100% wrong defense by GIB

[gwnn]

Check out trick 10:

 

[hv=http://www.bridgebase.com/tools/handviewer.html?sn=gwnn&s=SAT4HK964DAKC9872&wn=Roboter&w=S983HAJT75DJ4CKT5&nn=Roboter&n=SKQJ65HQ82DQ852CA&en=Roboter&e=S72H3DT9763CQJ643&d=s&v=b&b=7&a=1N(notrump%20opener.%20Could%20have%205M.%20--%202-5%20%21C)P2H!(Jacoby%20transfer%20--%205+%20%21S)P2S(Transfer%20completed%20to%20S%20--%202-5%20%21C%3B%202-5%20%21)P3D(New%20suit%20--%204+%20%21D%3B%205+%20%21S%3B%2010+%20total%20points)P3S(Support%203rd%20S.%20No%204th%20D%20--%202-5%20%21C%3B%202-3%20%21)P4C(Cue%20bid%20--%204+%20%21D%3B%205+%20%21S%3B%20%21CA%3B%2013+%20total%20points)P4S(2-5%20%21C%3B%202-3%20%21D%3B%202-5%20%21H%3B%203%20%21S%3B%2017%20HCP%3B%2018)P4N(Blackwood%20%5BS%5D%20--%204+%20%21D%3B%205+%20%21S%3B%20%21CA%3B%2015+%20total%20points)P5H(Two%20or%20five%20key%20cards%3B%20no%20queen%20--%202-5%20%21)D(6+%20HCP%3B%20rebiddable%20%21H%3B%20%21HKQ%3B%2020-%20total%20points)6S(4+%20%21D%3B%205+%20%21S%3B%20%21CA%3B%2015+%20total%20points)PPP&p=S3S5S2S4CAC3C2C5D2D3DAD4DKDJD5D7C7CTS6C6D8D9SAH5STS8SKS7SQC4C8S9DQDTC9H7H2H3HKHJH4HTHQD6H8CJH6HACKSJCQH9]400|300[/hv]

 

I opened 1NT on 14, true, but GIB doesn't know I haven't stupidly shut myself out of some club points (QJ or anything). All it needs to do is play the ace of hearts and return the jack of hearts. Then it is irrelevant what I have in clubs or indeed anything else in my hand.

 

Can anyone still go for the "limited simulations" excuse here? I think it is there can be no distribution where ducking the heart works and there can be no distribution where winning the heart and returning one does not work. So even if you just simulate a single hand, you would get this right as west. Am I missing something?

 

edit: of course also taking the Ace and returning the ♣K is a 100% single/double dummy play.

Instant tournament, IMP's

Edited April 23, 2017 by gwnn

[gwnn]

I am honestly wondering if anyone has ever seen this before, GIB choosing a literally 0% play over a literally 100% play. Just to put a fine point on it:

 

1. GIB knows where all the spades are (it had 3, declarer showed 3, partner showed 2, dummy showed 5).

2. GIB knows where all the diamonds are (dummy had 4, it had 2, declarer showed 2)

3. GIB does not know where all the clubs are, but only 4 are outstanding (I discarded two already).

 

So, what are my possible hands?

 

The "worst case scenario" is that I opened with:

a.

AT4

Kx

AK

QJ9872

Or the middling case:

b.

AT4

Kxx

AK

H9872 (H=Q/J)

 

On the other end of the spectrum is my actual hand:

c.

AT4

Kxxx

AK

9872

b. is a bona fide 1NT opening, a is debateable (not sure what GIB thinks?) and c is 14 so maybe impossible.

 

However, in all of a-c, ducking the ace of hearts is known to let the slam through (the human has the option of finessing on the way back) and winning and waiting for the heart trick is known to beat it.

 

So far, I was ignoring the 1NT opening and carding and talking about cold maths here (I cannot have less than 2 hearts, there aren't enough clubs outside).

 

So, what is going on? I assume the key to understanding this is trick 2, when EastGIB followed with the ♣3 and trick 5 when he followed with the ♣6. Maybe this means that EastGIB can only have 3 or 5 clubs? And because I "can't" open 1NT on 6 clubs, it decided that I must have hand 3b, which I also "can't have" so it just gave up simulating and started ducking everything??

 

If this is the case, there should be a backup plan for these "gave up simulating" situations, for example, GIB falling back on pure mathematical inferences (all suits still add up to 10 points and 13 cards, no matter how abusive humans are) and ignoring everything else.

 

edit: sorry, re-formatting after posting.

Edited April 24, 2017 by gwnn

[gwnn]

Sorry for the bump, is nobody else bothered by this? GIB playing a demonstrably 100% wrong card instead a demonstrably 100% right card? I've seen many unnecessary finesses taken by GIB that can always be dismissed as an unlucky set of simulations, but none that are just known to be the wrong card.

[Stephen Tu]

I think we're all bothered by this, it's just that none of us non-BBO personnel are really in a position to investigate or fix this sort of bug, and what else is there to say really? Does every "GIB shouldn't make completely idiotic plays like this" post require people to make "+1" posts?

