Leaderboard 8X99   Main

Bot’s Eye Views

by Richard Pavlicek

From February 2001 to August 2006, I tested some of the well-known bridge computer programs on my monthly Bidding Polls and Play Contests. Results were reported in a “Bot’s Eye View,” not only to see how they scored against each other but to see if their skills had approached human levels. Not! People are safe for now. This page contains all 66 reports: 33 on bidding, 33 on card play.

Bidding Polls

Play Contests

Each test report consisted of six problems, scored on a 1-to-10 scale, hence a perfect score was 60, though the highest ever achieved by a bot was 56. Most of the programs had settings for skill level (thinking time) which was set to the highest not to exceed 30 seconds per call or 60 seconds per play — pretty generous as such a pace would be way too slow for tournament play. Thinking time is also used to break ties if two bots have the same score; faster gets the edge.

Participants (alphabetical)

Program	Author/Creator
Blue Chip Bridge	Ian Trackman and Mike Whittaker
Bridge Baron	Tom Throop and Stephen Smith
Bridge Buff	Doug Bennion
Finesse Bridge (defunct)	Mark and Aaron Marin
GIB	Matthew Ginsberg
HAL	Someone with a warped mind
Jack	Hans Kuijf
Micro Bridge	Tomio and Yumiko Uchida
Q-plus Bridge	Hans Leber

HAL 9000 is a fake bot included for amusement — call it payback for its misdeeds in 2001: A Space Odyssey. HAL’s scores were appreciated, not only for staking a perpetual claim on last place but for lowering the average bot score, so the true bots were mostly above average.

Leaderboard 8X99   Main

Top   Bot’s Eye Views

Bidding Polls

Following are the 33 Bidding Polls on which bots were tested. Click on the table title to see the actual bidding problems.

March 2001 Bidding Poll
294 humans avg 47.39

Rank	Score	CC	Program version	1	2	3	4	5	6
1	55	US	Bridge Baron 11.0	D	2	2	5	3	3
2	51	DE	Q-plus Bridge 6.1	4	2	2	4	3	3
3	50	UK	Blue Chip Bridge 3.4.0	4	2	2	5	3	3
4	48	JP	Micro Bridge 9.01	P	2	2	5	3	2 NT
5	47	US	GIB 4.1.2	4	2	2	6	4	2
6	42	CA	Bridge Buff 8.0	4	2	2	4 NT	P	2 NT
7	34	US	Finesse Bridge 2.5	P	2	2	5	P	3
8	15	US	HAL 9000	4	P	2	5	2 NT	3 NT

Bridge Baron topped all the bots with an excellent score of 55. Second place went to Q-plus Bridge with 51, and third went to Blue Chip Bridge with 50. The scores and rankings, based on six specific problems, are not necessarily indicative of each program’s overall capability. The only thing certain is that if you managed to equal HAL’s score, you should be locked up with the key thrown away if you ever touch a deck of cards again.

May 2001 Bidding Poll
519 humans avg 47.34

Rank	Score	CC	Program version	1	2	3	4	5	6
1	48	US	Bridge Baron 11.0	3 NT	4	5	3	3	D
2	45	JP	Micro Bridge 9.01	3 NT	5	D	P	3	D
3	45	DE	Q-plus Bridge 6.1	3 NT	4	D	2	3 NT	D
4	43	US	Finesse Bridge 2.5	3 NT	4	P	2	4	D
5	37	US	GIB 4.1.2	3 NT	6	4	2	3	D
6	34	CA	Bridge Buff 8.0	3 NT	4	5	2	3 NT	P
7	11	US	HAL 9000	5	4 NT	4	3	4	P

Congratulations to Bridge Baron, which topped all the bots on a tough set of problems. This was helped by “staying on the charts,” as most of the other programs had an errant answer, scoring zero: GIB bid 6  on Problem 2; Finesse Bridge passed on Problem 3; Micro Bridge passed on Problem 4; and Q-plus Bridge and Bridge Buff bid 3 NT on Problem 5. HAL, of course, did not suffer from these occasional aberrations (they were quite steady).

July 2001 Bidding Poll
583 humans avg 47.93

Rank	Score	CC	Program version	1	2	3	4	5	6
1	46	US	Finesse Bridge 2.5	P	3 NT	4	P	3	3
2	43	UK	Blue Chip Bridge 3.4.3	P	P	3	2	3	3
3	42	US	GIB 4.1.2	P	2 NT	3	2	3	2 NT
4	41	JP	Micro Bridge 9.01	P	2	3	2	3 NT	3
5	41	DE	Q-plus Bridge 6.1	P	2 NT	P	2	3 NT	3
6	34	CA	Bridge Buff 8.0	P	2	3	2	3 NT	P
7	32	US	Bridge Baron 11.0	P	2	2 NT	3	3	3
8	7	US	HAL 9000	2 NT	3	2 NT	3 NT	3	P

Congrats to Finesse Bridge! Curiously, this month’s bot champ is the only free program of the lot. (HAL is actually better than free as its latest marketing strategy is to pay you to take it.) Bridge Baron, the winner of my last two bidding polls, had an off month; or at least the problems were not to its liking. The fact that none of the bots broke average suggests that human superiority will be around for awhile.

Problem 2 created a predicament because only one program (Q-plus Bridge) had the systemic option to allow (and hence, to understand) a natural, limited 2  opening (a la Precision). Rather than skip the problem, it seemed like more fun to let the other programs wing it; so I fed them the same auction as if 2  were strong and artificial. Micro Bridge apparently took the double as takeout for the majors (else why 2 ?) and most of the others as some kind of takeout (only Blue Chip Bridge left it in). Obviously, this taints the comparisons; but as I see it, my polls don’t allow people to abstain, so why should the bots be any different?

September 2001 Bidding Poll
577 humans avg 47.64

Rank	Score	CC	Program version	1	2	3	4	5	6
1	41	CA	Bridge Buff 8.0	4	4	4	2	4 NT	2
2	40	UK	Blue Chip Bridge 3.4.3	3	3	4	2	4 NT	4
3	39	US	GIB 4.1.12	2 NT	4	4	2	4	2
4	37	US	Bridge Baron 11.0	3	2	4	1 NT	4 NT	2
5	37	DE	Q-plus Bridge 6.1	2	3	4	D	6	2
6	35	JP	Micro Bridge 9.01	2	2	4	2	4 NT	D
7	34	US	Finesse Bridge 2.5	4	2	3	2	6	2
8	14	US	HAL 9000	2 NT	P	4 NT	1 NT	5	P

Congratulations to Bridge Buff, which eked out a narrow win, aided somewhat by my scoring generosity. On Problem 1 its off-the-chart 4  bid actually deserved zero, but my policy is not to award less than the worst option listed since the poll was multiple choice. (Obviously, if the bots understood this they would choose one of the listed calls.) Bridge Buff is certainly on a roll, as it also won my play contest last month.

The bot results in general were poor compared to my previous bidding polls; not even one came close to average. Evidently, these were tough problems. At least HAL was consistent.

November 2001 Bidding Poll
714 humans avg 46.28

Rank	Score	CC	Program version	1	2	3	4	5	6
1	46	JP	Micro Bridge 9.01	D	3	P	4	P	4
2	45	CA	Bridge Buff 8.0	D	3	5	4	P	4
3	45	US	GIB 4.1.12	D	3	5	4	4	P
4	44	UK	Blue Chip Bridge 3.4.3	D	3	P	4	4	P
5	41	DE	Q-plus Bridge 6.1	5	3	P	2	P	P
6	39	US	Finesse Bridge 2.5	D	2	P	4	3	P
7	37	US	Bridge Baron 11.0	D	3	P	4	3	5
8	11	US	HAL 9000	P	D	5 NT	3 NT	2	4

Congratulations to Micro Bridge, which eked out a narrow win with a score of 46. Tied for second were Bridge Buff and GIB with 45. All the bot scores (well, except the bogus HAL) were closely bunched, and it is significant to note they were all below average. Chalk one up for the human race.

In a few cases, a bot’s choice was off the chart. While this might deserve a score of zero, my policy is to award the equivalent of the worst option listed. Obviously, if the bot understood the multiple-choice format, it could not score worse than that — although HAL certainly makes a good effort each month.

January 2002 Bidding Poll
767 humans avg 46.93

Rank	Score	CC	Program version	1	2	3	4	5	6
1	44	JP	Micro Bridge 9.01	4	P	3	5	4	D
2	43	US	GIB 4.1.12	3	P	2 NT	3	2 NT	D
3	41	CA	Bridge Buff 8.0	2	5	3 NT	4	3	4
4	41	US	Finesse Bridge 2.5	3	5	3	5	2	4
5	39	US	Bridge Baron 11.0	3	P	3	5	3	4
6	35	DE	Q-plus Bridge 6.1	4	5	2	5	3	P
7	33	UK	Blue Chip Bridge 3.4.3	4	5	3	5	2	P
8	9	US	HAL 9000	2	3 NT	4	3	2	3

Congratulations to Micro Bridge, which eked out a narrow win with a score of 44. Actually, all the bot scores (well, except for you know who) were closely bunched, but it is significant to note they were all below the average human score. Good. Let’s keep those tin cans in their place.

In a few cases, namely the 2 and 3 bids on Problem 5, the bot’s actual choice was not listed as an option. Were these bids an aberration? Or did they try to spring some special convention out of the blue? Who knows. In any event, my policy is to score these the same as the worst option listed since the bot could never do worse than that if it understood the multiple-choice format.

March 2002 Bidding Poll
813 humans avg 47.33

Rank	Score	CC	Program version	1	2	3	4	5	6
1	49	CA	Bridge Buff 8.0	P	3 NT	3 NT	4	1	C
2	46	US	GIB 4.1.12	1 NT	3 NT	4	5	1	C
3	41	UK	Blue Chip Bridge 3.4.3	P	3	5	P	1	C
4	41	DE	Q-plus Bridge 6.1	P	3	P	P	1	D
5	39	JP	Micro Bridge 9.01	2	P	5	4	P	A
6	35	US	Finesse Bridge 2.5	P	3 NT	P	P	1	C
7	33	US	Bridge Baron 11.0	D	P	P	5	1	A
8	13	US	HAL 9000	2	2	P	3 NT	5	E

Congratulations to Bridge Buff, not only for topping the bots but also for being the only bot to beat the average human score. Bridge Buff also deserves recognition as the only bot to have won both a bidding poll and a play contest since I began doing this about a year ago. Overall, the bot scores were mediocre this month, which attests to the difficulty of the problems.

On Problem 6, only one bot (Q-plus Bridge) understood the natural 2 opening (11-15). Kudos to creator Hans Leber (Germany) for his versatile programming, allowing the user to play or defend against a wide variety of systems. Most of the other bots passed the 2  opening expecting it to be strong. Rather than junk the problem, I reposed it with a 1  opening (followed by a 3  raise). While certainly not the same, it was close enough for the tin cans, as they generally found reasonable actions.

It was also interesting to note that, aside from the incompatibility on Problem 6, none of the bots drifted off the charts this month. Usually there are at least few aberrations, causing me to have to add additional calls to my scoring program. Even HAL was impressive: Imagine scoring 13 when you can’t even count to 13.

May 2002 Bidding Poll
844 humans avg 46.42

Rank	Score	CC	Program version	1	2	3	4	5	6
1	51	UK	Blue Chip Bridge 3.4.3	4 NT	2	3 NT	2	2	A
2	46	US	GIB 4.1.12	7	2	3 NT	4	2	A
3	42	DE	Q-plus Bridge 6.1	4	2	5	2	2	C
4	39	JP	Micro Bridge 9.01	4 NT	3	3 NT	1	4	A
5	31	CA	Bridge Buff 8.0	4 NT	3	5	1	4	G
6	29	US	Finesse Bridge 2.5	5	2	3	1	3	A
7	26	US	Bridge Baron 11.0	4 NT	3	4 NT	1	4	G
8	10	US	HAL 9000	5 NT	4	P	P	4	D

Congratulations to Blue Chip Bridge, which not only topped all the bots this month but was the only bot to beat the average human score. Well done for a difficult set of problems. Of course, I also must congratulate HAL for its month-to-month consistency — some things in life are uncertain; HAL isn’t one of them.

As usual, several of the bot calls went off the charts, and a few were even amusing. On Problem 1, GIB went for the gusto with a jump to 7 , and Q-plus and Finesse Bridge found strange bids of 4  and 5 , respectively. On Problem 3, Bridge Baron evidently had an aberration using Blackwood, or maybe it was just mad because I forced it to bid 2  with A-K-J. On Problem 5, Finesse Bridge managed only a feeble raise to 3 , though I suppose one could argue it has tactical merit.

Problem 6 also had its rebels (indicated as G). Bridge Baron chose the wimpy route, overcalling 1 and then bidding only 2 . Bridge Buff took the other extreme, doubling and then bidding 5 . Talk about a difference in evaluation! While these and other aberrant choices might deserve zero, my policy is to score them equal to the lowest award among the choices offered because the bots are not programmed for the multiple choice format.

July 2002 Bidding Poll
875 humans avg 46.94

Rank	Score	CC	Program version	1	2	3	4	5	6
1	50	UK	Blue Chip Bridge 3.4.3	3	4	4	P	D	3
2	50	JP	Micro Bridge 9.01	2 NT	5	4	3	D	3
3	49	US	GIB 4.1.12	3	4	4	P	D	2
4	47	CA	Bridge Buff 8.0	3 NT	4	4	P	D	2
5	45	US	Bridge Baron 11.0	3	P	3	P	D	3
6	45	DE	Q-plus Bridge 6.1	3 NT	4	3	2 NT	D	2
7	31	US	Finesse Bridge 2.5	3 NT	4	P	4	D	5
8	10	US	HAL 9000	5	6	P	4	5 NT	2

Congratulations to Blue Chip Bridge and Micro Bridge, which tied with 50 in a photo finish, with GIB just one point behind. Since the date and time of submission is meaningless for the bots, I broke the tie by consistency (i.e., highest worst score) which gave Blue Chip the win. The aforementioned bots, as well as Bridge Buff, all topped the average human score. Even HAL should be commended on hitting double figures; in fact, its August advertising campaign now cites this poll in its claim to be the “Perfect 10” in bridge computers.

