[KiwiBridge]

Playing in money tournaments, and I've tried quite a few, I find it very frustrating. In addition to the random deals, there doesn't seem to be much logic displayed by these robots. Some examples....

After two passes, my robot opens 2♥, RH robot doubles, I bid 2♠ with my QT9xx, LH robot jumps to 4♠, my robot doubles and to my great surprise all pass. The robots proceed to play 4♠X on their 3-3 fit, 865 opposite K42 for -800. A bonus for me (this time).

Last night, playing for money, I open 1NT, my robot bids 2♥ transfer and I accept with 2♠, now robot bids 3♠. The mouseover explanation shows 9+ invitational so I bid game. Down goes dummy with a three count and I'm minus 300.

Ever tried to sign off in game in a minor? GIB invariably bids slam, which usually fails.

Ever discouraged a GIB lead having shown suits of your own? GIB woodenly continues the suit, setting up discards and letting contracts make when the obvious switch would have allowed it to be defeated.

Then again, my GIB often doubles cold contracts, thus costing me even more cash.

[bid_em_up]

Yea, I opened 1C, 2D by my bot pard, explained as Soloway Jump Shift 17+ hcp rebiddable diamonds, 3C by me, 4H by bot explained as splinter accepting clubs, 4N by me expecting it to be RKC for clubs (I guess 4S would have been?), and what does the bot do?

Passes.

Making 7.

Oh well, at least 4N was makeable.

[hotShot]

When you hover over you bids it shows what they mean.
You will be less surprised, what the bot does when you read what you are promissing .

[bid_em_up]

hotShot, on May 18 2006, 01:36 PM, said:

When you hover over you bids it shows what they mean.
You will be less surprised, what the bot does when you read what you are promissing .

Maybe they do for you, but I have yet to see it......

[bid_em_up]

Ok, now I am home and can actually attempt this. Lets say I wanted to find out how the bot would interpret 4N....I click on 4, then hover the mouse over NT? Nothing.

Cant click on NT, then the bid is made and its too late. So what are you saying?

Telling me how the bot interpreted the bid after the fact is kinda pointless, except I will have an idea of what the bot may do the next time I make that bid.

[uday]

Make sure you are running version 4.7.5 of BBO ( the current one, from
http://online.bridgebase.com/ )

Then try it in a simpler setting. Example, bid 2C when bot opens 1N or something. Most of the time, GIB has some idea of what your bids mean. Maybe it doesnt for obscure sequences.

If you still dont see the popup explanantion (it is FD-style) try clicking on CONV and make sure the FD options to show explanations are set

[bid_em_up]

uday, on May 18 2006, 06:05 PM, said:

Make sure you are running version 4.7.5 of BBO ( the current one, from
http://online.bridgebase.com/ )

Then try it in a simpler setting. Example, bid 2C when bot opens 1N or something. Most of the time, GIB has some idea of what your bids mean. Maybe it doesnt for obscure sequences.

If you still dont see the popup explanantion (it is FD-style) try clicking on CONV and make sure the FD options to show explanations are set

I am running 4.7.5, but had the box for "Display for bids you might make" unchecked. Thanks.

[zenko]

OK, this has to be a bug, I played MBT and the bidding goes: 2Cl (STR) by LHO GIB, 3 Sp (Pree) by my GIB partner, 4 Sp (with singleton) by RHO GIB, and LHO GIB passed with xxx in spades?? I expect my GIB partner to be confused with my bidding (frankly, most of my human partners are too), but to see GIB-GIB partnership not to understand each another is quite strange, almost human...

Zenko

[qing]

Concerning stories about GIB, I have one.
Playing real $, in one hand vul vs not, I opened 1♣, my lefty human overcalled 1♥, GIB pd jumped to 2♠, presumably strong. I had Axxx of ♠, but a minimum hand so I bid 3♠. Lefty tried 4♦, and GIB cuebid 4♥, which gets doubled enthusiastically by my righty GIB.
I have low doubleton of ♦, and am still not very interested in slam, so I retreated to 4♠. Now lefty persists with 5♦, to which my GIB cuebid 5♥again, and again righty doubled. This time I think, oh well this GIB is just going to carry to slam anyway if I retreat? So I tried the effect of passing -- after all in that 5♥ it said it's forcing to 5♠ -- and lo and behold, so did GIB!
So instead of playing in our 8-4 ♠ fit, we played in this wonderful 3-1 fit, and GIB went down 8 (-2300) when we could make 5♠ (but not 6♠).
That is an expensive lesson with GIB!

[barmar]

I just won an MBT because my opposing GIB did one of these strange things. I opened 1NT in 2nd seat, LHO GIB bid 2NT, which supposedly showed minors. His hand was ♠AKx ♥AKTxx ♦x ♣Txxx, and RHO ended up playing 3♦X down 4 vulnerable instead of setting me 1 vulnerable.

Unfortunately, it seems like CHO GIB tends to make more strange bids that LHO/RHO. This was a rare gift.

