Bidding Deviation in ACBL Robot Duplicate (BUG?)

[jdolson]

Today while playing in an ACBL Robot Matchpoint Duplicate (Tournament #4292, GIB is W, N, and E), I had this auction - South(me) Dealer:

 

1♥ P 2♥ 2♠

3♥ P P 3♠

P 4♠ All Pass

 

At seven other tables, the human player sitting South also bid 3♥ over 2♠ but the auction continued differently??!?:

 

1♥ P 2♥ 2♠

3♥ P P Dbl

P 5♦ P P

 

I have never seen this happen in an ACBL Robot Matchpoint duplicate; usually if the auction is the same up to GIB, GIB will ALWAYS make the same bid. In fact, I thought the ACBL Robot game was set up this way to make sure that the GIBs make the same bid on the same auction.

 

Has anyone else seen this happen...?

 

Regards,

 

Jim Olson

 

To check the Bridge Movies, go to 4-18-2010 Tournament #4292, Board 3 and compare No. 1-7 with No. 26, e.g. BOOKER1 with jdolson (Traveler 4292-1271597401-86585611).

 

Note that when I practice with a human partner against two GIBs in the Main Bridge Club, it is routine to see GIB bid differently on the same auction when reviewing the Bridge Movie - but NOT in the ACBL Matchpoint Robot duplicate.

[JoAnneM]

Yes, I have had it happen, that is why I once made a post asking if the GIB's "learned" how to bid a hand during a tournament, and if that was why some players waited and played the hands late. The answer was "no".

[barmar]

Note that when I practice with a human partner against two GIBs in the Main Bridge Club, it is routine to see GIB bid differently on the same auction when reviewing the Bridge Movie - but NOT in the ACBL Matchpoint Robot duplicate.

It's the same GIB program, so why would you expect it to operate differently in duplicate than in the MBC?

 

Actually, that's not quite right. If you use the Windows client, the version of GIB in the MBC is the version you downloaded, which doesn't contain all the fixes that are in the online version. But when you play robot tourneys, everyone uses the online GIB.

 

The reason why GIB can behave different in identical auctions is because it often uses simulations in addition to the bidding database. Simulations use random hands, and different runs may produce hands that skew differently, resulting in different decisions.

[TylerE]

I wonder...if the GIBs are running server-side would it not be possible to set the random number generator to the same seed in all gibs? If the generator was reseeded to a known value for each bid in the auction, and for each card in the play, this would effectively make GIB derterminate, since each gib would generate the same set of random hands. Wouldn't *improve* things, but would make it hurt a bit less if you knew all the other gibs would be making the exact same mistake.

[barmar]

If there's a two-way finesse, and it's really a guess which way to go, you'd expect about half the field to make the right guess and half to guess wrong. Should it be different when the field is full of robots?

[TylerE]

In a live game....no.

 

In a game where the stated goal is to determine the relative skills of *1* player at each table... sure, why not?

[Bbradley62]

Maybe you hesitated more than the other Souths before bidding 3♥. :)