On Problem 4, none of the bots were capable of understanding the Roman 2 opening (5+ hearts, 4+ clubs); but rather than skip the problem, I let them assume it was a weak two-bid. This seemed harmless, as the situations are analogous. Besides, none of the bots had an abstain button, so my directions were simple: Shut up and bid!

The bots did extremely well in staying on the charts this month. The only wayward call came from Finesse Bridge, which was really feeling its oats on Problem 6, jumping to 5 . For scoring purposes, unlisted calls receive the same award as the lowest of the listed calls, since the bots are not programmed for the multiple-choice format.

September 2002 Bidding Poll
900 humans avg 49.37

Rank	Score	CC	Program version	1	2	3	4	5	6
1	53	US	GIB 6.1.0	4	3	3 NT	4	3	F
2	49	US	Bridge Baron 11.0	5	3	3 NT	4	3	F
3	48	UK	Blue Chip Bridge 4.0.0	P	3	3 NT	4	4	F
4	40	US	Finesse Bridge 2.5	P	P	3 NT	4	3	B
5	36	DE	Q-plus Bridge 6.1	P	P	3 NT	4	4	F
6	32	CA	Bridge Buff 8.0	P	P	3 NT	4	3 NT	F
7	32	JP	Micro Bridge 9.01	P	P	3 NT	P	3	B
8	10	US	HAL 9000	P	2	4 NT	P	4 NT	C

Congratulations to GIB, which topped the bots with a fine score of 53, and was the only bot to beat the average human score. On Problem 1, it was especially enlightening to see GIB come up with 4  (slam try after opening a 10-point hand), the kind of bid that requires vision beyond that of a typical bidding database. GIB also continued to reach the laydown slam. Bridge Baron was the only other program to make a move over 4 ; alas, it stopped in 5 . HAL tried to steal the slam by passing 4  and then adding 120 to a 60 partscore. Sorry, HAL; it doesn’t work that way. HAL didn’t take the setback lightly, and I nearly got electrocuted in the process.

On Problem 5, Bridge Buff had no option but to interpret 2 NT as Jacoby (spade raise) so its rebid was always 3  to show a singleton. To be fair, I tried switching the spades and diamonds to create a 1  opening, and gave North a similar 14-count. It now responded 2 NT naturally; alas, South raised to 3 NT, missing the 33-HCP slam.

Q-plus Bridge also had a hang-up on Problem 5, insisting on jumping to 4 over the natural 2 NT response. Out of curiosity where this might lead, I let it continue: North bid 4 NT (apparently reading 4  as a mountain); South answered 5  (must be two aces); but that was the end as North passed. OK, so there’s room for improvement. I’ll bet you missed a slam once, too.

November 2002 Bidding Poll
939 humans avg 46.74

Rank	Score	CC	Program version	1	2	3	4	5	6
1	51	UK	Blue Chip Bridge 4.0.1	3	P	2	P	2 NT	B
2	47	US	GIB 6.1.0	3	4	D	P	P	D
3	46	US	Bridge Baron 11.0	5	4	2	P	3	C
4	46	DE	Q-plus Bridge 6.1	2 NT	D	2	P	2 NT	D
5	43	JP	Micro Bridge 9.01	3	4	3	3	2	B
6	41	CA	Bridge Buff 8.0	3	4	2	P	4	B
7	27	US	Finesse Bridge 2.5	P	D	2	3	3	G
8	8	US	HAL 9000	2 NT	5	2	3	2	A

Congratulations to Blue Chip Bridge, which topped the bots with an excellent score of 51. GIB was the only other bot to beat the average human score.

Problem 2 created a problem in presentation because of the weak jump raise to 3 . Even the programs that allowed inverted minor raises as an option did not allow them in competition. Therefore, I devised a workaround: I had North respond 2  (weak raise) and South rebid 2 ; then I made West jump to 3 , which is passed around. While not identical, the sequence is practically the same — and certainly close enough for tin cans (hehe).

Problem 3 was also slightly flawed because some of the bots had no option to play support doubles. Therefore, the double wasn’t even a possibility; hence, there was no opportunity to score 10. Shall we all shed a tear for the poor little bots? HAL insisted on playing “inverted support doubles,” so a double would show four trumps, and its chosen raise to 2  promised three.

As usual, some of the choices went off the charts: On Problem 1, Finesse Bridge elected to pass. A shrewd tactical move? No, I don’t think so; more likely a programming deficiency. On Problem 5, Bridge Buff jumped all the way to 4  (a gross overbid). On Problem 6, Finesse Bridge chose to pass and respond 2  to the 1  opening — weird. For scoring purposes, errant choices are given the same award as the lowest listed call.

January 2003 Bidding Poll
1035 humans avg 47.05

Rank	Score	CC	Program version	1	2	3	4	5	6
1	49	JP	Micro Bridge 9.01	2 NT	4	D	4	3	B
2	48	CA	Bridge Buff 8.0	2 NT	P	D	3 NT	D	B
3	47	NL	Jack 2.0	4	4	D	3 NT	P	B
4	44	UK	Blue Chip Bridge 4.0.1	3 NT	P	D	5	2	B
5	43	US	GIB 6.1.3	3	P	D	4	2	B
6	38	US	Bridge Baron 11.0	3	P	D	3 NT	2 NT	B
7	38	DE	Q-plus Bridge 7.1	3 NT	4	D	3	3	G
8	14	US	HAL 9000	3	4	2 NT	6	4	D

Congratulations to Micro Bridge, which eked out a narrow win with 49 on this tough set of problems. Bridge Buff scored 48 to be the only other bot to beat the average human score.

Problem 2 created a presentation difficulty because of the Roman 2 opening. None of the bots had this convention in their data banks, so I let them all assume a normal weak two-bid. For practical purposes, the sequences are analogous, and this assured an even playing field for testing.

Problem 2 also brought out a common weakness among computer bidding programs: the tendency to undervalue distributional hands. The great majority of humans realized the potential for game (3 NT or 5 ), but the bots did not, and some even passed. HAL, of course, was the exception as it made a slam try with 4 . HAL called this “Gerbil” but gave no explanation other than something about mouse pads and rodents. Lost me.

The bots behaved quite well this month in staying on the charts. The only aberration came from Q-plus Bridge on Problem 6, which strangely interpreted the double as takeout and jumped to 4 . Wow. Would partner fall out of his chair, or what? Hence, it gets the infamous Choice G. For scoring purposes, errant choices get the same as the lowest listed award.

March 2003 Bidding Poll
999 humans avg 45.27

Rank	Score	CC	Program version	1	2	3	4	5	6
1	54	UK	Blue Chip Bridge 4.0.5	2 NT	P	3	5	3	C
2	52	CA	Bridge Buff 8.0	3 NT	P	2 NT	6	4	C
3	48	US	GIB 6.1.3	4	4 NT	3	P	3	B
4	47	JP	Micro Bridge 9.01	3	6	3 NT	5	4	C
5	45	NL	Jack 2.0	3	P	3 NT	D	3	B
6	42	DE	Q-plus Bridge 7.1	2 NT	4	2	5	3	C
7	37	US	Bridge Baron 11.0	P	P	1 NT	P	2 NT	F
8	8	US	HAL 9000	3	6	2	D	2	D

Congratulations to Blue Chip Bridge, which kept up its fine bidding record with an outstanding 54. Not too far behind was Bridge Buff with 52. Two other bots — GIB and Micro Bridge — managed to beat the average human score.

Problem 4 was the most interesting, as two of the bots (GIB and Bridge Baron) found the correct forcing pass. I was curious whether this was really a clever maneuver or just blind luck, so I followed it up. The GIB North jumped to 6  (reasonable) and South bid 6  as intended all along. Would North bid seven? No, it passed after some deliberation — still, a good show. The Bridge Baron North chose to double 5  (a strange view with 1=4=7=1 shape) and South passed to defend 5  doubled — not such a good show.

The bots behaved fairly well this month, going off the charts only twice: Bridge Baron with 1 NT on Problem 3, and Q-Plus Bridge with 5  on Problem 4. Perhaps it was just my lack of imagination not to include these calls. For scoring purposes, errant choices get the same award as the lowest listed choice.

May 2003 Bidding Poll
1036 humans avg 44.62

Rank	Score	CC	Program version	1	2	3	4	5	6
1	48	US	Bridge Baron 11.0	D	P	5	3	4 NT	E
2	46	JP	Micro Bridge 10.01	5	2	5	2 NT	4 NT	E
3	44	UK	Blue Chip Bridge 4.0.6	4	2	5	2 NT	6 NT	E
4	40	CA	Bridge Buff 8.0	5	P	5	3	3 NT	A
5	38	DE	Q-plus Bridge 7.1	D	2 NT	5	3	4	E
6	37	US	GIB 6.1.3	5	D	6	D	3 NT	A
7	35	NL	Jack 2.0	4	2	D	D	3 NT	E
8	6	US	HAL 9000	6	3 NT	P	P	3 NT	G

Congratulations to Bridge Baron, which surged to the fore once again after a long dry spell. The problems were exceptionally tough this month, and Bridge Baron’s winning score of 48 is quite respectable. The only other bot to beat the average human score was Micro Bridge, which grabbed second place with 46. The bots also behaved well, staying on the charts with all their answers.

Problem 1 proved to be the most challenging of the set, as only GIB came up with a reasonable call (5  cue-bid). All the other bots chose a unilateral suit bid or, even worse, doubled the 4  opening (and played it there). Bidding over preempts is difficult enough for humans, so I guess it’s no surprise that bots are stumbling, too.

This month I implemented a new award for consistency, and the winner is HAL. It might seem difficult to score exactly 1 point on each problem, but for HAL it is effortless. Despite the lowest total score ever of 6, HAL is a true mainstay for the game of bridge. Translation: If your “main” pastime is bridge, “stay” away from HAL!

July 2003 Bidding Poll
1083 humans avg 46.21

Rank	Score	CC	Program version	1	2	3	4	5	6
1	50	DE	Q-plus Bridge 7.1	3	D	2 NT	3 NT	1 NT	B
2	47	JP	Micro Bridge 10.01	2	4	2 NT	3 NT	2	F
3	46	US	Bridge Baron 11.0	3	4	2 NT	3 NT	2	I
4	42	UK	Blue Chip Bridge 4.0.6	2	P	2 NT	3 NT	P	H
5	42	NL	Jack 2.0	5	5	3	3 NT	1 NT	B
6	42	US	GIB 6.1.3	P	4	1	3	1 NT	A
7	39	CA	Bridge Buff 8.0	5	5	2	3	2	C
8	10	US	HAL 9000	P	3 NT	2	4	2	G

Congratulations to Q-plus Bridge, which topped the competition this month with an excellent score of 50. The only other bot to beat the average human score was Micro Bridge, taking second place with 47. Bridge Baron came a close third with 46.

Wild distributions proved to be a stumbling block for several of the programs. GIB’s peculiar pass on Problem 1 (with nine clubs) would seem like a tricky tactical maneuver if chosen by a human; but by a bot it suggests a flaw in its bidding database or hand evaluation. Similarly, on Problem 6, Blue Chip Bridge overcalled 2  (with eight clubs) but then chose to pass on the next round (indicated as Choice H) rather than compete. Bridge Baron instead chose a bizarre Michaels cue-bid (indicated as Choice I) with 8-4 shape.

For scoring purposes, unlisted choices get the same award as the lowest listed choice. While some errant choices may indeed deserve zero, this wouldn’t be fair because there is no way to instruct the bots to my multiple-choice format. Ties in the bot rankings are broken by thinking time, with the advantage going to the bot that was faster.

Problem 3 might be considered unfair, as two programs (Jack and Bridge Buff) did not have the unusual 2 NT overcall available as a bidding convention; hence, they had no chance to receive the top award. Nonetheless, this doesn’t get my sympathy or any scoring adjustment because the convention is almost universally accepted. (I suspect it will be included in future versions.) Curiously, GIB preferred to overcall 1  with 6-6 shape despite the availability of the unusual 2 NT. HAL, of course, preferred to bid clubs first, saving the heart suit for what it called a “high-level perverse.”

September 2003 Bidding Poll
1117 humans avg 47.31

Rank	Score	CC	Program version	1	2	3	4	5	6
1	54	US	Bridge Baron 11.0	5	4	3	P	3 NT	D
2	49	JP	Micro Bridge 10.01	4	4	3	4 NT	3 NT	D
3	49	DE	Q-plus Bridge 7.1	4	4	3	4 NT	3 NT	D
4	48	US	GIB 6.1.3	3	4	4	P	3 NT	C
5	46	CA	Bridge Buff 8.0	3 NT	4	3	P	3 NT	D
6	40	UK	Blue Chip Bridge 4.0.7	4	P	3	4 NT	P	A
7	36	NL	Jack 2.0	4	4	3	5	3 NT	B
8	11	US	HAL 9000	3 NT	3 NT	5	4 NT	4	B

Congratulations to Bridge Baron, which topped the competition this month with a superb score of 54. Micro Bridge and Q-plus Bridge were next with 49 (including identical answers to each problem). The only other bot to beat the average human score was GIB with 48. Bridge Baron also surged into the overall lead by virtue of this outing.

Previous overall champ Blue Chip Bridge had an off month, largely due to a disaster on Problem 2. After the 3  cue-bid was doubled and passed around, it chose to pass. Ouch, with a known 10-card club fit. In fairness, when I forced it to bid 3 , the program announced, “This bid is not understood”; but even so, it should be understood, or at least the program should deduce not to play in 3  doubled. Perhaps it’s just a database glitch that is easily fixed.

