Robot Defence

[wynsten]

This hand

What is the rationale for East playing ♦J? Double dummy it makes no difference, but Declarer doesn't have xray eyes. Pitching Jack just gives declarers who chose this line of play an extra trick they don't deserve. Ideally defending Robots should play double dummy defence, but when multiple lines are possible, simple bridge (lead top of sequence, otherwise lowest card that works).

[G_R__E_G]

This hand

What is the rationale for East playing ♦J? Double dummy it makes no difference, but Declarer doesn't have xray eyes. Pitching Jack just gives declarers who chose this line of play an extra trick they don't deserve. Ideally defending Robots should play double dummy defence, but when multiple lines are possible, simple bridge (lead top of sequence, otherwise lowest card that works).

 

 

I'd assume that GIB was showing count and in the simulations it ran it decided that it could afford to play the jack.

[wynsten]

Sure it could "afford" to play the Jack - against an opponent who can see through the cards (or can't see at all). But aganst an opponent who sees what you play, and can't see what you don't play (and that's the usual kind) the Jack is a bad play.

 

As for showing count, does its GIB partner recognize count signals?

[G_R__E_G]

Yes, GIB absolutely recognizes count signals. What you need to realize is that GIB doesn't know "good plays" or "bad plays". What it does is run simulations and plays accordingly. For example you might have AKJ109 opposite 8765 and have evidence that the opponents hands are fairly flat. GIB might finesse the queen and you'd say "Any idiot knows that with a 9 card fit playing for the drop is the percentage play" and you'd be right but what GIB's done is run the simulation multiple times and as chance would have it this time around it's calculated that in more of those simulations the finesse worked better.

[wynsten]

"Yes, GIB absolutely recognizes count signals" Thanks - I hadn't realized that.

"What it does is run simulations" Fair enough. Are the opponents in these simulations playing double dummy, or at random, or doing simulations of their own...? If they are playing double dummy or at random, then the simulations will rate unreasonably high plays that give away the locations of cards.

[pigpenz]

how about the lead!!!! :ph34r:

GIB just never likes to lead what it has!

most humans are either leading a ♦ which brings up back to the J♦

most likely getting played at trick one

or people follow the Soloway rule that holding 5-4 lead 4 bagger are leading a ♠

but come on 8♥ lets get real

[Stephen Tu]

The opponents are assumed to play double dummy, so GIB sometimes gives away the placement of the cards by randomly popping honors. At some point the plan was to simulate with the opps playing *single-dummy*, which would in theory solve this kind of problem, but I think the computers weren't fast enough yet and it was never implemented.

[wynsten]

Thanks, Stephen - that makes sense. I guess it depends what you are trying to achieve. As a competitor in a "robot tourney" I don't really care whether the robot defenders signal one another - what I want is to have my good play rewarded and my opponents' bad play punished. The current setup doesn't do that. Better (though not perfect) would be to have the robot defenders play double dummy, and where equal plays are available, play "human-like" (lead top of sequence, otherwise play low; lead longest suit against no-trump, etc).

[Stephen Tu]

I don't think we want the defenders playing true double dummy (i.e. cheating). But maybe the code should be altered so that in a case where it determines multiple cards are equal, instead of playing total random, it play spots instead of honors.

 

This isn't ideal (there are some combos where it is right to pop the honor with some appreciable frequency), but probably better than what it is doing now.

[wynsten]

Stephen: Playing double dummy isn't cheating (I'm not competing against the robots; I'm competing against the other humans). If, as declarer, I find a winning line that can't be beat double dummy, and you don't, then I should score higher than you.

 

If, as a defender, I find a brilliant return (double dummy-like, due to my insightful card reading) then I should not be penalized because my robot partner can't play equally brilliantly. Please give me double dummy robot defenders (as partners or opponents).

 

As for the declarer, if I am dummy, let me play the hand instead of the robot. If I am a defender (with a double dummy robot partner) then the robot declarer should just play good single dummy.

