The Pattern of Friendly GIB Defense

[iandayre]

The general consensus here has been that GIB chooses randomly between equal cards when it perceives that declarer can always make the hand at double dummy. I do not agree with that theory. Here is an example that confirms many others I have seen.

 

Dummy after Trick 10: H x, D AQ Declarer's hand HJ, D xx. East GIB to pitch from H QT, D Kx.

 

I have never seen GIB bare its King in this type of position. Most of us would pitch the HT, allowing Declarer to endplay us, but leaving him the option of finessing. But GIB pitches the HQ, leaving declarer no losing option. It is as if it has an emotional aversion to being endplayed.

 

My point is not that this happened once, it is that this is GIB's standard modus operandi. Am I correct?

[1eyedjack]

I can't comment on the particular hand, but I share your suspicion that it is not random. I believe for example that given a choice of irrelevant small cards it avoids playing the lowest card with a frequency inconsistent with random selection. Against rabbits who can't count that is probably a winning action. Against good players who are wise to the tendency it should be a long term loser.

[Siegmund]

I have noticed that I usually know a couple tricks ahead of time if a squeeze is going to work, because the GIB makes a "gift" discard that was in fact going to be forced a few rounds later.

I could be convinced that it really is random, and it just catches our eye because a human (at regional or lower level) would always play the card that gave an imperfect declarer a chance to fail to properly execute the squeeze or endplay, so half the time GIB's play looks normal and the other half it looks like a silly mistake.

 

It annoys me, because it means I am getting a 50% board in a robot tournament that would have been a 70% in a weak live game -- but usually if GIB makes a fatal early discard it's because he really was going to be doomed unless I did something really stupid.

 

From a programming standpoint, I don't see a good way for GIB to evaluate what kind of mistakes it thinks a human declarer would make in situations that even a 'dumb' computer would never get wrong. It certainly doesn't seem like a priority, compared to the other improvements to GIB's bidding and play that could be made.