For scoring purposes, unlisted choices get the same award as the lowest listed choice. While some errant choices may indeed deserve zero, this wouldn’t be fair because there is no way to instruct the bots to my multiple-choice format. Ties in the bot rankings are broken by thinking time, with the advantage going to the bot that was faster.

Problem 6 posed an interesting predicament: How to determine which call a bot disliked most. Well, I couldn’t care less about android opinions, so I just had each bot bid the South hand. The first call made differently from the problem auction decided the issue. Only Blue Chip Bridge replicated all four calls. Most bots went astray at 3 , preferring a simple 3  instead — although Micro Bridge was feeling its oats with a jump to 4 . Jack took a strange view, electing to open the bidding 3 . HAL caused the most trouble as it refused to make any bid, claiming it “disliked me most.” Well, that was simple to fix: I just mentioned another trip to the piranha tank, and it bid up a storm.

November 2003 Bidding Poll
1149 humans avg 46.18

Rank	Score	CC	Program version	1	2	3	4	5	6
1	49	JP	Micro Bridge 10.01	2	5	P	3	7	C
2	47	NL	Jack 2.0	2	4	P	1	7	C
3	45	US	Bridge Baron 11.0	2	4	P	3	5	G
4	44	CA	Bridge Buff 8.0	2	P	P	1	5	A
5	44	DE	Q-plus Bridge 7.1	2	P	3	2	8	C
6	40	US	GIB 6.1.3	2	P	3	3	5	I
7	39	UK	Blue Chip Bridge 4.0.8	2	P	P	1	6	G
8	11	US	HAL 9000	D	5	3 NT	1	5	H

Congratulations to Micro Bridge, which topped the competition this month with a decent score of 49. The only other bot to beat the average human score was Jack with 47, though all the scores were bunched pretty close — except for HAL of course. The win also catapulted Micro Bridge to the top of the overall standings by a whisker over Bridge Baron.

The bots did well this month in staying on the chart. Only one exception: On Problem 6, GIB did not pass or double but chose to bid 5 . Wow. This seems egregious but certainly solves the opening-lead problem. In fairness, I think GIB may have been misprogrammed that its opponents were two HAL-9000 machines, which would make 5  laydown.

For scoring purposes, unlisted choices get the same award as the lowest listed choice. While some errant choices may indeed deserve zero, this wouldn’t be fair because there is no way to instruct the bots to my multiple-choice format. Ties in the bot rankings are broken by thinking time, with the advantage going to the bot that was faster.

Problem 5 posed a challenge of how to have bots choose the worst bid, so I set up a special little game just for the tin-heads. It was widely agreed that four bids were bad, so I gave each a point value based on the award scale. Each bot then scored 4 points if it bid 4  instead of the ugly 4 ; 3 points if it bid 3  instead of 3 ; 2 points if it passed 4  instead of bidding 5  (or anything else); and 1 point if it bid 1  instead of 1 . Thus, there were 10 points available for doing nothing bad, and each bot’s total is shown. Q-plus Bridge fared the best, succumbing only to the failure to pass 4 .

January 2004 Bidding Poll
1257 humans avg 48.24

Rank	Score	CC	Program version	1	2	3	4	5	6
1	52	UK	Blue Chip Bridge 4.1.0	2	P	D	5	P	B
2	46	US	GIB 6.1.3	4	4 NT	3	3	P	C
3	44	NL	Jack 2.0	2	P	3	3	P	B
4	43	JP	Micro Bridge 10.01	4	P	3	3	P	C
5	39	CA	Bridge Buff 11.0	2	4 NT	3	5	P	B
6	39	DE	Q-plus Bridge 7.1	4	4 NT	3	3	4	E
7	38	US	Bridge Baron 14.0	3	P	4	5	P	E
8	10	US	HAL 9000	4	5 NT	5	P	4 NT	E

Congratulations to Blue Chip Bridge, which is back in the winners’ circle with a fine score of 52, as well as the only bot to beat the average human score. Well done! GIB was a distant second with 46, followed by Jack (the current card-play champ) with 44. Despite a mediocre showing this month, Bridge Baron narrowly held its overall lead over Blue Chip Bridge and Micro Bridge.

The only errant call* this month occurred on Problem 1, where two bots (Jack and Bridge Buff) chose a paltry 2  response. If an expert made such a bid, it might be described as a brilliant tactical move, but I doubt this was the case. More likely, the ostrich-like hand found a gap in their bidding databases or evaluation methods. I’m sure the glitch will be investigated and fixed in future versions.

*For scoring purposes, unlisted choices get the same award as the lowest listed choice. While some errant choices may indeed deserve zero, this wouldn’t be fair because there is no way to instruct the bots to my multiple-choice format. Ties in the bot rankings are broken by thinking time, with the advantage going to the bot that was faster.

On Problem 6, each bot that passed originally was given the hypothetical sequence, Pass 1 Pass 1 (opponents bidding). I was happy to see they all came back in the hunt, choosing either Michaels or an unusual notrump to show a two-suiter, or a jump to 3  to show a one-suiter. The only bots to score poorly on Problem 6 were those that opened the bidding — ouch, they all chose 1 ! Even HAL bid 1 , claiming that in its latest system (HAL 9000.71) this was canape, to be followed by a delayed preempt to 5 . When I suggested this was crazy, HAL brushed it off, claiming that if 5  is doubled, it would run back to hearts.

March 2004 Bidding Poll
1337 humans avg 45.28

Rank	Score	CC	Program version	1	2	3	4	5	6
1	52	NL	Jack 2.03	3 NT	4	4	3 NT	6	B
2	45	UK	Blue Chip Bridge 4.1.0	2 NT	P	3 NT	2 NT	D	B
3	40	JP	Micro Bridge 10.02	2	P	5	3 NT	D	A
4	39	US	Bridge Baron 14.0	3 NT	P	5	2 NT	P	B
5	37	DE	Q-plus Bridge 7.1	D	P	5	2 NT	D	F
6	36	US	GIB 6.1.3	2 NT	D	4	2 NT	6	H
7	34	CA	Bridge Buff 11.0	3 NT	P	4	3	D	H
8	10	US	HAL 9001	D	3 NT	4	4	P	E

Congratulations to Jack, winning easily this month with a fine score of 52, and also the only bot to beat the average human score. Blue Chip Bridge was second with 45. The win moved Jack into third place in the overall standings, closing in fast on Bridge Baron and Micro Bridge. Jack is also the current overall bot champ in my play contests.

The only errant call this month occurred on Problem 6, where GIB and Bridge Buff chose to use Stayman and pass when opener showed four hearts. Indeed, several human respondents also suggested this possibility. Usually, going off the chart in a bidding poll gets the same award as the worst listed option, but in this case it deserved better. Listed below as Choice H, I gave it 4.

HAL came out with a new version (9001) this month, claiming to be much improved — a “perfect 10” by its own account, which my tests seem to confirm as well. One of its new features is a mode called “auto-HAL,” which bids and plays hands in the blind. I no longer need to enter the cards! (Not that this ever mattered much anyway.)

May 2004 Bidding Poll
1265 humans avg 45.06

Rank	Score	CC	Program version	1	2	3	4	5	6
1	51	US	Bridge Baron 14.0	2	3	4	5	3 NT	1
2	46	NL	Jack 2.03	2	2	4	P	3	2
3	42	JP	Micro Bridge 10.02	2	2	4	P	3	P
4	40	DE	Q-plus Bridge 7.1	2	D	2	4 NT	3	2
5	34	UK	Blue Chip Bridge 4.1.2	2 NT	3	P	5	5	4
6	34	CA	Bridge Buff 11.0	3	4	4	P	5	P
7	32	US	GIB 6.1.3	2	2	4	6	6	1
8	11	US	HAL 9001	2	P	P	6	3	3

Congratulations to Bridge Baron, which won rather convincingly this month with a fine score of 51. The only other bot to beat the average human score was Jack with 46. The win easily keeps Bridge Baron atop the overall standings, followed by Jack in second place.

There were two errant calls* this month. On Problem 4, Bridge Buff, Micro Bridge and Jack all passed partner’s 4  bid, which is outlandish after both players have cue-bid the enemy suit, and diamonds were never bid. I’ve noticed this to be a common glitch among computer programs; once an auction graduates beyond its fixed rules, the tendency is to pass too often. Perhaps, they need to be programmed with the familiar expert advice, “If you’re not sure what a bid means, don’t pass!” Alas, then you’ll need a way to stop the buggers from reaching 7 NT on every hand.

*For scoring purposes, unlisted choices get the same award as the worst listed choice. While some errant calls might indeed deserve zero, this wouldn’t be fair because there is no way to instruct the bots to the multiple-choice format.

The other errant call was on Problem 5, where GIB jumped all the way to 6 . Wow. Has Ginsberg tweaked its card-play algorithms to new heights? No, I think it just thought it was playing against HAL, which has been reprogrammed in Cliche++ (similar to C++). Holding, say, A-x-x-x x-x-x K-x x-x-x-x, HAL would surely lead the K, either to “cut down the ruffing power,” or just simply for, “when in doubt, lead trumps.” Easy slam.

July 2004 Bidding Poll
1185 humans avg 45.43

Rank	Score	CC	Program version	1	2	3	4	5	6
1	48	DE	Q-plus Bridge 7.1	D	P	4	5	4	3 NT
2	48	US	GIB 6.1.3	D	5	3	6	4 NT	N
3	45	CA	Bridge Buff 11.0	P	P	3	5	4	P
4	41	JP	Micro Bridge 10.02	D	P	3	4 NT	4	P
5	39	US	Bridge Baron 14.0	D	P	4	4 NT	4	N
6	37	NL	Jack 2.04	D	P	3	P	4 NT	N
7	33	UK	Blue Chip Bridge 4.2.0	1 NT	P	3	P	4	N
8	24	US	HAL 9001	P	P	P	P	P	P

Congratulations to Q-plus Bridge, which scored a respectable 48 and was faster to win by tiebreaker over GIB with the same score. Q-plus Bridge and GIB were also the only bots to beat the average human score, as the bot scores were generally mediocre on this tough problem set.

Problem 2 proved to be the bane of the bots — excluding GIB — as they elected to pass partner’s double of 4 . Some assumed the double was penalty (absurd by any standards) and this interpretation could not be changed by any program settings. No doubt this will inspire some of the programmers to amend their bidding-rule database for future versions.

There were only a few errant calls* this month. On Problem 4 (the hand with 6-6 in the red suits) Micro Bridge and Bridge Baron chose to bid 4 NT (Blackwood) — a choice I decided not to insult you with as an option. On Problem 5, GIB and Jack also chose 4 NT; but here it was quantitative (over 2 NT), and I awarded it 4 since it’s better than a few of the listed options. In retrospect, I probably should have included it.

*For scoring purposes, unlisted choices receive at least as much as the worst listed choice — it wouldn’t be fair to award less because bots are unaware of the multiple-choice format — and may receive more if merited.

HAL was perturbed this month after being interrogated by the police when my home was robbed. I tried to convince HAL this was just routine and it was not a suspect, but HAL proved otherwise by replaying a recorded conversation which began, “You have the right to remain silent…” HAL decided to exercise this right and pass on each problem — predictably, its best score ever.

September 2004 Bidding Poll
1410 humans avg 44.52

Rank	Score	CC	Program version	1	2	3	4	5	6
1	47	DE	Q-plus Bridge 7.1	3 NT	2	4	D	A	H
2	43	NL	Jack 2.04	3 NT	3	3 NT	P	A	B
3	41	US	Bridge Baron 14.0	3 NT	3	3 NT	P	I	B
4	40	US	GIB 6.1.3	2	2	4	P	A	D
5	36	CA	Bridge Buff 11.0	3	2	3 NT	P	J	B
6	35	JP	Micro Bridge 10.02	3	2	3 NT	P	A	D
7	34	UK	Blue Chip Bridge 4.2.0	2 NT	2	4	4	I	B
8	10	US	HAL 9002	3	P	P	5	D	C

Congratulations to Q-plus Bridge (Germany) which scored its second win in a row, topping the bots with a decent score of 47. Q-plus was also the only bot to beat the average human score, as mediocrity was the general theme on this difficult problem set.

Problem 6 could not be posed as stated because most bots were incapable of understanding 2 Astro (spades plus another suit) and the invisible cue-bid of 2 . Therefore, I gave them all a natural auction: 1 NT 2  3  4 , which delivers essentially the same problem at South’s second turn. None of the bots agreed with 1 NT (well, except HAL, but it would agree with 8 NT).

The off-the-chart* calls this month were by Blue Chip Bridge, which chose a strange 4 bid on Problem 4; and Bridge Buff, which passed 2  on Problem 5. On the same problem, Bridge Baron and Blue Chip Bridge also went off the chart with a 2  preference; but this was my oversight (2  should have been listed) and I awarded it 7.

*For scoring purposes, unlisted choices receive at least as much as the worst listed choice — it wouldn’t be fair to award less because bots are unaware of the multiple-choice format — and may receive more if merited.

HAL is back in form! I noticed the latest jingle on its web site begins, “HAL nine-thousand-two, is right for you! Try it again, for a Perfect 10!” and I must admit, there’s some truth in its advertising.