[Stephen Tu]

IMO if you want to play vs. double dummy defenders just buy & play bridge master 2K. Even better than double dummy, they swap cards under the table so you go down if you weren't supposed to play a particular line!

 

Changing the bots to play DD makes for a weird game IMO. Part of declaring is playing in such a way that opps can make mistakes and taking advantage. Opps who never make mistakes and make the killing lead 100% of the time are just unrealistic. They would unjustly train people to be far less aggressive in bidding than they should be. I definitely, DO NOT WANT THIS, for tournaments or MBC hands. If you want the option for this, in say training mode type of room, that would be fine, but it's not real bridge any longer IMO.

 

Lots of people have asked for allowing human to declare all hands by human-robot side, that's reasonable, but they've found it hard to program.

[pigpenz]

IMO if you want to play vs. double dummy defenders just buy & play bridge master 2K. Even better than double dummy, they swap cards under the table so you go down if you weren't supposed to!

 

Changing the bots to play DD makes for a weird game IMO. Part of declaring is playing in such a way that opps can make mistakes and taking advantage. Opps who never make mistakes and make the killing lead 100% of the time are just unrealistic. They would unjustly train people to be far less aggressive in bidding than they should be. I definitely, DO NOT WANT THIS, for tournaments or MBC hands. If you want the option for this, in say training mode type of room, that would be fine, but it's not real bridge any longer IMO.

 

Lots of people have asked for allowing human to declare all hands by human-robot side, that's reasonable, but they've found it hard to program.

and thats the part I hate about the commercial GIB

is having to declare when GIB should be playing

sorry but it has never just felt right when the table gets rotated

must be some sort of mental block :rolleyes:

[wynsten]

No, Stephen I don't "want to play vs. double dummy defenders". Nor do I want the cards swapped beneath the table. I want to compete against humans with an even playing field, eliminating as much as possible, the luck factor. Duplicate bridge does that already, by dealing us the same hands. Currently, however, the robot defenders play like a pair of dummies. On average I can make at least one extra trick against them (especially in notrump) than I could against decent human players. But the problem is that they dole out this generosity on a rather random basis. (Of course they play identically against identical declarer play, but small variations in play can lead to silly play like the ♦j in my example.)

 

You have a point that you can't trick a robot that is playing double dummy, so you do lose that aspect. But you can't trick them now. The defender in my example wasn't tricked into playing his Jack - he just played it at random. However you are right about the opening lead. No need for that to be double dummy since it happens before the human declarer gets in. But the current logic is worse. Look at the lead in this hand. As pigpenz correctly notes, it is outrageously bad. You are unjustly training people to be overly aggressive. I don't claim my suggestion is perfect; just better than what we have now.

[johnu]

Yes, GIB absolutely recognizes count signals.

 

Is that really true? I thought that GIB played random cards when it thinks it doesn't matter, so half the time you might get the correct count and half the time you would get the wrong count.

[dwar0123]

Is that really true? I thought that GIB played random cards when it thinks it doesn't matter, so half the time you might get the correct count and half the time you would get the wrong count.

I am very doubtful that gib gives count.

[cloa513]

I am very doubtful that gib gives count.

GIB can't signal nor understand signals- BBO Yellows and admin have said so. GIB plays like its opponents can see through the back of cards.

[pigpenz]

GIB can't signal nor understand signals- BBO Yellows and admin have said so. GIB plays like its opponents can see through the back of cards.

yes I thought in earlier posts on the forum it was stated GIB doesnt play count or suit preference or anything, I dont even know if it would play a lightner double if made?

[barmar]

GIB does signal, but it doesn't generally understand partner's signals. It mostly signals count, except that it will give attitude on opening lead if partner leads a high honor. It never gives suit preference, and obviously won't interpret partner's suit preference signals.

[fuburules3]

GIB does signal, but it doesn't generally understand partner's signals. It mostly signals count, except that it will give attitude on opening lead if partner leads a high honor. It never gives suit preference, and obviously won't interpret partner's suit preference signals.

 

I always just thought partner was being ornery when I give him a ruff with the 2 and he returns a spade :rolleyes: