, +660 and 5♦, +620.  Bridge Baron picking up 1 IMP against Micro Bridge.

In the human championships, 6♦ was reached at many tables, and for the big clubbers with South as declarer.  When South was declarer, West led a heart more often than not, but with North as declarer, the opposite was true.  The human play (except in Daily Bulletin articles) is not revealed, so the opening lead cannot be analyzed.  In 6♦ from the North side, the human defense got it right some times (25%), but the robots were 0 for 3.

The final KO between Jack and WBridge5 saw many swing and potential swing boards.

An excellent grand was bid and made at both tables on board 11 of the 3^rd quarter.

¹singelton or void; ² 1 or 4 key cards; ³ one king

¹ one key card; ² trump queen and club king (one king, not in hearts)

Made 7 at both tables, E/W +1510, no swing

The last board of the match shows an interesting inference that humans sometimes get right and sometimes get wrong but robots always get wrong.

The play often depends on declarer's estimate of the level of an opponent's play.  Robots don't make such a judgment.

¹ Jacoby, 11-14 hcp; ² asking; ³ 13-14 hcp with four spades

Opening lead ♥K, Made 5: N/S +450

¹ Singleton plus additional values; ² relay; ³ singleton heart;  ⁴ 2 of 5 keycards

The play comes down to finding the ♣10.  The holdings that matter to declarer are West holding ♣ K, ♣ K 10, K x and K x x x.  With ♣ K x x, West will always duck the king and defeat the game.  With K x x x, West will never cover with the king as the 10 will show up.  With K 10, West will always cover.  Now with K x, do you cover or not?  If it is known that West always covers with K x declarer will always get K x x x right (when the king doesn't appear play for the K x x x as opposed to K x); have a guess for K 10, singleton K and K x when it does appear, with the finesse of the 10 approximately four times as likely to win as the drop of the 10 from K 10.  If it is known that West never covers with K x then declarer will always get K 10 right and singleton K wrong, and have a guess for K x and K x x x when the king doesn't appear.  Of course on any given hand declarer might be able to determine the count on the suit in question, as on this hand, after playing the side suits, west cannot have 4 clubs, so declarer will play for K x.  A cat and mouse game…or game theory.  If defender assumes, from K x, that if covering with the king, the declarer will almost always finesse for the 10, then defender will never cover from K x.  If declarer knows this is defender's thinking, then declarer will always get K 10 right and singleton king wrong!  A matter of game theory (judging the probability of your opponent's play from certain holdings) which the robots don't do!  The robots will always finesse for the 10 when the king is played and apparently (most robots) play the K from K x.  In human play, with two known experts against each other, duck from K x most of the time (Edgar Kaplan would give the percentage quite accurately (80% of the time?)...as he did in his famous "The Deep Nine" article (The Bridge World, October 1973).

Board 3 of the 2nd quarter may have determined the winner.

Opening lead ♣5, making 6: E/W +680

¹2 of 5 key cards without trump queen; ² no kings

The probability of making 6♠ is 41.4% (56% of the 67.8% of the times trumps are 3-2 [whenever the club king is onside, with 5 spades, 3 diamonds, 3 clubs and 1 heart; and 1/8th the time the club king is offside], plus 12.5% of the 28.3% of the times trumps are 4-1; plus none of the 3.9% of the times trumps are 5-0.).  The so called "rub of the green" in bridge!  A long match can be decided by one slightly lucky deal.  Of course, the luck tends to balance out and without a complete analysis it can't be determined which side had the better of it.

Keen analysis of board 3 of the 4th quarter can be found in the New York Times Bridge Column, by Phillip Alder.

opening lead ♦8

Making 12 tricks, E/W +1430

opening lead ♦7

Down one,, E/W -50

In summary...
At table 1, declarer played a heart to the ace at trick 2.  When South followed, declarer was assured of 12 tricks.  A finesse at trick two would guarantee the contract against a four card heart holding by North, barring an unlikely ruff at trick three.  At table two, Wbridge5 would be assured of the contract with a heart finesse at trick two.  Instead declarer took a spade finesse at trick two and went down when the hearts did not come home.

For complete results go to www.computerbridge.com  You will find the 17 year history of the event, along with many publications and descriptions of computer play.  This year’s results also offer an opportunity to compare robot play against human play, with two sets of 16 boards to compare.  The complete robot play of the two sets are shown.  The complete play of the final KO is also shown along with some highlights.