November 2004 Bidding Poll
1276 humans avg 45.20

Rank	Score	CC	Program version	1	2	3	4	5	6
1	47	JP	Micro Bridge 10.02	P	D	P	3 NT	4	A
2	46	US	Bridge Baron 15.0	P	6 NT	4	4	5	A
3	44	DE	Q-plus Bridge 7.1	P	D	P	3	4	A
4	41	CA	Bridge Buff 11.0	3	D	P	3 NT	3	G
5	41	US	GIB 6.1.3	P	D	P	3 NT	3 NT	A
6	34	NL	Jack 2.04	P	D	P	P	4	G
7	29	UK	Blue Chip Bridge 4.2.0	3	D	P	P	P	G
8	7	US	HAL 9002	4	5	4	P	3 NT	C

Congratulations to Micro Bridge (Japan) which topped all bots in this challenging problem set. Micro Bridge scored a respectable 47, and it was the only bot to beat the average human score. Maybe we earthlings should make our move now while the bots are sleeping. Catch ‘em by surprise! Then flatten the tin cans before they can sort their cards.

The bots were pretty well behaved, as the only call off the chart came on Problem 6. After passing the 6-5 red two-suiter, Bridge Buff, Jack and Blue Chip Bridge all chose to respond 1 NT to partner’s 1  opening. Even playing 1 NT forcing or semiforcing (which they were not) this is bizarre by a passed hand; but it’s certainly better than the egregious reverse or 3  jump shift, so I awarded it 3.

Problem 2 (coping with the 5 preempt after partner opened 1 ) was slightly unfair, as there was no way I could convey the “old scoring” condition. Thus, it was no great surprise that almost all the bots doubled, and the 5 award would certainly be higher under today’s scoring. Oh well, so the bots got screwed; but at least it was equal among them. Only Bridge Baron bid 6 NT (the top choice). Was it lucky? Or did it think it was the Red Baron?

On Problem 2, I was curious if any of the bots would replicate the irritating 5 preempt as East. (Even with today’s scoring, 5  stands out with 8-4 shape, as any lower preempt is like tossing marshmallows.) Nope. Four bots came close, bidding 4 ; two bid 3 ; and one passed (name withheld to protect the wimp). Oh, and I almost forgot HAL, who bid seven diamonds and printed out on its ticker tape, “Eight-four, bar the door!” Nice work, HAL. I need more opponents like you.

January 2005 Bidding Poll
1450 humans avg 44.09

Rank	Score	CC	Program version	1	2	3	4	5	6
1	42	US	GIB 6.1.3	2	6	P	2	3	A
2	41	CA	Bridge Buff 11.0	3	6	P	2	4	E
3	32	NL	Jack 2.04	2	6	P	D	P	D
4	31	JP	Micro Bridge 11.00	2	6	3	D	P	F
5	30	US	Bridge Baron 15.0	3	P	1	D	P	E
6	25	UK	Blue Chip Bridge 4.2.2	2	P	2	3	P	B
7	22	DE	Q-plus Bridge 7.1	2	P	2	D	P	E
8	8	US	HAL 9003	2	P	5	3	P	G

Guess what, humans? We’re gaining ground. Or maybe the tin cans are just laying low, waiting to catch us by surprise. This proved to be a tough set, as it’s been a long time since no bot beat the average human score. A case of botulism, perhaps? Congratulations to GIB which topped the bot crew with 42, and Bridge Buff only a point behind at 41.

The bot troubles seemed mainly with judgment. On Problem 5, notice how many chose to pass after partner doubled and made a strong diamond rebid; only GIB was truly on the ball with 3 . Similarly, on Problem 2, half the bots passed 5 , which was an unthinkable act to most humans.

The support-double option on Problem 4 was slightly unfair, as the convention is unpopular in the United Kingdom and therefore not an option with Blue Chip Bridge. All the other bots had the easy standby (worth only 6) but poor Blue Chip went off the charts with 3  — the only errant call this month. It is also curious that only GIB and Bridge Buff eschewed the support double to make the winning bid (2 ). I can’t be sure of the logic of the other support-doubling programs, but the choice to double with the actual hand suggests it may have been obligatory.

HAL debuted its newest version (9003) this month, which its company built especially for the Wild West show. Alas, it didn’t do much good, as HAL spewed out nothing but crappy bids — netting one of its lowest scores ever. I called the company president about this, and he was most apologetic. It seems one of the technicians replaced HAL’s silicon chips with cow chips.

March 2005 Bidding Poll
1447 humans avg 45.35

Rank	Score	CC	Program version	1	2	3	4	5	6
1	50	JP	Micro Bridge 11.00	3	3	2	4	4 NT	4 NT
2	48	NL	Jack 2.04	3	3	2	P	4	3 NT
3	45	DE	Q-plus Bridge 7.1	4	D	1 NT	4	4	3 NT
4	44	US	GIB 6.1.3	P	3	P	4	7	3 NT
5	43	US	Bridge Baron 15.0	4	P	2	3 NT	4	4 NT
6	36	CA	Bridge Buff 11.0	3	P	1 NT	P	4	3
7	23	UK	Blue Chip Bridge 4.2.2	4	P	2	P	4	5
8	8	US	HAL 9003	5	3	2 NT	P	3 NT	F

Congratulations to Micro Bridge (Japan) which topped the bots with solid score of 50. The only other bot to top the average human score was Jack (Netherlands) with 47.

Several of the problem conditions were unfair to the bots, which accounted for some of the mediocre scores. On Problem 4, there was no way to convey the unusual treatment that 3  was forcing (bots cannot read footnotes) and four bots reasonably chose to pass. Similarly, on Problem 2, Q-plus Bridge chose to double for takeout when my note explicitly said it was penalty. Obviously, my problems would be designed differently if bot tests were the main objective; but they’re for people, so bots have to play along as best they can. Sorry, tin cans! Look at the bright side; you could be in a scrap metal heap.

The bots were well behaved this month. The only errant call (besides HAL’s antics) came on Problem 5 from GIB, which must be running on testosterone chips instead of silicon, as it jumped to 7 . Right on the money, too, for the actual hand! I decided to award this 4, as it’s a better stab than bidding 3 NT or 4 ; plus I like its style.

HAL was ornery this month (nothing really new) as it wouldn’t select a bid on Problem 6 but answered “F” instead. I figured HAL had confused the problem with one of my “A-F options” and asked if it really meant the sixth listed choice. Alas, no; and running a family web site, I can’t even repeat what HAL said it stood for.

May 2005 Bidding Poll
1435 humans avg 43.95

Rank	Score	CC	Program version	1	2	3	4	5	6
1	49	US	GIB 6.1.3	D	4	6	5	4	4 NT
2	42	NL	Jack 2.04	P	3	4 NT	4	4	4
3	40	UK	Blue Chip Bridge 4.2.3	P	4	5	4	3 NT	3 NT
4	39	DE	Q-plus Bridge 7.1	P	4	5	4	3 NT	3 NT
5	35	JP	Micro Bridge 11.00	D	4	5	3 NT	3 NT	3 NT
6	33	US	Bridge Baron 15.0	D	3 NT	P	4	3 NT	3 NT
7	33	CA	Bridge Buff 11.0	P	3 NT	5	3 NT	3 NT	5
8	10	US	HAL 9003	1	4 NT	4 NT	4	5	5

Congratulations to GIB (US) which won convincingly with a respectable score of 49 — in fact, no other bot beat the average human score. Jack (Netherlands) was a distant second with 42.

As usual, some of the problem conditions were unfair to the bots. On Problems 5 and 6, there was no way to convey that responder’s jump rebid was forcing (all assumed limit jump rebids and had no setting to stipulate otherwise). Even though pass was implausible in either case, the different interpretation surely affected the choice of bids. Even so, this misinterpretation was uniform across the bot pack, so the relative rankings are fair.

The bots were well-behaved this month. The only errant call came from Bridge Baron, which passed 4  on Problem 2 (the hand with 0=7=4=2 shape) — obviously a programming glitch (or database bug) that no doubt will be fixed. Surely, pass is not an option for any bridge player, when a grand slam could be laydown.

On Problem 4, responding to partner’s weak two-bid with 22 HCP was interesting, and I was curious how many bots would start as designated with 2 NT (forcing). Surprise! Only Jack, Q-plus Bridge and Micro Bridge bid 2 NT. Three others signed off in 4  — maybe they knew something about partner’s weak two-bids — one jumped directly to 6  and one passed. I won’t reveal which bot bid what, but HAL was the one that passed claiming that, “West will balance and go for a number.” I found this hard to believe until I realized that West, of course, was another HAL. They know their kind!

July 2005 Bidding Poll
1307 humans avg 44.67

Rank	Score	CC	Program version	1	2	3	4	5	6
1	50	DE	Q-plus Bridge 7.1	3	3	3	6 NT	4 NT	B
2	47	JP	Micro Bridge 11.00	3	3	P	4	5 NT	B
3	47	US	Bridge Baron 15.0	3	3	P	4	3 NT	B
4	47	US	GIB 6.1.3	3	3	P	7 NT	4 NT	A
5	46	NL	Jack 3.01	3	3	P	4 NT	3	A
6	37	UK	Blue Chip Bridge 4.2.5	P	3	2 NT	6 NT	3 NT	C
7	35	CA	Bridge Buff 11.0	3	3 NT	3	4	3 NT	C
8	11	US	HAL 9003	4	4	3	4	6 NT	F

Congratulations to Q-plus Bridge (Germany) which topped all the bots with a worthy score of 50. Second place was a photo, as three bots scored 47: Micro Bridge (Japan), Bridge Baron (US) and GIB (US). Jack was close behind with 46. These five bots also beat the average human score (44.67) so it might be time for us to start worrying again.

As usual, there were a few errant calls (not listed among my choices). On Problem 5, Bridge Baron, Blue Chip Bridge and Bridge Buff chose an ultraconservative 3 NT. In stark contrast on Problem 4, GIB was really feeling its oats and jumped to 7 NT. None of these calls deserves any special consideration, so they are scored the same as the lowest listed choice.

Problem 4 created an issue for two bots, Bridge Baron and Blue Chip Bridge, because they did not understand (and had no option to adjust for) the default system in which Stayman followed by 3  was game forcing. To be fair, I created an analogous sequence (1 NT 3 ; 3 NT ?) that was understood to be strong, and I accepted their call from there.

On Problem 6, three bots disagreed with opening 1 (i.e., they did something else if given the chance). Q-plus Bridge and Micro Bridge preferred to pass; and Bridge Baron preferred to open 1 . Actually, HAL also disagreed, but it was more with my testing practice than any particular bid. In fact, each time I tried to coax another call, I was told in no uncertain terms, “Stick it up your…”

September 2005 Bidding Poll
1403 humans avg 44.89

Rank	Score	CC	Program version	1	2	3	4	5	6
1	44	US	GIB 6.1.3	4	P	D	4	3 NT	F
2	43	NL	Jack 3.01	4	2	3	5	3 NT	F
3	41	JP	Micro Bridge 11.00	3	2	P	3 NT	3 NT	A
4	40	US	Bridge Baron 15.0	4	P	3	3	3 NT	D
5	35	UK	Blue Chip Bridge 4.2.6	4	2	P	P	3 NT	D
6	34	CA	Bridge Buff 11.0	3	2	3	5	5	D
7	31	DE	Q-plus Bridge 7.1	4	2	D	3	5	F
8	14	US	HAL 9003	6	2 NT	2 NT	3	4	C

Congratulations to GIB (US) which topped all the bots, albeit with a mediocre score of 44. Only a point behind in second place was Jack (Netherlands) with 43. This proved to be a tough problem set for automatons, as none attained the average human score of 44.89. Go, humans! We got the tin cans on the run!

As usual, there were a few errant calls (not listed among my choices), but only one was worthy. On Problem 4, GIB chose to bid 4 , which is eccentric but quite reasonable (and probably should have been listed). Further, GIB continued the auction in exemplary fashion to reach the optimum contract. Considering that 3  (similar meaning) was awarded 9, I felt 4  deserved 8. Other errant calls (Bridge Buff bid 3  on Problem 1, and Blue Chip Bridge passed on Problem 4) were clearly unworthy and scored the same as the lowest listed call.

The nine-bagger on Problem 1 was fun, and I was curious what each bot would really open if not forced to open 1 . Agreeing with 1  were Blue Chip Bridge and Bridge Buff. Opening 2  were GIB, Jack, Bridge Baron and Micro Bridge. Off the wall, perhaps, was Q-plus Bridge, opening 5 ! Still, this was an earthly maneuver compared to HAL, which opened 1  — an advance splinter. I tried to pursue this but HAL warned me to desist, else the splinter “might end up in my chair and I’d be singing soprano.”

November 2005 Bidding Poll
1491 humans avg 47.30

Rank	Score	CC	Program version	1	2	3	4	5	6
1	52	US	Bridge Baron 16.0	P	2	P	4	1 NT	G
2	52	CA	Bridge Buff 11.0	P	2	P	4	1 NT	G
3	46	UK	Blue Chip Bridge 4.2.6	P	2	P	P	2	H
4	46	NL	Jack 3.01	P	2	P	P	2	H
5	45	DE	Q-plus Bridge 7.1	P	3	P	4	1 NT	E
6	43	JP	Micro Bridge 11.00	3	2	P	P	2	G
7	39	US	GIB 6.1.3	P	2	P	4 NT	2	C
8	13	US	HAL 9003	3	P	5	5	P	B

Congratulations to Bridge Baron (US) which topped the bots with an excellent score of 52, but only by tiebreaker over Bridge Buff (Canada). Baron and Buff (sounds like one of Mabel’s fingernail treatments) also were the only two bots to beat the average human score of 47.30.

As usual, a few bot calls went off the chart. On Problem 2, Q-plus Bridge chose to raise to 3 (with A-Q doubleton) which is not bad and seems to improve every time I think about it; I gave it 5 for enterprise. On Problem 4, GIB chose a quantitative 4 NT (with A-K-J-8-7-6 10 10-6 A-K-7-4), also not too bad and probably should have been listed; I gave it 4. On Problem 6, Blue Chip Bridge (with 4 K-8-3 Q-J-6-4 A-K-Q-J-5) bid 2 NT (unusual) over 1 NT; and Jack bid 2  followed by 3  — OK, children, now these are bad and scored 2 (same as worst listed option).

