[Cascade]

This was from a teaching table but I do have the lin file generated if needed.

On this auction (1♣) 1♠ (1NT) pass; (pass) 2♥ (pass) ?

With

♠54
♥A842
♦T74
♣K542

GIB gave a preference to 2♠.

[Cascade]

I was practicing these auctions 1m (1♠) 1NT ...

Here is another one:

1♦ 1♠ 1NT P P 2♥ P ?

Preference with

♠ 54
♥ 10982
♦ K108
♣ Q84

[Cascade]

A third time - I won't report any more - with:

♠ J10
♥ 9872
♦ J743
♣ AK2

1♦ 1♠ 1NT P P 2♥ P 2♠!?

[johnu]

This is a generic problem (in competitive auctions?) where GIB makes a hopeless preference to the wrong suit. If it could have been easily fixed I'm sure that it would have been fixed years ago. I'm guessing the problem is that there is no meta-rule that controls these awful preferences, and that each individual auction needs to fixed. I'm also guessing that fixes cause other cascading bugs that would need to be patched so it's a never ending cycle.

[smerriman]

It even bids spades with 5 hearts.

Edit - actually, it does with 6 hearts too. Seems it requires 7 hearts before deciding hearts are better.

[Cascade]

And just to compound things on related auctions like 1♦ 1♠ 1NT Pass 2♦ Dbl Pass ?

GIB bids 2♥ on three card suits with a doubleton spade.

[cvcherry]

Yes, sometimes GIB bid suit with not enough length which is a disaster already, leave me no reasonable options. Here's an example:



Why not just PASS here???

[thepossum]

It often chooses the wrong bid. It is based on a completely fundamental flaw in the philosophy and design about how to simulate human bridge play with a computer program. Until this fundamental flaw is fixed robots will always stay in the ordinary category compared to humans. There are many difficult decisions trying to bid to a system. Humans have to be very flexible to deal with them. Even humans struggle with some situations. To expect a computer program to cope with those complex situations where its hand, the bidding, the system, and the potential candidate bids are all in conflict leads to an "incorrect" bid, not to mention the fact that the use of small numbers of double dummy simulations to fill information gaps, and no understanding of human bidding, and no goals (ie knowledge, intention etc). Its amazing it bids so well most of the time. However there clearly is a problem with the number of times it chooses the wrong suit in a preference situation. Maybe sometimes its due to a small simulation sample. I must say the benefit is that my play of hands such as Moysian (and other low trump fit) hands has improved massively over the last year since I started playing GiB