Mike Whittaker, one of the people responsible for developing the program Blue Chip Bridge, reports on the second World Championship for computers.

The American Summer North American Championships in Chicago also played host to the World Computer Championships, sponsored by Baron Barclay and OKbridge. Seven programs took part in the main event and two others joined for the separate bidding competition. Despite the relatively small turnout the event was international. Programs from America, Canada, Germany, Japan and the UK took part. The favourite for the title was the American program GIB. UK hopes rested on Blue Chip Bridge, the program I have been involved in for the past four years, so please forgive the inevitable bias to this report.

A round-robin of 10-board matches produced the four semifinalists. Each match involved playing the boards twice, as N/S and as E/W. The resulting IMP score was converted to Victory Points. Blue Chip had a tough draw on Day 1 against GIB and last year's winner, Bridge Baron. The GIB match was fairly close but GIB's better card play earned it a deserved 15-5 VP win. A difference in hand evaluation earned GIB 6 IMP's on this deal:

ª	Q 6 3			ª	K 4 2
©	Q J 3			©	A 10 8 7 2
¨	A Q 10 8 3			¨	9 5 4
§	A Q			§	6 5

Both programs began 1¨-1©-2NT. What would you do now with East's hand? Blue Chip passed but GIB raised to 3NT. The opening diamond lead produced the same 11 tricks for both programs. An opening spade lead would have been best but the contract could still have been made. This was one example of GIB's ability to appreciate the overall potential of its hand and not just simply how many points it had. The program's deal simulator and ultra fast double dummy analyser combine to add a degree of judgement to GIB's bidding which, as I understand it, can sometimes overrule the initial suggested bid.

A different example of GIB's bidding capabilities appeared on this hand:

After (1ª)-2¨-(2ª) GIB psyched 2NT!

Blue Chip ignored it and jumped to a good 4ª contract but misplayed it to go one down. This cost 12 IMPs when, in the replay, GIB stopped in 3ª and made ten tricks.

With the development of faster computers, bridge programs are now more able to calculate their play. It is quite possible for them to deliberately play a squeeze. For example, Bridge Baron had such an opportunity on this hand. Playing in 3NT, Baron (East) arrived at this position after seven tricks:

ª — © — ¨ 5 2 § 10 9 3 2 ª — ª 3 © — © 6 ¨ A 10 9 6 4 ¨ K 8 § A § J 5 ª 8 © — ¨ Q J 7 § K Q

Baron led the ©6 and South was in trouble. A diamond discard would establish the suit. A club discard would allow Baron to cash the §A before returning to hand with the ¨K in order to cash the §J, squeezing South in spades and diamonds. When South discarded the ª8 Baron followed up with the ª3. South was criss-cross squeezed in the minors. However, Baron took pity after South threw the §K and simply cashed the king and ace of diamonds, conceding a diamond to South.

Did Baron not see the squeeze? We will never know because a) it did not need all six tricks to make the contract, and b) computers have not yet learned how to be sadistic. Blue Chip appeared to have shaken off any effects of jet lag and began Day 2 with a 20-0 demolition job of Genii. Good bidding judgement was evident on this hand:

ª	Q 5 2			ª	J 9 7
©	K Q 10 3			©	A 8 6 2
¨	J 8 7			¨	K 4 3
§	A K 8			§	Q J 3

Most programs reached 4© which has no real chance. Bridge Genii and Blue Chip reached the better 3NT game; Genii via 1NT(15-17)-3NT and Blue Chip after 1©-3©-3NT-Pass. Unfortunately, neither game made. The first contender to drop out was Bridge Genii. Nothing seemed to be going right for it. Mind you, it had to take the blame for one of its bad results when a dubious weak 2© opening, holding Kxxx in spades, was doubled for takeout but left in for penalties. When the puff of smoke had cleared the Genii had conceded an 800 penalty and the match.

Meanwhile, the German program, Q+ Bridge, which had done well in the 1997 event, was recovering after early losses to GIB and Bridge Baron 8. Successive wins of 20-0, 19-1 and 14-6 put it well in contention. With two rounds left to play, GIB, Bridge Baron 8 and Q+ Bridge looked certain to reach the semis. The fourth qualifying spot was a three-horse race between the UK (Blue Chip Bridge), Japan (MicroBridge) and the USA (Meadowlark Bridge).

Blue Chip made the first move by defeating Meadowlark 14-6. Andrew Robson had come over to watch the last few hands and Blue Chip immediately took the opportunity to show off with a flashy (though unnecessary) play in a 4© contract. This was the position after six tricks with North on lead:

ª 5 4 3 © K ¨ — § A 6 5 ª K J 9 8 ª 7 6 2 © — © — ¨ — ¨ J 7 5 4 § K 10 4 § — ª A Q 10 © 9 6 ¨ Q 10 § —

A small club was led from dummy and, when East showed out, Blue Chip discarded the ª10. West won with the §10 but was endplayed!