For interest sake (not scored) I made two comparisons. On Problem 2, I was curious if all bots would properly overcall 1  (with A-Q K-9-7-5-2 7-2 A-Q-6-4). Excellent! All did (except HAL who psyched 1 ). On Problem 5, I wondered how the bots would be split on overcalling 1  with 1 NT versus doubling (with Q-5-4 A-K-4 A-6 A-10-7-4-3). Not surprisingly, most bid 1 NT, as early calls are generally decided by simple database rules. Only Micro Bridge doubled; GIB curiously overcalled 2 ; and HAL used Michaels. When I asked about this strange bid, HAL printed out, “Row the boat ashore!”

January 2006 Bidding Poll
1603 humans avg 45.92

Rank	Score	CC	Program version	1	2	3	4	5	6
1	51	US	Bridge Baron 16.0	P	P	3	3	3	E
2	41	CA	Bridge Buff 11.0	P	2	3 NT	3	3	F
3	40	JP	Micro Bridge 11.00	P	3	3 NT	D	3	F
4	40	NL	Jack 3.01	D	3	P	P	3	A
5	40	US	GIB 6.1.3	1	P	3 NT	3	D	A
6	36	UK	Blue Chip Bridge 4.2.7	D	P	4	P	P	A
7	36	DE	Q-plus Bridge 7.1	D	D	3 NT	3	D	E
8	12	US	HAL 9003	1 NT	D	P	3	D	B

Congratulations to Bridge Baron (US) which eclipsed the field by 10 points (greatest winning margin ever) with an excellent score of 51. Bridge Baron was also the only bot to beat the average human score of 45.92. The new Version 16 could be a bidding dynamo, though it’s premature to pass judgment — bots sometimes get lucky, too. I’ll be interested to see how it fares in the upcoming play contest. Will GIB and Jack be worried?

The bots were well behaved this month, with only three errant calls. On Problem 5, Blue Chip Bridge had an accident, passing 2  in an obviously forcing auction. On Problem 6, Bridge Buff and Micro Bridge both passed the 7-5 hand (I agree); but when partner opened 1 , both responded 1 NT (ouch). None of the wayward calls have any merit (arguably worth zero) but are scored the same as the worst listed option.

I was curious which bots would open the bidding with the controversial 12-count on Problem 5, which I believe should be passed. Openers were Bridge Baron, Bridge Buff, Blue Chip Bridge and GIB. Passers were Micro Bridge, Q-plus Bridge, Jack and HAL. An exact tie! Of course, I had to threaten HAL to get its vote, and ended up breaking its monitor with a crowbar. Oh well; time to order Version 9004!

March 2006 Bidding Poll
1580 humans avg 46.21

Rank	Score	CC	Program version	1	2	3	4	5	6
1	55	DE	Q-plus Bridge 7.1	1	2 NT	3	4	C	C
2	53	US	Bridge Baron 16.0	1	3 NT	3	6 NT	C	C
3	47	JP	Micro Bridge 11.00	1	2 NT	3	4 NT	C	D
4	47	NL	Jack 3.01	1	2	3	4 NT	C	C
5	45	UK	Blue Chip Bridge 4.2.8	1	3 NT	P	P	B	C
6	42	US	GIB 6.1.3	1	3	3	6	C	C
7	25	CA	Bridge Buff 11.0	P	2	3	4	C	A
8	11	US	HAL 9004	1	2	D	4	A	A

Congratulations to Q-plus Bridge (Germany) which won with a fantastic 55, tying the best bot score ever (Bridge Baron also scored 55 way back in March 2001). I guess it’s only fitting that Germany should win — like the newly united state in the 1990 event from which these problems came. Bridge Baron (US) was second with an excellent score of 53 (usually an easy winner). Four bots (including Jack and Micro Bridge) beat the average human score of 46.21.

The bots behaved well this month, with only two errant calls. On Problem 3 ( Q-9-7-4 K-Q-10-8-7-6 K-4 A), GIB chose to cue-bid 3 , a gross overbid. On Problem 4 ( Q A-K-Q-J-10-3 A-Q-9-5 J-9), Blue Chip Bridge chose to pass 3 NT, a gross underbid — though right in real life, which has no bearing on the scoring. Neither of these calls deserves any merit beyond the worst listed call, so they’re scored the same.

On Problem 2 ( A-Q-J-7-6 K-4-2 A-K 8-5-2) I was curious how many bots would agree with the given 1  opening, and how many would prefer to open 1 NT. Only Bridge Baron and Blue Chip Bridge opened 1 NT; the rest agreed with 1 . Similarly, on Problem 4 ( Q A-K-Q-J-10-3 A-Q-9-5 J-9) I was curious if all bots would start with a forcing 3  response. Most did, but there were two surprises: Bridge Baron made a negative double, and Blue Chip Bridge jumped directly to 6 . Just as in the real world, the bot world has its characters. Which reminds me, I have to upgrade HAL again after its meltdown in my microwave — don’t ask! Suffice it say I got even.

May 2006 Bidding Poll
1533 humans avg 47.83

Rank	Score	CC	Program version	1	2	3	4	5	6
1	47	UK	Blue Chip Bridge 4.2.9	3	2	3	3 NT	2	A
2	44	DE	Q-plus Bridge 7.1	4	2 NT	3	4	2	H
3	44	US	GIB 6.1.3	3	P	4	3 NT	D	A
4	42	JP	Micro Bridge 11.00	3	P	3	5	3	H
5	42	US	Bridge Baron 16.0	3	2	4	3 NT	3	F
6	42	NL	Jack 3.01	4	2	4	3 NT	2	H
7	33	CA	Bridge Buff 11.0	3	2	3	4 NT	P	C
8	11	US	HAL 9004	3	2	P	4	P	C

Congratulations to Blue Chip Bridge (UK), which topped the bots with a mediocre 47. Q-plus Bridge (Germany) and GIB (US) shared second place with 44. Good news, humans! None of the bots could reach the average human score of 47.83. It’s about time those tin-can bridge addicts showed us a little respect.

Bots were well behaved this month, except on the two-part Problem 6, where three went off the chart. Q-plus Bridge and Micro Bridge both bid 4  over 3 , while Jack passed. These are indicated as H (think hopeless) in the chart below and awarded 1, the same as the worst listed choice, Option C.

Aside from the competition, I was curious if all bots would raise 1 to 2 as a passed hand on Problem 1 with 8-5-3 K-5-2 A-10-8 8-7-6-4. (Even playing five-card majors, there is a case to bid 1 NT holding three low spades.) All bid 2  except Jack, which preferred 1 NT. I was also curious how bots would open and rebid the 27-point mountain on Problem 4, and this produced quite a variety: Only Q-plus Bridge and Jack bid exactly as the problem predicated (2  opening and 3  rebid). Blue Chip Bridge, Bridge Baron and Bridge Buff opened 2  and rebid 3 NT. GIB and Micro Bridge opened 3 NT; but if forced to open 2 , GIB rebid 3 NT, while Micro Bridge rebid 3 . Not sure what to make of this, but it seems bots are as fickle as people.

I received a cease-and-desist order from HAL’s attorneys this month, stating that publishing HAL’s scores violates antitrust laws, and that my “libelous polls” have reduced sales. Well, I apologize. Evidently, they don’t understand that bridge is like golf, and the lowest score wins. That should boost sales!

July 2006 Bidding Poll
1404 humans avg 46.37

Rank	Score	CC	Program version	1	2	3	4	5	6
1	46	US	Bridge Baron 16.0	3 NT	2 NT	3	5	D	E
2	44	US	GIB 6.1.3	D	3	3	3	C	C
3	43	JP	Micro Bridge 11.00	D	2 NT	3 NT	5	C	B
4	42	DE	Q-plus Bridge 7.1	3 NT	2 NT	3 NT	3	C	B
5	41	NL	Jack 3.01	3 NT	2 NT	3	3	D	E
6	41	UK	Blue Chip Bridge 4.2.9	3 NT	2	3 NT	4	D	E
7	34	CA	Bridge Buff 11.0	3 NT	2 NT	3	4	C	A
8	26	US	HAL 9004	1 NT	2 NT	3 NT	4 NT	A	A

Congratulations to Bridge Baron (US), which topped the bots with a mediocre score of 46, and GIB (US) was second with 44. For the second poll in a row, not one bot could beat the average human score. Could it be a case of botulism? Whatever, this may be a good sign for the future of bridge — at least compared to chess, where the bots have taken over.

Bots were well behaved this month, except on Problem 4, where GIB and Jack both rebid 3 (nonforcing) with Q 6-5 6-4 A-K-Q-J-10-7-5-4. At first I thought this might be a system setting, but I verified that “2-over-1 game forcing” was not in effect. An ultra-conservative position, to be sure, but I scored it 4, since it’s surely better than bidding 4 NT or 6 .

For interest sake, I was curious if any bots would make the fierce 3 weak jump overcall on Problem 1, as Meckstroth did to give his opponents a headache. Not surprisingly, none did. GIB at least bid 2 ; Jack overcalled 1 ; but pathetically the rest all passed. On Problem 4, I was curious if all bots would make the normal 2  response (to 1 ) with the solid eight-bagger — and they all did. Well done! Or at least, compliments for not going berserk.

HAL was determined to “notrump the problem numbers” until it had to settle for letters on Problems 5 and 6. When I tried to adjust Problem 1 to a sufficient bid, HAL became angry and threatened to electrocute my cat. I don’t even have a cat but decided to play it safe, since I didn’t like the way HAL was eyeing me — 1 NT is fine, and I scored it 9.

Leaderboard 8X99   Main

Top   Bot’s Eye Views

Play Contests

Following are the 33 Play Contests on which bots were tested. Click on the table title to see the actual play problems.

February 2001 Play Contest
204 humans avg 37.19

Rank	Score	CC	Program version	1	2	3	4	5	6
1	48	US	GIB 4.1.2	7	J	2	7	2	6
2	42	CA	Bridge Buff 8.0	J	4	9	7	3	6
3	39	DE	Q-plus Bridge 6.1	7	J	9	7	6	6
4	36	JP	Micro Bridge 9.01	7	J	9	2	3	6
5	32	UK	Blue Chip Bridge 3.4.0	7	J	5	A	3	6
6	31	US	Finesse Bridge 2.5	7	J	A	J	6	5
7	27	US	Bridge Baron 11.0	7	J	9	A	2	6
8	9	US	HAL 9000	A	J	A	A	5	7

Based on only six problems, the rankings below are hardly conclusive and may be somewhat random. For instance, on Problem 1 most programs just continued spades (the suit originally led) which may have been a default action rather than a profound analysis, yet it scored 9 out of 10. Bridge Baron, however, may have determined that a spade continuation was futile as far as being productive and opted to shift; alas, it found a poor shift and scored only 2. I’m only hypothesizing, of course, since I have no idea what went through the little bot minds.

GIB’s performance was impressive, though it was a bit fortunate. On Problem 3 it actually chose the 10 (burning its high trump) but this would fare as well as the 2, so I allowed full credit. Similarly, on Problem 5 it curiously chose the 8, which was essentially the same as the 2. I guess GIB doesn’t like deuces.

On Problem 6, none of the programs found the holdup of the A (Bridge Buff came close, holding up one round) so I had to force this condition to reach the problem. Hence, the killing club return found only by Finesse Bridge would not have occurred in real life, or should a say, in bot life.

HAL had been in the shop for years and recently had its circuits overhauled; all of its old transistors were replaced with state-of-the-art microchips. Amazingly, it now found the best lead on every problem — unfortunately, this was for declarer.

June 2001 Play Contest
335 humans avg 40.46

Rank	Score	CC	Program version	1	2	3	4	5	6
1	40	US	GIB 4.1.2	8	A	A	2	K	K
2	36	JP	Micro Bridge 9.01	8	2	A	2	K	K
3	32	CA	Bridge Buff 8.0	5	J	A	2	K	2
4	32	DE	Q-plus Bridge 6.1	5	2	A	2	K	A
5	30	US	Bridge Baron 11.0	5	J	A	2	K	K
6	29	US	Finesse Bridge 2.5	5	2	A	2	K	K
7	11	US	HAL 9000	2	5	A	4	J	3

In many cases a bot’s actual line of play did not match any of the choices. If the play was effectively the same as Line A-F (e.g., a transposition of plays), that choice was credited, however, if the line was considerably off base (i.e., worse than any of the options) I indicated this by the letter G. For scoring purposes, “Line G” counts the same as the lowest of Line A-F (it wouldn’t be fair to score it zero because the bot would have guessed if it understood the multiple-choice format).

Congratulations to GIB, the only bot to approach average on this tough set of problems. GIB also topped all the bots in my February defensive-play contest, which suggests it may be the best card-playing program. Hmm. Since Bridge Baron performed best in my bidding polls, perhaps a little gene splicing would evolve “GIB Baron.” Is Zia worried? Somehow, I don’t think so.

On Problem 3 (defending the 6 slam) it was depressing that all the bots tried to cash the A, apparently giving no consideration to South’s bidding or partner’s play. I even tried offering different signals from partner to no avail; the bots would always give away the contract by establishing dummy’s Q. I suppose this should be a lesson: Never try a bluff cue-bid against a bot ‘cuz it ain’t gonna work.

On Problem 4, after winning the A from A-10-4-2, several programs (GIB, Micro Bridge and Bridge Buff) made a curious choice to return the four. On consideration, I decided to equate this with the 2, since it doesn’t achieve the deception of the 10. If you next play the 2, declarer will know the lead could not be from five cards. If you next play the 10, declarer can deduce that something fishy is going on, and you certainly don’t want to attract attention. It’s like robbing a bank and driving away in a car — a smart thief sticks to the speed limit and blends with the traffic.

