[ArtK78]

I rarely comment about GIB's bidding, but I thought that this one was really absurd:

My link

Is there some reason why GIB can't ask for aces (key cards) rather than leap to a grand?

This hand demonstrates the classic use of Blackwood - not as a method for trying for slam, but for checking that cashing aces are not missing in order to avoid bidding a bad slam. Apparently, GIB missed that lesson.

[barmar]

This is coming from simulations (the book bid is 5♥, asking how good your trumps are). It's apparently assuming that since you made a help suit try in spades, that you don't need any help in the other suits. Since it holds ♣ AK, you must have a singleton or void there, and it's placing you with at least ♦ AQxx (and more often AKxx). Since it's assuming there are no losers in the minors, Blackwood won't tell it what it needs to know to decide between 6 and 7. Since 7 makes in most of the simulated hands, it just bids it.

Hands that it dealt you looked like:

S A K 3
H A K Q 9 8
D A K Q 2
C 7
--
S A K T 8
H A K Q 9 2
D A K 7 3
C ---
--
S A K T
H A K 9 7 4
D A K Q 9 8
C ---
--
S A K 3
H A K Q 9 8
D A K Q 2
C 7
--
S A K T 8
H A K Q 9 2
D A K 7 3
C ---

[manudude03]

If the description is followed by the book, then all 5 of those example hands are impossible (12 8421 points when the HSST is 2-11). I think what Art is getting at is that even if blackwood will help occasionally, why not use it?

[ArtK78]

barmar, on 2012-January-09, 22:39, said:

This is coming from simulations (the book bid is 5♥, asking how good your trumps are). It's apparently assuming that since you made a help suit try in spades, that you don't need any help in the other suits. Since it holds ♣ AK, you must have a singleton or void there, and it's placing you with at least ♦ AQxx (and more often AKxx). Since it's assuming there are no losers in the minors, Blackwood won't tell it what it needs to know to decide between 6 and 7. Since 7 makes in most of the simulated hands, it just bids it.

So, what you are saying is the only way I can avoid a grand slam off an ace is if I bid Blackwood myself.

I had a problem rebid over 2♦. I don't have assurance that hearts is an adequate trump suit for slam. So bidding 4NT seems precipitous.

There is nothing I can do to prevent this result other than either bid Blackwood or just bid game?

Assuming that I grant that GIB has decided that a grand will make more often than not on hands that are consistent with my bidding. Does it hurt to check for key cards first before bidding the grand? Obviously, it cannot hurt, and it can avoid this type of disaster.

[barmar]

ArtK78, on 2012-January-10, 08:41, said:

Assuming that I grant that GIB has decided that a grand will make more often than not on hands that are consistent with my bidding. Does it hurt to check for key cards first before bidding the grand? Obviously, it cannot hurt, and it can avoid this type of disaster.

Unfortunately, that's not the way GIB works. There's no way to tell it "If your simulations tell you to bid X, do Y instead." All it can tell is that none of the bidding rules for continuations after Blackwood will get it to the grand, so the only way to get there is to blast.

We could disable simulations in this area, but slam bidding is hard to write precise rules for -- there are lots of judgement calls and assumptions required. Simulations are what GIB uses in place of actual judgement and intuition. We recognize that this is not a perfect system, and GIB doesn't make these crazy leaps as much as it used to because we've blocked some of the simulations already.

If you bid 4♦ instead of 3♠, it doesn't do this. GIB's understanding of a help suit is 2-11 8421 points in the suit -- you actually had 13 in ♠, but only 7 in ♦. I think this is what made it expect your ♦ to be better than they were.

[gwnn]

Art, I think if those are the hands GIB thinks you have, it would be best if you didn't bid 3S at all. Just bid 4H as you have a minimumish 2C opener. Barmar, wouldn't it be better if the description of 3S included "super monster hand with basically 0 losers"?