Meadowlark's last chance was against MicroBridge but the Japanese program won easily, 18-2. Meadowlark found itself in an awkward position on this hand:

With the opponents vulnerable, 1¨ is opened on your right. Assuming you overcall 1©, what do you do next when LHO raises to 4¨ passed back to you?

Just about anything was the winning answer. Eleven tricks were easy in both hearts and clubs. Meadowlark passed and lost 12 IMPs. This did not surprise me too much. Programs which use a database for bidding tend to be less impressive on very shapely hands and/or quick, preemptive auctions. They simply can't find a particularly close match within the database. This is sensible enough on grounds of frequency, reserving database space for the more common hand types, but it doesn't look too good when it happens. In the crucial last match between Blue Chip and MicroBridge, Blue Chip was 15 IMPs up after five boards but 10 IMPs down after eight. What would you lead as North after a 1ª-2©-3NT auction?

ª 9 8 7 © 10 7 ¨ A 9 3 § A J 10 6 3 ª A K Q 2 ª J 10 4 © 4 2 © A Q J 9 6 ¨ K Q J 8 ¨ 10 7 5 4 § Q 4 2 § K ª 6 5 3 © K 8 5 3 ¨ 6 2 § 9 8 7 5

MicroBridge chose to lead a spade. Blue Chip won, established its three diamond tricks immediately and finished with nine tricks. Against MicroBridge, Blue Chip led the §J. MicroBridge could have made the same nine tricks but it won, cashed four spade tricks and took the heart finesse. All square with one to play. The last hand was a 4ª contract. With a few tricks left Blue Chip looked home and dry but it overlooked the need to ruff high, got overruffed with the last trump, and it was MicroBridge that went through to the semis.

The semifinals were played over 34 boards. I have no hand records but GIB defeated Bridge Baron and Q+ won against Microbridge 8, neither match being very close. In the 54-board final GIB came through to win comfortably by 181-118 IMPs and take the title. Hans Leber, creator of Q+ Bridge, told me afterwards that GIB had gained many IMPs through superior bidding and not, as had been expected, from its card play. Could GIB do the double and add the bidding prize to its world championship title?

Each program had to bid the same 20 hands. The contracts reached were IMP-scored against the actual result when the hands had been played in a world class (human) competition. Any score approaching zero IMPs would therefore be an excellent result. Q+ had won the event last year with -40 IMPs. This year it was GIB, with a staggering +2 IMPs, that won. Blue Chip also did well and finished in second place. The top five scores were: GIB +2, Blue Chip Bridge -30, Q+ Bridge -57, Bridge Baron -84, MicroBridge -88.

MicroBridge and Q+ Bridge joined forces for a 12-board match against a team of American junior players. The juniors managed a 16-IMP win, mainly due to a slam hand. GIB had the nerve to challenge one of the world's top pairs, Zia Mahmood and Michael Rosenberg, to a match played live on the internet on OK Bridge. Zia and Rosenberg won, but not by a lot. The GIBs combined effectively on this hand, where Rosenberg (East) had to play 4©:

ª K Q 4 © A 10 ¨ 9 6 § 8 7 6 5 4 2 ª A J 9 ª 8 5 3 © 7 3 © K Q 9 8 5 2 ¨ K Q J 10 8 5 3 ¨ A 4 § K § Q J ª 10 7 6 2 © J 6 4 ¨ 7 2 § A 10 9 3

GIB1, South, led a small spade, won by the queen. GIB2 switched to a diamond, won by dummy's jack. A heart to the king won, and Rosenberg then led a club, won by GIB1. A second diamond lead was won by the ace and Rosenberg tried a spade to the jack, losing to the king. GIB2 cashed the ©A before leading a small spade to dummy's ace. Rosenberg found himself locked in dummy, forced to lead a diamond. This had the effect of promoting the ©J for GIB2 and Rosenberg finished two down.

Finally, we come to the FAQ: (Frequently Asked Question) Will computers ever triumph against top quality human opposition ? The idea has always been laughed at but I would not be too complacent. Before long the sheer calculating power of the computer will give it a definite edge over even the best human declarer in contracts that require technical expertise. However, I think that the complexities of the bidding language, the use of deception in play and defence and some abstract qualities, such as table presence, will keep the humans ahead, at least for a while.

The next computer world championships are scheduled for January 2000 in Bermuda. Blue Chip has already booked his ticket and is hard at work improving his game, and I have less than 18 months to dream up an excuse to attend. I'm sure I'll think of something.