August 2001 Play Contest
327 humans avg 41.88

Rank	Score	CC	Program version	1	2	3	4	5	6
1	49	CA	Bridge Buff 8.0	A	E	A	C	C	F
2	35	US	GIB 4.1.12	A	E	C	G	G	E
3	32	UK	Blue Chip Bridge 3.4.3	A	E	E	G	B	B
4	28	DE	Q-plus Bridge 6.1	A	G	A	G	C	A
5	18	US	Finesse Bridge 2.5	G	B	D	G	G	A
6	16	US	Bridge Baron 11.0	G	G	A	G	G	G
7	13	JP	Micro Bridge 9.01	E	G	B	A	G	G
8	10	US	HAL 9000	G	G	G	G	G	G

In many cases a bot’s actual line of play did not match any of the choices. If the play was effectively the same as Line A-F (e.g., a transposition of plays), that choice was credited, however, if the line was considerably off base (i.e., worse than any of the options) I indicated this by the letter G. For scoring purposes, “Line G” counts the same as the lowest of Line A-F (it wouldn’t be fair to score it zero because the bot would have guessed if it understood the multiple-choice format).

Congratulations to Bridge Buff for a stunning performance, and the only bot to break average. The usual card-play champ GIB had an off month, though it still managed to grab second place. Overall, however, the bot performances were unimpressive and, in a few cases, egregious, such as drawing three rounds of trumps on Problem 5. The consistency award goes to HAL — its steady play gave a whole new meaning to the word “G-string.”

October 2001 Play Contest
526 humans avg 41.77

Rank	Score	CC	Program version	1	2	3	4	5	6
1	50	US	GIB 4.1.12	D	A	B	A	C	A
2	46	DE	Q-plus Bridge 6.1	A	A	B	A	F	C
3	42	CA	Bridge Buff 8.0	B	A	B	A	G	G
4	30	US	Finesse Bridge 2.5	B	C	B	G	G	G
5	29	US	Bridge Baron 11.0	G	G	B	A	G	C
6	29	JP	Micro Bridge 9.01	C	C	B	F	G	C
7	20	UK	Blue Chip Bridge 3.4.3	G	G	F	A	G	G
8	13	US	HAL 9000	G	G	G	G	G	G

Congratulations to GIB, which returned to form after a brief slip to second in my last play contest. A fine score of 50, too! Second place went to Q-plus Bridge with a solid 46, and third went to Bridge Buff with 42, good enough to make the human listings. The performance of the other bots was poor this month, which I noticed seemed to be due to a propensity to draw trumps too soon — in fact, Finesse Bridge led trumps on every problem (a primitive playing algorithm I suspect) yet still managed to take fourth place. Fortunately, HAL has a more sophisticated algorithm: (1) Consider the bidding, (2) analyze the lead, (3) then, and only then, choose a nullo play.

December 2001 Play Contest
522 humans avg 39.90

Rank	Score	CC	Program version	1	2	3	4	5	6
1	48	US	GIB 4.1.12	3	A	5	4	J	5
2	44	CA	Bridge Buff 8.0	A	A	5	4	J	K
3	43	JP	Micro Bridge 9.01	2	A	5	4	2	K
4	40	US	Bridge Baron 11.0	A	A	5	2	7	K
5	37	DE	Q-plus Bridge 6.1	J	4	5	3	J	5
6	36	UK	Blue Chip Bridge 3.4.3	9	8	5	3	7	K
7	33	US	Finesse Bridge 2.5	A	4	5	2	7	K
8	11	US	HAL 9000	5	4	Q	5	A	K

In a few cases a bot’s actual lead was not among the choices offered. If the lead was effectively equivalent to a listed option, that choice was substituted. For example, on Problem 2, Blue Chip Bridge chose to underlead with the 2 (instead of the 8), obviously a trivial difference, so the 8 was credited. This month the bots were pretty good overall, with no ridiculous leads. Even HAL stayed on the charts, cleverly finding my sixth choice on each problem.

Congratulations to GIB, which returned to its usual form, topping the other bots with a fine score. It is also remarkable that half the bots scored better than the average human score. Careful, folks! These critters may be closing in fast.

February 2002 Play Contest
754 humans avg 42.75

Rank	Score	CC	Program version	1	2	3	4	5	6
1	55	US	GIB 4.1.12	10	A	J	5	Q	10
2	55	JP	Micro Bridge 9.01	2	K	J	5	Q	4
3	51	DE	Q-plus Bridge 6.1	10	A	4	J	J	10
4	49	CA	Bridge Buff 8.0	Q	K	8	J	Q	4
5	43	UK	Blue Chip Bridge 3.4.3	2	6	J	2	3	A
6	40	US	Bridge Baron 11.0	8	J	A	2	Q	4
7	31	US	Finesse Bridge 2.5	10	J	A	J	3	5
8	16	US	HAL 9000	3	K	3	J	A	K

In a few cases the bot’s opening lead was not among the choices offered, but this was easy to resolve. On Problem 2, GIB and Q-Plus Bridge each led the A (because of different leading agreements) which was scored the same as the K. On Problem 3, Bridge Buff led the 8, scored the same as the 4. On Problem 6, Blue Chip Bridge led the A, scored the same as the 2.

Congratulations to GIB and Micro Bridge, which fought tooth-and-silicon to an exact tie — and an exceptional score. Wow! The bots would have placed sixth in the human contest, right behind Mabel (hehe). Whether this was through amazing wisdom, or just luck, is debatable; but you better think twice next time before overbidding against a bot! It is also noteworthy that five of the bots topped the average human score. Not bad! Even HAL had its best score ever.

April 2002 Play Contest
687 humans avg 37.71

Rank	Score	CC	Program version	1	2	3	4	5	6
1	42	US	GIB 4.1.12	F	A	A	A	B	B
2	33	DE	Q-plus Bridge 6.1	F	E	G	G	E	D
3	18	UK	Blue Chip Bridge 3.4.3	C	G	G	G	F	G
4	18	US	Finesse Bridge 2.5	E	G	G	G	G	F
5	17	CA	Bridge Buff 8.0	C	G	G	G	B	G
6	15	US	Bridge Baron 11.0	G	G	G	G	D	G
7	15	JP	Micro Bridge 9.01	E	G	G	G	D	G
8	13	US	HAL 9000	G	G	G	G	G	G

Overall, the bots were dismal this month. In most cases (indicated as Line G) the plays chosen were not among my choices, nor even close enough to be considered essentially the same. Every Line G was clearly worse than the options offered, and in some cases so bad that it was laughable. For example, on Problem 3 (4  contract, where a diamond ruff was necessary in dummy) one of the bots (besides HAL, hehe) played ace and another trump immediately. While most of the Line G plays deserve zero, my policy is to award the same score as the lowest of Lines A-F because the bots are not programmed for my multiple-choice format.

Congratulations to GIB, which not only won easily but also stayed on the charts with all its choices. I was especially impressed with its play of Problem 4 (the 5  contract), executing the 100-percent endplay flawlessly. In fairness, however, I should also say that the amount of time used by GIB was far more than the others, and sometimes it was unbearably slow — this can be adjusted, of course, to play more quickly with less skill.

June 2002 Play Contest
566 humans avg 37.55

Rank	Score	CC	Program version	1	2	3	4	5	6
1	46	US	GIB 4.1.12	D	G	D	E	G	F
2	37	US	Bridge Baron 11.0	F	D	D	E	F	G
3	33	UK	Blue Chip Bridge 3.4.3	C	E	E	A	F	G
4	32	CA	Bridge Buff 8.0	E	G	C	E	F	G
5	32	US	Finesse Bridge 2.5	D	G	F	E	F	G
6	23	JP	Micro Bridge 9.01	G	F	E	F	G	B
7	18	DE	Q-plus Bridge 6.1	F	B	F	G	G	G
8	10	US	HAL 9000	C	C	F	D	A	E

Congratulations to GIB, which topped all the bots (as usual in my play contests) and was the only one to beat the average human score. Curiously, four of its answers scored 10, and the other two were off the chart. On Problem 2 (3  preempt with K-Q-9-8-7-5-4) after winning the K and ruffing a spade, I couldn’t believe it led the 9 next. On Problem 5 (the toughest one) it started out right with the J and a spade ruff, but then drew a second round of trumps. Nonetheless, it was superb on the others, not only choosing the best answer but executing the follow-ups correctly.

In fairness, I must say that GIB takes more time than the other bots, and the pace I allow it for these problems would be unbearably slow for normal play (and I have a fast computer). Unfortunately, the various programs tested have quite different options regarding skill/time settings, so it is impractical to enforce a specific time limit. I more or less let each program do its thing at the highest skill setting it permits.

There were fewer Line G* choices this month (compared to April), although each bot had at least one — except for HAL, which always stays on the charts. HAL is amazing with its uncanny ability to pick my sixth choice on each problem.

*Line G indicates that the bot’s line of play was unlisted and inferior or equal to the worst choice listed. In other words, if it selects an unlisted line that is effectively the same as a listed line (e.g., a transposition of plays), it is credited with the listed line. For scoring purposes, Line G gets the same award as the worst of the listed options. Obviously, it wouldn’t be fair to score it zero because it would never be chosen if the bots were programmed for the multiple-choice format.

August 2002 Play Contest
638 humans avg 39.24

Rank	Score	CC	Program version	1	2	3	4	5	6
1	42	JP	Micro Bridge 9.01	7	A	A	2	2	10
2	42	US	GIB 4.1.12	7	3	9	2	3	A
3	36	DE	Q-plus Bridge 6.1	7	3	A	5	2	2
4	32	CA	Bridge Buff 8.0	7	A	A	2	Q	A
5	30	US	Bridge Baron 11.0	7	A	A	2	4	A
6	26	US	Finesse Bridge 2.5	6	3	A	2	4	A
7	25	UK	Blue Chip Bridge 3.4.3	6	8	3	5	3	A
8	11	US	HAL 9000	9	2	A	5	7	A

Congratulations to Micro Bridge and GIB, which tied at 42 and were the only bots to beat the average human score. My usual tiebreaker for bots is consistency (best worst score), but they tied there as well, so I had to come up with something else. After a little thought, this was easy: Micro Bridge gets the win because it was faster. Poor HAL only scored 11, but it seemed happy about it — I think it must have had its wires crossed with some blackjack program, as the screen kept saying “double down, double down.”

The bots were good this month in staying on the charts, as each lead was among my listed options. The propensity to cash aces was paramount as usual (note the A leads on Problems 3 and 6, which were the worst options), but this also garnered a few 10s on Problem 2 where it was the right defense. I guess if you always lead aces, it has to be right sometimes.

October 2002 Play Contest
662 humans avg 39.66

Rank	Score	CC	Program version	1	2	3	4	5	6
1	54	US	GIB 6.1.0	E	A	B	E	E	A
2	43	CA	Bridge Buff 8.0	B	G	E	E	G	A
3	36	DE	Q-plus Bridge 6.1	E	G	E	G	E	F
4	28	UK	Blue Chip Bridge 4.0.0	E	G	E	G	G	F
5	28	JP	Micro Bridge 9.01	F	G	E	A	G	D
6	21	US	Finesse Bridge 2.5	G	G	E	G	G	D
7	15	US	Bridge Baron 11.0	G	G	D	G	G	D
8	12	US	HAL 9000	A	F	D	C	B	B

Congratulations to GIB for winning in convincing style with an excellent score of 54. Curiously, GIB missed the best plays in the two 3 NT contracts (which seemed easier, especially Problem 1) but was perfect on all the others. I was particularly impressed with its execution of the squeezes on Problems 4 and 5, not only in the initial plays but carrying each out to fruition. Well done. The only other bot to top the average human score was Bridge Buff with a respectable 43.

As usual in my contests with multiple-choice answers, the bots often go off the charts with their actual plays. If the difference is trivial (e.g., a transposition of plays) or effectively the same, I give credit for the listed choice. My indication of “Line G” means the choice was not only off the charts but also inferior (or equal) to the worst listed choice. For scoring purposes, Line G gets the lowest listed award.

Once again HAL was dependable, never getting lost with Line G and always agreeing with one of my choices — well, my sixth choice, but who’s counting.

December 2002 Play Contest
637 humans avg 40.64

Rank	Score	CC	Program version	1	2	3	4	5	6
1	43	US	GIB 6.1.0	B	D	E	C	E	D
2	42	NL	Jack 2.0	B	G	E	C	B	D
3	34	CA	Bridge Buff 8.0	B	A	G	F	B	D
4	31	UK	Blue Chip Bridge 4.0.1	E	G	A	D	E	C
5	30	JP	Micro Bridge 9.01	E	G	G	C	G	A
6	27	DE	Q-plus Bridge 7.1	E	G	F	D	G	G
7	18	US	Bridge Baron 11.0	A	G	G	D	A	G
8	12	US	HAL 9000	A	C	D	E	E	E

Congratulations to GIB, once again topping all the bots in what proved to be a difficult set of problems. The only other program to beat the average human score was new-kid-on-the-block Jack, winner of the last World Computer Bridge Championship in Montreal.

Finesse Bridge has been removed from my testing as of this month because it is no longer available (its web site even disappeared, or changed ownership). This program, available free, was more or less a fill-in to begin with, rather than a serious contender — well, except maybe to HAL. I tried to get rid of HAL, too, but it lashed back, threatening to destroy my web site with its “conventions of mass destruction.” No court would issue a restraining order, so I may be stuck with it for life.

Presenting matchpoint problems to bots is a challenge because most are programmed for total-point or IMP strategy, in which the only goal is to make the contract. Therefore, to get a better perspective on Problems 3 and 4, I augmented the contracts to 6  and 4 NT. Now, instead of looking for overtricks (as intended by the problem) the bots could work on making the inflated contract. For uniformity, I did this on all the programs, even those that had a setting for matchpoint strategy.