 

Assume we agree with you, assume that someone at BBO will eventually read the post and look into why it happened.

[gwnn]

OK sorry for insisting, I wasn't looking for positive reinforcement, but I thought every time an "idiotic play" is brought up, it's been just dismissed as a sampling error, and it seemed to me that this particular case is impossible to dismiss as such. I was not using a hyperbole for the thread title, but it is actually the case that any mathematically correct distribution simulated would tell GIB W to take the ♥A (nothing even breaks even). I have not seen any other threads on such an obvious bug, but I guess I missed some.

 

If GIB is not always playing based on distributions simulated (however few those may be) and sometimes just starts playing random cards and/or starts ducking everything, I think I would like to know that this is the case.

[Stephen Tu]

This is a weird one, I tested it out on my old PC GIB, it makes same error! Also in the command line mode it was giving weird results at the end for what it thought the expected results of playing HJ vs. HA were, like -650/-680, like it forgot it was defending a slam. Definitely worth the programmers looking into the deal in depth.

[uday]

Noted. Don't know yet why it did what it did, but on our list :)

[gwnn]

Thanks to both of you. Stephen: Weird indeed!

[steve2005]

There have been several examples in this forum recently where GIB has ducked the setting trick even.This sort of defensive error is reasonably common.

[gwnn]

There have been several examples in this forum recently where GIB has ducked the setting trick even.This sort of defensive error is reasonably common.

Any specific one you have in mind? If GIB is ducking a setting trick, it might be that it is "trying to get" -2 or that it "thinks" that the contract is going down no matter what, or in suit contracts it is simulating only hands in which declarer is void in that suit, however unlikely that would be. Now you may argue that is absurd (and I will 99% of the time agree with you) but I have never seen a hand where playing card A is mathematically certain to let the contract through and card B is mathematically certain to beat it and GIB playing card A.

[steve2005]

Any specific one you have in mind? If GIB is ducking a setting trick, it might be that it is "trying to get" -2 or that it "thinks" that the contract is going down no matter what, or in suit contracts it is simulating only hands in which declarer is void in that suit, however unlikely that would be. Now you may argue that is absurd (and I will 99% of the time agree with you) but I have never seen a hand where playing card A is mathematically certain to let the contract through and card B is mathematically certain to beat it and GIB playing card A.

 

http://www.bridgebas...ng-after-the-q/

this one is just bizarre

http://www.bridgebase.com/forums/topic/76514-gib-failing-to-cash-slam-setting-ace-in-4th-seat-and-more/

ducking setting trick and should have cashed it earlier anyways

 

[gwnn]

Sorry for being technical but these aren't as strong as my example, where literally any distribution where everyone has 13 cards is beaten, i.e., a literally 100% wrong defense was chosen over a 100% correct defense. I don't wanna be all "OMG my example is so great" but this aspect (that no conceivable simulation would ever tell GIB to duck the ace, and that this is provable) is the only reason why I even created this thread. I knew that GIB makes many, many bad defenses every day, but I did not know that it makes literally 0% plays as well.

http://www.bridgebas...ng-after-the-q/

this one is just bizarre

I agree it's bizarre, but it's not as bad as mine as technically declarer could be void. I'm not saying that this is how GIB should work and I know the auction makes this impossible, but mathematically declarer could be.

http://www.bridgebase.com/forums/topic/76514-gib-failing-to-cash-slam-setting-ace-in-4th-seat-and-more/

ducking setting trick and should have cashed it earlier anyways

Again, declarer could be void. Not 100% wrong. Yes it's wrong and I hope GIB can do better in the future, but it's not as purely wrong as my post.

[smerriman]

Noted. Don't know yet why it did what it did, but on our list :)

So, has been 4 months now and a new version of GIB released. What did you discover when investigating this bug, and was it fixed?

[pigpenz]

GWN

i had lots of hands where GIB makes idiot plays when he has the contract beat in his own hand.

 

EG.

 

 

dummy has Jxxx♥ GIB has AKtx♥

you lead small GIB wins T♥gets in later with A♠then blocks ♥ or crashes your Qxx♥by playing AK♥.

 

when those happen I just shake my head, if there was a simulaton for beating 2 dont care beating 1 is better then letting them make 3NT

[smerriman]

As gwnn pointed out above, this thread isn't about idiotic plays, which are very common - it's about impossible plays, which aren't. So your example is completely unrelated.

 

The lack of response from BBO is rather sad :(

[gwnn]

Yes, the 100% in the title and the text in this case is not hyperbole, but demonstrably correct. Any simulated deal where each player has 13 cards dealt to them and no one revoked would lead W to take his ace of hearts and that taking it would defeat the contract perforce and would tell W that ducking it would let the contract through perforce. This is based only on players following suit, not on any higher inferences. So this is a clear example that proves that GIB does not always base its decisions on simulations.