[ArtK78]

Rather than quibble with the description of what a help-suit slam try should consist of, couldn't GIB be programmed so that if it is going to leap to a slam (whether small or grand) it would bid RKCB first just to check if a key card or two was missing?

In this case, the program tells GIB to leap to 7♥. Can't there be an override which requires GIB to bid RKCB first? If there is a key card missing, then it bids a small slam instead. If no key card is missing, GIB can go ahead and bid the grand.

[barmar]

gwnn, on 2012-January-11, 03:52, said:

Art, I think if those are the hands GIB thinks you have, it would be best if you didn't bid 3S at all. Just bid 4H as you have a minimumish 2C opener. Barmar, wouldn't it be better if the description of 3S included "super monster hand with basically 0 losers"?

GIB doesn't understand multi-meaning bids. If you can make the same bid with two different types of hands, its understanding of your hand is the disjunction of the hand types.

You can see this more commonly in the description of the auction 2M-4M. Responder could have a strong hand with 2+ support (bidding to make) or a weak hand with 4+ support (continuing the preempt). GIB's description of 4M is 2+M, 4+ total points, because that's the least common denominator of the two meanings.

In the 2M-4M case, it's not really a problem because opener isn't expected to bid on, so it doesn't care what the bid means. But multiple meanings for constructive bids would be a problem, because it doesn't help GIB figure out what to do next.

Another place where this comes up is with takeout doubles, which could either show normal takeout shape or a strong hand. We solve the problem there by leaving out the strong hand in the description of the double. If doubler later bids a new suit freely we cancel the original meaning and substitute the new one. But I don't think this approach can be used usefully for this 3♠ bid.

[barmar]

ArtK78, on 2012-January-11, 11:36, said:

Rather than quibble with the description of what a help-suit slam try should consist of, couldn't GIB be programmed so that if it is going to leap to a slam (whether small or grand) it would bid RKCB first just to check if a key card or two was missing?

In this case, the program tells GIB to leap to 7♥. Can't there be an override which requires GIB to bid RKCB first? If there is a key card missing, then it bids a small slam instead. If no key card is missing, GIB can go ahead and bid the grand.

In general, we try to avoid reprogramming GIB unless it's really necessary. We prefer to do what we can with the bidding rules.

I've been studying the code since I came on the job over 2 months ago, and I still find it incredibly confusing. Ginsberg may have had some neat AI ideas, but his coding style is among the worst I've ever seen. So I'm very reluctant to tweak it, since it's so hard to tell what the side effects might be. Most of the changes that have been made to the code have been simple, well-contained changes (e.g. the change to calculation of total points to devalue short suits with face cards). Changing the general bidding or play algorithms are much scarier.

[gwnn]

barmar, on 2012-January-11, 15:29, said:

GIB doesn't understand multi-meaning bids. If you can make the same bid with two different types of hands, its understanding of your hand is the disjunction of the hand types.

You can see this more commonly in the description of the auction 2M-4M. Responder could have a strong hand with 2+ support (bidding to make) or a weak hand with 4+ support (continuing the preempt). GIB's description of 4M is 2+M, 4+ total points, because that's the least common denominator of the two meanings.

In the 2M-4M case, it's not really a problem because opener isn't expected to bid on, so it doesn't care what the bid means. But multiple meanings for constructive bids would be a problem, because it doesn't help GIB figure out what to do next.

Another place where this comes up is with takeout doubles, which could either show normal takeout shape or a strong hand. We solve the problem there by leaving out the strong hand in the description of the double. If doubler later bids a new suit freely we cancel the original meaning and substitute the new one. But I don't think this approach can be used usefully for this 3♠ bid.