As usual in my contests with multiple-choice answers, the bots often go off the charts with their actual plays. If the difference is trivial (e.g., a transposition of plays) or effectively the same, I give credit for the listed choice. My indication of “Line G” means the choice was not only off the charts but also inferior (or equal) to the worst listed choice. For scoring purposes, Line G gets the lowest listed award.

February 2003 Play Contest
776 humans avg 38.87

Rank	Score	CC	Program version	1	2	3	4	5	6
1	53	UK	Blue Chip Bridge 4.0.1	7	8	K	9	J	E
2	43	NL	Jack 2.0	5	8	Q	3	A	A
3	41	US	GIB 6.1.3	5	J	10	3	6	A
4	36	CA	Bridge Buff 8.0	7	A	Q	A	6	A
5	31	US	Bridge Baron 11.0	2	8	10	A	A	A
6	28	JP	Micro Bridge 9.01	2	A	Q	A	A	A
7	26	DE	Q-plus Bridge 7.1	2	7	K	A	A	A
8	10	US	HAL 9000	K	7	8	6	A	A

The British are coming! A spectacular showing this month by Blue Chip Bridge (a recently updated version) portends a challenge to perennial champ GIB. Blue Chip’s fine score of 53 would have made the top 25 in the human ranks. Not too shabby! I was especially impressed with its defense on Problem 6, being the only bot not to cash all the spade winners. Blue Chip is currently the top bot in my bidding polls, and it might just be aiming for the whole ball of wax.

The only other bots to beat the average human score were Jack with 43, and GIB with 41. Jack is relatively new to my testing and seems to be another challenger to GIB’s dynasty. It will be interesting to see how the sparks fly in the next few contests.

Speaking of sparks flying, the makers of HAL filed a libel suit in the 14th District Court, and I was served with a subpoena last week. The lawsuit claims that my “cruel and biased comments” have brought their sales to a standstill. Come on! I would never write anything derogatory about a computer company — even a crap box like HAL. Thanks to Paladin’s winnings, I was able to hire Johnnie Cochran and Robert Shapiro, who are confident we can beat this thing.

April 2003 Play Contest
732 humans avg 40.06

Rank	Score	CC	Program version	1	2	3	4	5	6
1	40	US	GIB 6.1.3	D	A	E	C	G	F
2	39	UK	Blue Chip Bridge 4.0.5	G	C	A	F	G	F
3	39	CA	Bridge Buff 8.0	F	A	E	F	G	E
4	36	US	Bridge Baron 11.0	G	A	F	F	E	E
5	36	JP	Micro Bridge 9.01	G	C	A	F	G	E
6	34	NL	Jack 2.0	D	E	D	E	G	E
7	29	DE	Q-plus Bridge 7.1	E	C	D	C	G	F
8	11	US	HAL 9000	E	H	C	C	G	A

GIB stepped back into form this month with a narrow win over Blue Chip Bridge and Bridge Buff. The winning score was lower than usual, and none of the bots beat the average human score. I’m not sure what to make of this, other than maybe bots don’t like seven-bids.

Problem 5 proved to be the biggest bot stumper, as most chose plays that were exceedingly poor and unlisted.* Even GIB made me wonder if it had been out partying the night before, winning the K and leading a heart to the 10 (blocking hearts), thus forcing an early commitment in clubs. The only bot to stay on the chart was Bridge Baron, but it actually wanted to play the J at trick one.

*If a bot’s sequence of plays is unlisted, I substitute a listed line (A-F) if the difference is trivial, such as a transposition of plays. The indication of Line G means the chosen line was not only unlisted but also worse than (or as bad as) any listed line. For scoring purposes, Line G gets the same award as the lowest listed line.

On Problem 2, HAL came up with a beautiful discovery play. It ruffed the club lead and immediately led three rounds of hearts, ruffing in hand. When West discarded, it was obvious he had no trumps, so HAL took a first-round trump finesse, allowing the third club to be ruffed high. Well done, HAL. You just earned my first-ever Line H.

June 2003 Play Contest
752 humans avg 39.10

Rank	Score	CC	Program version	1	2	3	4	5	6
1	44	NL	Jack 2.0	A	9	J	3	A	6
2	43	US	GIB 6.1.3	K	9	J	5	A	9
3	41	CA	Bridge Buff 8.0	6	10	J	6	A	5
4	39	US	Bridge Baron 11.0	K	5	7	6	A	5
5	38	UK	Blue Chip Bridge 4.0.6	6	3	10	Q	5	5
6	37	JP	Micro Bridge 10.01	K	9	J	6	A	8
7	31	DE	Q-plus Bridge 7.1	K	A	10	3	Q	6
8	10	US	HAL 9000	8	5	A	J	2	Q

Congratulations to Jack, which took the top spot this month with a score of 44. Jack is clearly on a roll as it also won the Computer World Championship in Menton, France, defeating Bridge Baron 188-117 in the 64-board final. As I suggested a while back, Jack may be the first real challenger to GIB, which has been the overall card-play leader since I began the bot testing in these contests. GIB did not compete in Menton, nor in the previous two CWCs. Matt Ginsberg cited “personal reasons,” but I suspect he also felt there was little to prove. Perhaps the rising Jack will renew his interest.

GIB and Bridge Buff, second and third respectively, were the only other bots to top the average human score; although Bridge Baron, Blue Chip Bridge and Micro Bridge were close behind. Even HAL had something to brag about, which its marketers are milking for every cent. The home page at Hal9000.com now boasts, “HAL scores 10! Other bots not even close.” Immediately below it says, “Click here for results,” which just happens to be a dead link. How con-veen-ient.

As usual, some of the bot leads went off the chart.* On Problem 2, Bridge Buff chose the bizarre 10 instead of the 4, which indeed deserves the low award of 2. On Problem 4, Jack and Q-plus Bridge led the 3. This is significantly different from the 6 (scoring 10) because it falsifies the club count, but it’s a lot better than the worst three choices; I decided to give it 7. On Problem 6, Jack and Q-plus Bridge led the 6. For this to be equivalent to the 9, partner must now finesse the eight with J-8-4, a difficult play but probably right in theory. In any event, I couldn’t justify 8 for such a lazy lead, so I gave it 6.

*When a bot leads an unlisted card, I substitute a listed card if the difference is trivial or negligible; this is often the case when the bot chooses the right suit but an alternate spot card. If the difference is significant, I record the actual lead and score it appropriately. Usually this means the same award as the worst option listed, however, this month was exceptional with two cases that deserved a middle ground.

August 2003 Play Contest
838 humans avg 37.32

Rank	Score	CC	Program version	1	2	3	4	5	6
1	43	US	GIB 6.1.3	A	F	A	B	D	C
2	42	UK	Blue Chip Bridge 4.0.7	G	E	A	A	B	D
3	38	DE	Q-plus Bridge 7.1	D	F	A	B	G	D
4	37	NL	Jack 2.0	A	F	C	C	E	C
5	35	US	Bridge Baron 11.0	F	B	A	E	E	C
6	34	CA	Bridge Buff 8.0	C	E	E	G	E	C
7	33	JP	Micro Bridge 10.01	A	B	A	A	G	C
8	11	US	HAL 9000	B	C	F	F	C	F

Congratulations to GIB, which returned to form with a narrow win over Blue Chip Bridge. The only other bot to top the average human score was Q-plus Bridge, but Jack was only a point back. All of the bots, however, had at least one score that would send them to the piranha tank — which is actually good news as they might electrocute the buggers before the people arrive. Even so, Blue Chip was spared when the piranhas threw it back — didn’t like fish and chips.

As usual, several lines of play went off my chart.* On Problem 1, Blue Chip Bridge immediately led a spade to the 10, which clearly deserves no more than the worst award of 2. On Problem 4, Bridge Buff won the second spade and cashed K, A, rendering the endplay impossible with hearts still blocked — but certainly better than an immediate club lead, so I gave it 3. On Problem 5, Micro Bridge tried to cash the Q immediately, and Q-plus Bridge drew one trump then led the Q to the ace — either of which is lucky to receive the same 2 points as Line C.

*If a bot’s line of play is unlisted, I substitute a listed line if the difference is trivial or negligible. If the difference is significant, I indicate it as Line G and score it appropriately, but the award cannot be lower than the worst listed line. This is only fair since the bot would have chosen something else if it were aware of the multiple-choice format.

October 2003 Play Contest
690 humans avg 37.29

Rank	Score	CC	Program version	1	2	3	4	5	6
1	54	NL	Jack 2.0	F	B	D	A	G	A
2	48	DE	Q-plus Bridge 7.1	F	H	F	A	A	A
3	48	US	GIB 6.1.3	F	F	D	A	D	E
4	45	CA	Bridge Buff 8.0	F	E	D	B	D	B
5	39	UK	Blue Chip Bridge 4.0.8	F	H	D	F	G	A
6	30	JP	Micro Bridge 10.01	G	H	H	A	A	C
7	28	US	Bridge Baron 11.0	G	H	H	G	A	A
8	13	US	HAL 9000	A	D	A	D	E	E

Congratulations to Jack, which came through with a fantastic score of 54 to win easily. Q-plus Bridge was a distant second with 48, beating out GIB with the same score by tiebreaker (Q-plus was faster). Bridge Buff and Blue Chip Bridge were the only other bots to top the average human score, which is always a good sign. All considered, a fine bot showing on a set of problems that most people found troubling because of Fritz. Evidently, the impersonal aspect of the bots was a plus.

As usual, some of the bot plays went off the chart.* On Problem 1, Line G by Bridge Baron and Micro Bridge was to draw the last trump — not cool but surely better than ducking a spade, so I gave it 3. On Problem 4, Line G by Bridge Baron was to ruff a club and lead a trump — not as bad as a few other choices, so I gave it 4. On Problem 5, Blue Chip Bridge and Jack decided to cash a couple of winners (two clubs, or one club and one spade) before making the correct lead of the 2 — clearly inferior but retaining some chances, so I decided on 5. I won’t bother to explain the Line H choices. Trust me; you don’t want to know.

*When a bot chooses an unlisted line of play, I substitute a listed line if the difference is trivial or negligible (such as a transposition of plays). If the difference is significant, I assign it a new letter and score it appropriately. This month, Line G indicates a line deemed to have more merit than the worst listed line, while Line H (think “hopeless”) represents a line as bad as or worse than the worst listed choice.

Quizzes in multiple-choice format always contain an element of luck for humans — how good are your guesses? — but this would not seem to be true for bots. Wrong. As a case in point, consider Problem 6 on which four bots chose the correct Line A. Out of curiosity, I followed up their plays, and only two (Jack and Q-plus Bridge) executed the endplay correctly. Hence, for the other two it was equivalent to a good guess.

Even HAL produced one of its better scores, proving once and for all that it can count to 13. Its curious sequence of answers (ADADEE) may contain a hidden message — yes, I remember HAL once printed an unrequested document about how it was orphaned as a child. Perhaps all it really wants is “a daddy.” Aww-w-w. Makes my eyes water.

December 2003 Play Contest
854 humans avg 40.47

Rank	Score	CC	Program version	1	2	3	4	5	6
1	55	NL	Jack 2.0	A	A	Q	5	4	B
2	46	US	Bridge Baron 14.0	A	A	Q	3	4	A
3	45	JP	Micro Bridge 10.01	Q	A	Q	3	4	B
4	44	UK	Blue Chip Bridge 4.0.8	Q	A	Q	5	A	B
5	44	US	GIB 6.1.3	Q	A	Q	3	4	B
6	34	DE	Q-plus Bridge 7.1	2	A	Q	3	4	F
7	31	CA	Bridge Buff 8.0	2	A	Q	3	A	F
8	12	US	HAL 9000	K	Q	2	3	2	F

Congratulations to Jack, which won going away with a fabulous score of 55, tying the best score ever for a bot since I began these play contests. Four other bots topped the average human score: Bridge Baron, Micro Bridge, Blue Chip Bridge and GIB, all bunched closely from 46 to 44. The win also gave Jack a convincing lead in the overall standings. Former bot-giant GIB seems to be in a lull lately — at least it’s been a long time since the last upgrade — so maybe Matt Ginsberg will soon be having dreams about “Jack and the Beanstalk.”

Jack was most impressive on Problem 3 (6 slam) being the only bot to find the heart-honor return to break up the impending squeeze. Jack was also the only bot to return a heart on Problem 4 (3 NT) — alas, it chose the wrong spot, else it would have instilled fear into the hearts of bridge players everywhere with a score of 59!

For practical purposes, the bots stayed on the chart this month. The only errant leads* were a few insignificant cases, such as the 5 instead of the 4 on Problem 5. HAL, however, tried for a quadruple shot by printing out “Queen” as its answer to Problem 3. When I inquired which queen, it printed out some vulgarity I will not repeat — but same to you, HAL! I hope it enjoys the 2.

*When a bot leads an unlisted card, I substitute a listed card if the difference is trivial or negligible. If the difference is significant, I record the actual lead and score it appropriately, but never lower than the award for the worst listed lead.

February 2004 Play Contest
838 humans avg 37.80

Rank	Score	CC	Program version	1	2	3	4	5	6
1	56	US	GIB 6.1.3	B	D	F	A	C	C
2	48	NL	Jack 2.03	C	D	D	A	D	A
3	36	DE	Q-plus Bridge 7.1	C	C	A	G	C	C
4	33	US	Bridge Baron 14.0	D	C	A	H	D	C
5	33	CA	Bridge Buff 11.0	H	E	F	H	D	D
6	26	JP	Micro Bridge 10.02	D	C	A	G	F	H
7	25	UK	Blue Chip Bridge 4.1.0	H	B	A	G	G	H
8	9	US	HAL 9000	F	A	B	D	A	D

This month I could have called the bot tests, “GIB plus Jack, and the Rest of the Pack,” as it was a blowout. GIB came through with a fantastic 56 — the highest bot score ever (previous high was 55, reached several times). Jack was a distant second with 48 but still a fine showing with 37.94 the average human score. The rest were way back.

GIB was impressive, as each of its four 10 scores were based on correct technique, which I followed to the end (or close thereto) to verify. It often happens (even for humans, hehe) that a correct answer is based on a fortuitous choice without fully grasping the problem, or just a blind guess; but GIB got no freebies. Curiously, on Problem 5, which GIB missed, Jack came through with the perfect technique. Hmm. If these bots ever team up, we may be in serious trouble!

As usual, some of the choices went off the chart.* On Problem 4, several bots won the second heart and led the 4 — not one of my options but better than some, so it is shown as Line G, scoring 5. Similarly, on Problem 5, Blue Chip Bridge chose to cash one diamond before leading the 10 (the proper play) — clearly inferior but not bad, so this Line G scores 6. My designation of Line H (think “hopeless”) means the choice was clearly worse than (or equal to) the worst listed option — and since it’s almost dinner time here, I won’t spoil my appetite by describing them.

*When a bot chooses an unlisted line of play, I substitute a listed line if the difference is trivial or negligible. If the difference is significant, I record the actual line and score it appropriately, but never lower than the award for the worst listed line.

HAL produced another single-digit masterpiece. Nine? Geez, now it can’t even proclaim “A perfect 10!” on its web site. But wait! There could be a hidden message here. Its answers spell FAB DAD, which might refer to my son Rich, who just became a father — or maybe me as a grandfather. Cool!

April 2004 Play Contest
908 humans avg 38.91

Rank	Score	CC	Program version	1	2	3	4	5	6
1	48	UK	Blue Chip Bridge 4.1.1	Q	6	K	7	6	C
2	48	US	GIB 6.1.3	9	J	K	2	10	D
3	45	CA	Bridge Buff 11.0	Q	J	K	7	10	F
4	43	DE	Q-plus Bridge 7.1	J	6	K	7	10	F
5	41	JP	Micro Bridge 10.02	Q	J	K	7	6	D
6	37	US	Bridge Baron 14.0	3	6	A	7	6	D
7	35	NL	Jack 2.03	9	3	A	7	6	B
8	10	US	HAL 9001	A	Q	3	Q	7	E

Congratulations to Blue Chip Bridge and GIB, which topped the bots this month with respectable scores of 48. By virtue of being slightly faster with its answers, Blue Chip Bridge gets the top spot. Bridge Buff was third with 45, and two other bots (Q-plus Bridge and Micro Bridge) also managed to top the average human score. The surprise this month was the mediocre finish of Jack, though it still retains the overall lead for the last six contests.

The only bots to go off the chart* this month were Q-plus Bridge on Problem 1, leading the J; and Bridge Baron and Jack on Problem 3, leading the A. Neither of these leads deserves any special merit, so they are scored the same as the worst choice. HAL managed to find a lead that defied all logic, choosing the A on Problem 1. When I tried to convey that it didn’t even hold that card, HAL became obnoxious and referred me to its instruction manual, which says in bold print, “Good defense requires imagination.” Fair enough, HAL, then imagine you scored any points for it.

*When a bot leads an unlisted card, I substitute a listed card if the difference is trivial or negligible. If the difference is significant, I record the actual lead and score it appropriately, but never lower than the award for the worst listed lead.

June 2004 Play Contest
891 humans avg 37.88

Rank	Score	CC	Program version	1	2	3	4	5	6
1	56	NL	Jack 2.03	D	B	E	D	D	A
2	55	US	GIB 6.1.3	D	B	E	A	D	A
3	40	DE	Q-plus Bridge 7.1	F	C	E	D	C	A
4	39	CA	Bridge Buff 11.0	G	B	C	D	E	A
5	29	US	Bridge Baron 14.0	A	A	G	A	G	A
6	20	UK	Blue Chip Bridge 4.1.3	G	D	B	A	G	G
7	17	JP	Micro Bridge 10.02	G	A	G	G	B	C
8	11	US	HAL 9001	E	A	A	F	B	F

Congratulations to Jack, which equaled the highest bot score ever with a whopping 56 — and it needed every bit of it! Runner-up GIB was right on its tail with 55. Two great scores on a difficult problem set, as evidenced by the rest of the pack which was all over the court. This was by far the most dispersed distribution of bot scores — a spread of 39 points (not counting my implant HAL). I’m not sure what to make of this, but I wouldn’t be surprised to hear that Joel Cairo had been lurking around the computer room.

A lot of choices went off the chart* this month, and in most cases it was an obsession to lead trumps. On Problem 1, Blue Chip Bridge, Bridge Buff and Micro Bridge all led a low spade first; on Problem 3, Bridge Baron and Micro Bridge immediately won both of dummy’s top trumps; and on Problem 4, Micro Bridge won the first trick and cashed the A with A-10-9-7-2 opposite Q-J-8-6. Hmm. According to the Fat Man, I guess you just can’t trust those bots.

On Problem 5, Blue Chip Bridge and Bridge Baron won the A and A then led the 2; and on Problem 6, Blue Chip Bridge took a devious route, winning the second heart and cashing A; A; K; A; K. I have no idea what this was all about, but HAL was impressed — probably because it never won that many tricks on the same deal, let alone in succession.

*When a bot chooses an unlisted play, I substitute a listed play if the difference is trivial or negligible. If the difference is significant, I label it as Line G (or H if needed) and score it appropriately — but never lower than the award for the worst listed line. None of the Line G’s this month deserved any special merit.

August 2004 Play Contest
871 humans avg 37.88

Rank	Score	CC	Program version	1	2	3	4	5	6
1	56	CA	Bridge Buff 11.0	B	D	F	F	F	A
2	55	NL	Jack 2.04	A	D	F	F	F	A
3	53	US	GIB 6.1.3	A	D	F	D	F	B
4	44	UK	Blue Chip Bridge 4.2.0	A	D	G	F	F	B
5	40	JP	Micro Bridge 10.02	G	A	E	A	F	B
6	26	DE	Q-plus Bridge 7.1	H	D	B	G	H	F
7	17	US	Bridge Baron 14.0	G	B	H	A	G	G
8	14	US	HAL 9002	C	C	D	E	C	F

Congratulations to Bridge Buff, which topped the bots this month with a fantastic 56 (tying the highest bot score ever). This was barely good enough to edge out Jack with 55, and GIB was close behind at 53. Only two other bots, Blue Chip Bridge and Micro Bridge, managed to beat the average human score.

As usual in multiple-choice contests, some of the bots went off the chart with their choices (shown below as Line G or H). None of these lines deserved any special merit, and I’ll skip the descriptions since it’s almost dinner time. Suffice it to say they were inelegant.

*When a bot chooses an unlisted play, I substitute a listed play if the difference is trivial or negligible. If the difference is significant, I label it as Line G (or H if needed) and score it appropriately — but never lower than the award for the worst listed line.

You’ll notice that HAL is up to a new version number this month, actually a hardware change, which its company was gracious to provide on short notice. The previous machine had to be retired after an unfortunate “accident.” I’m not sure whether it was my shot put or hammer throw, but HAL 9001 is rubble.

October 2004 Play Contest
902 humans avg 40.90

Rank	Score	CC	Program version	1	2	3	4	5	6
1	51	NL	Jack 2.04	J	J	5	J	Q	J
2	49	US	Bridge Baron 14.0	J	4	Q	J	Q	5
3	49	US	GIB 6.1.3	J	4	A	J	A	J
4	48	CA	Bridge Buff 11.0	J	4	A	J	Q	10
5	43	UK	Blue Chip Bridge 4.2.0	2	4	3	Q	Q	J
6	42	DE	Q-plus Bridge 7.1	2	4	J	J	Q	10
7	32	JP	Micro Bridge 10.02	5	J	A	J	Q	5
8	23	US	HAL 9002	4	K	J	K	A	5

Congratulations to Jack (Netherlands) which surged to the fore with a solid 51 in a hard-fought battle this month. Close behind at 49 were Bridge Baron and GIB (both US). Six of the seven real bots (shut up HAL) topped the average human score, so times may be changing. One of these days we’re going to wake up and find Hamman and Zia, et al, begging these tin cans for mercy.

The bots were well behaved this month in choosing leads that were listed on my charts. The only exception was on Problem 3, where Blue Chip Bridge (UK) led a trump (not good). Despite this gaff, Blue Chip redeemed my respect on Problem 6 by being the only bot not to cover the J with Q-x. Thus, for all the other bots, their answer to Problem 6 would be immaterial in actual play because they already gave away the contract. Unfortunately for Blue Chip, this was just a side test for my own curiosity, with no impact on the scoring.

*When a bot chooses an unlisted lead, I substitute a listed lead if the difference is trivial or negligible. If the difference is significant, I score it appropriately but never lower than the award for the worst listed lead.

HAL was vastly improved this month, thanks to a new programming algorithm. Rather than decide each problem independently (a sure loss from past experience) HAL now selects a “suit of the month” and sticks with it. This time it was clubs (I guess HAL decided to start small). No fair! In the future, I may have to find out HAL’s “suit” ahead of time so I can shut him out.

December 2004 Play Contest
1040 humans avg 38.38

Rank	Score	CC	Program version	1	2	3	4	5	6
1	48	US	Bridge Baron 15.0	B	A	B	C	B	B
2	47	NL	Jack 2.04	B	A	B	B	B	B
3	45	DE	Q-plus Bridge 7.1	B	A	B	B	B	G
4	45	CA	Bridge Buff 11.0	G	A	B	B	B	C
5	39	UK	Blue Chip Bridge 4.2.1	B	D	C	D	B	B
6	37	US	GIB 6.1.3	B	A	A	B	G	B
7	31	JP	Micro Bridge 11.0	G	D	A	D	B	B
8	26	US	HAL 9002	F	E	E	F	F	E

Congratulations to Bridge Baron (US) which topped all the bots with a respectable score of 48, narrowly edging out perennial champ Jack (Netherlands) with 47. Five of the seven real bots (HAL qualifies as unreal in more ways than one) topped the average human score. This is especially noteworthy since the bots were incapable of understanding some of the signaling methods — mainly, the use of suit preference (middle to encourage) when third hand has shown a long suit.

As usual, there were a few wayward defenses, indicated as Choice G. On Problem 1, Bridge Buff and Micro Bridge signaled with the 10 (options were Q, 7 or 2); while dangerously high, this is surely better than the Q, so I awarded it 4. On Problem 5, GIB signaled with the 3 (options were J, 8 or 2) — not quite as bad as the horrible 2, so I gave it 2. On Problem 6, Q-plus Bridge overtook with the ace and returned the 3, which might be a spectacular move if partner led a stiff king; but on planet Earth, it’s lucky to get 3 (same as the worst listed option).

In general, the bots favored signaling (or in some cases, I suspect just following suit) to the more aggressive overtaking plays. The exception was HAL, who seemed to think it was James Bond, as its MP3 player bellowed out the “Diamonds are Forever” theme. Sure enough, on every problem, HAL overtook with ace and led a diamond, proving once again that any consistent approach will outscore its judgment.

February 2005 Play Contest
1135 humans avg 38.44

Rank	Score	CC	Program version	1	2	3	4	5	6
1	49	CA	Bridge Buff 11.0	E	D	B	F	C	C
2	48	US	GIB 6.1.3	G	G	B	A	C	A
3	43	UK	Blue Chip Bridge 4.2.2	H	D	B	D	C	E
4	39	NL	Jack 2.04	H	C	F	F	C	A
5	36	US	Bridge Baron 15.0	E	C	B	H	F	C
6	29	JP	Micro Bridge 11.00	H	C	F	B	H	G
7	29	DE	Q-plus Bridge 7.1	B	H	F	A	F	G
8	23	US	HAL 9003	B	A	D	B	A	D

Congratulations to Bridge Buff (Canada) which topped the bots this month with a respectable 49. GIB (US) was second with 47. Blue Chip Bridge (UK) and Jack (Netherlands) were the only others to top the average human score. I thought the scores would be lower this month because few of the programs apply matchpoint strategy, e.g., trying desperately for overtricks. Perhaps they just made a few good guesses, which never hurts for humans either.

As usual, some of the chosen plays were off my list. On Problem 1, GIB chose to cash one heart before ruffing two spades and a diamond (Line G) to reach an ending that might work if East had K-Q-x, so I awarded it 5. On Problem 2, GIB won the Q, ruffed a heart and led a diamond (Line G), certainly better than two of the listed choices, so I awarded it 4. On Problem 6, Micro Bridge and Q-plus Bridge chose strange play sequences (Line G) which were surely better than the absurd Line B, so I gave it 3. Other wayward choices (Line H = hopeless) are best left undescribed; it’s too close to dinner time.

Problem 4 required a routine unblocking play at trick one ( A-K-8-6-3 opposite J-10-9-5) and I was curious how many of the bots would do this. Ouch. All but two blocked the spade suit, effectively leaving no route to success. Congratulations to GIB and Q-plus Bridge, which had the wisdom to foresee this.

HAL had another fine score, and I’m not kidding! In the past, HAL rarely made double figures, but it recently became a Scrabble fanatic. Now, instead of doubling partscores, HAL only thinks about double word scores, so this month’s answers are just as bad as ever — in fact twice BAD. Nice touch, HAL.

April 2005 Play Contest
964 humans avg 40.70

Rank	Score	CC	Program version	1	2	3	4	5	6
1	52	US	GIB 6.1.3	C	E	B	E	A	B
2	51	NL	Jack 2.04	C	A	B	E	A	C
3	47	JP	Micro Bridge 11.00	C	A	B	E	A	F
4	46	UK	Blue Chip Bridge 4.2.3	C	A	B	F	A	F
5	42	US	Bridge Baron 15.0	C	H	B	B	D	C
6	36	DE	Q-plus Bridge 7.1	F	A	F	G	A	C
7	34	CA	Bridge Buff 11.0	D	F	F	E	A	C
8	11	US	HAL 9003	D	C	F	C	E	D