That is interesting and I see it is a confusing issue. However, I don't see how this relates to the case at hand. In this case it seems like GIB understands (please don't take these numbers too seriously):

2♣: 19+ hcp, 23+ total points
2♥: 19+ hcp, 23+ total points, 5+hearts (so far, so good)
3♠: 23+ hcp, 28+ total points, 5+ hearts, 2-11 8421 in spades (the italics are not in the description, but somehow GIB assumes it to be true)

Unsuspecting humans would just bid 3♠ to continue to describe their hand, but GIB thinks they promise an ace more than a minimum 2♣ opening. Shouldn't that be in the description? I don't see how multiple meanings for constructive bids come into play. 2♥ had the meaning of (say) 19-30 hcp, 3♠ had a higher minimum, making the range 23-30 or so. 3♠ has a more precise meaning. Which multiple meanings are we talking about?

[barmar]

You said 'wouldn't it be better if the description of 3S included "super monster hand with basically 0 losers"'. When you said "included", I thought you meant that this would be another meaning, in addition to help suit in ♠. If it could be either, that's "multiple meanings". Did you mean that this should replace the current meaning?

I fully admit that GIB's current "help suit slam try" bidding style is not great -- good human players generally make cue bids. However, they also often use Italian-style cue bidding (first or second round controls), which GIB doesn't understand and would be very tricky to add.

GIB's 7♥ bid was not based on South showing a particular strength, but inferences from the fact that he invited slam only needing help in ♠. That's what caused it to deal hands with at most a singleton ♣ and such good ♦ in the simulations.

[calm01]

Barmar said "Since 7 makes in most of the simulated hands, it just bids it."

Using this logic, next time I am driving towards an American STOP sign and I know there is rarely any cross traffic I will not bother to stop and check for traffic.

[fred]

calm01, on 2012-January-17, 10:31, said:

Barmar said "Since 7 makes in most of the simulated hands, it just bids it."

Using this logic, next time I am driving towards an American STOP sign and I know there is rarely any cross traffic I will not bother to stop and check for traffic.

Unlike you apparently, GIB has some appreciation for the notion of risk/reward.

Fred Gitelman
Bridge Base Inc.
www.bridgebase.com

[calm01]

Fred,

You wrote:

"Unlike you apparently, GIB has some appreciation for the notion of risk/reward.

Fred Gitelman
Bridge Base Inc.
www.bridgebase.com"


I am sorry you feel under personal attack - I hope all is well in other areas of your life.

To shoot the messenger is a strategy with a long history - a history with few victories and all of them pyrrhic.

The best place to look for the source of any anger is often in the mirror.

Unconditional love enables one to distinguish between the actor and any perceived sins of the actor.

With unconditional love, Ric.

[fred]

calm01, on 2012-January-17, 14:52, said:

Fred,

You wrote:

"Unlike you apparently, GIB has some appreciation for the notion of risk/reward.

Fred Gitelman
Bridge Base Inc.
www.bridgebase.com"


I am sorry you feel under personal attack - I hope all is well in other areas of your life.

To shoot the messenger is a strategy with a long history - a history with few victories and all of them pyrrhic.

The best place to look for the source of any anger is often in the mirror.

Unconditional love enables one to distinguish between the actor and any perceived sins of the actor.

With unconditional love, Ric.

Constructive criticsm is very welcome. That is why we have this Forum and why we have included facilities in our software itself so that users can let us know when they disagree with GIB's actions. Our strong user community has certainly made it easier for us to do our jobs. We greatly appreciate that.

That being said, I find that many of your posts are not constructive. They are frequently dripping with sarcasm and frequently contain highly questionable (and sometimes downright incorrect) statements regarding bridge and/or bridge software that tend to be expressed as if they were FACTS.

We have excellent people on our staff who are working very hard to improve GIB. Your posts give me (and others) the impression that you sometimes intentionally try to make these people look foolish in public. In general our staff members are too polite to defend themselves, but I have no qualms about doing it for them.

So if you publicly post something stupid and rude, don't be surprised if I am publicly rude (and quite possibly stupid) to you in response.

Fred Gitelman
Bridge Base Inc.
www.bridgebase.com

[cherdano]

Don't make slam tries in this auction when you have a minimum 2♣ opener.