[barmar]

One thing I've noticed is that GIB does not seem to be good at recognizing cue bids. I've ended up in more 3-3 contracts at high levels because GIB thought I was showing a real suit after I'd already raised him. When the agreed suit is minors, be careful cue bidding a major because every time you try to go back to the minor he'll bid the major over you.

Luckily it seems like the opposing GIBs don't realize that the wheels have come off, either, so they don't double these stupid contracts too much. But going off 4 or 5 when you could have been making a game or only down 1 is still bad.

[TrialBid]

Scoring: Rubber

---

GIB N held this. I opened 1NT and GIB W Dbl'd. GIB N Redbl'd, a bid oddly described as "long suit 5-11 HCP, 12-" The only thing that seemed clear enough was with 5 points minimum it was penalty oriented. With the hand you see 1NTxx was not a success. I'm only glad I hadn't put up real money to see and rely on this kind of bizarre and misleading explanation.

How quickly can we expect to get basic sequences more reliably described? A system called Full Disclosure is sadly misnamed if that's the kind of thing we must be prepared to cope with.

[fred]

TrialBid: It makes me feel a lot better about spending thousands of dollars every month on the servers that are used to provide you with free MBTs to know that the horrible experience you describe did not cost you even $1.

Everyone knows that GIB sometimes makes horrible (or even unexplainable) bids. We even warn our members about this in various places in our documentation. However, in my opinion, its overall level of play is considerably higher than that of the average BBO member. Even if you play so well that GIB's level is not good enough for you, the nature of MB and MBTs on BBO is that they are fair contests - every player has the same advantage (or disadvantage).

One of the things that makes GIB special is that, unlike most computer bridge programs, it tries to exercise its "judgment". That is, it does not always make the book bid if it "thinks" that another bid will work out better. That is probably what happened on the hand you describe.

The nature of bridge is that not all bridge players (or robots) agree with all of each others' judgment calls. Furthermore, when a player (or robot) uses its judgment, the results are not always good. I am quite sure, however, that if you continue to play against GIB and if you keep an open mind, you will see that GIB judges brilliantly at least as often as it judges poorly. Furthermore, I think you will find that certain aspects of GIB's card play are stronger than those of any human bridge player who has ever lived (seriously).

Matt Ginsberg accomplishment in creating GIB is nothing short of remarkable and he has been most generous with respect to the terms under which he is allowing our members to use his program. You (and others who feel the need to complain publicly when GIB does something you disagree with) are doing both Matt and BBO a disservice.

If you would like to help us to improve GIB then send us an e-mail if you see it do something that you don't like. There is no need to show our dirty laundry to the general public.

Making further improvements to GIB will take a lot of time, effort, and skill. That in turn will cost us money. We are a lot more likely to spend this money if people like you stop complaining and start throwing the occasional $1 towards our MB efforts.

You might even find that you enjoy the experience. Rumor has it that I am a pretty good bridge player and I love playing in MBTs. If GIB is good enough for me, then maybe (if you adopt a different attitude) you will find that it is good enough for you too.

If not then don't play, but posts like yours are not going to improve things. The most likely result of these public complaints is to drive people away from MB and MBTs. That is about the best way to ensure that GIB will never improve.

Fred Gitelman
Bridge Base Inc.
www.bridgebase.com

[TrialBid]

I'm sorry if it seemed I was criticizing GIB--unless you are defending the explanantion vis-a-vis the bidding. If I'm criticizing anything it is the apparent disconnect between the explanantion and the hand.

Specifically: what is it about the explanation that suggests that when GIB redoubles 1NT that I should not pass with most normal hands? If this was truly an anomalous bid and the explanation really suggested that most times the final contract should be 1NTxx, that is one thing. And it's one I can live with.

On the other hand, if that was the explanation I received and my hand had run, then as an opponent I would have called for the director. There's the principle that says if both sides take abnormal actions on the same hand that vary in the same direction, there is a presumption of an undisclosed partnership agreement. Fred, I know it's familiar to you. I'll add for other readers the common example: You state that your 1NT opening is 15-17 and you open what you think is an "attractive" 13. When the hand is over you see that responding hand had 10HCP and passed quickly--making 1NT. Director!

Please rest assured that I'm fully supportive of your MBTs and MB games. The issue is one of disclosure. Surely GIB must be held to the same standards as human players in giving information that describes how it bids!

It's possible that this is monumentally difficult. Possibly, though, the answer lies in the automated nature of the whole system. The most exact answer to "what does Rdbl mean?" would be (hypothetically) that in 119 instances where the auction had started this way, GIB had between 3 and 11 points with an average of 6.4. 1NT was makable on 68 percent of the hands and the average number of tricks available was 7.6. That would be "Full Disclosure."

[fred]

I am not trying to defend GIB's action. My bridge judgment also suggests that GIB made a poor bid. However, GIB "thinks" about bridge very differently than we do and there have been occasions in which GIB has made an action that looked bizarre to me, but on further reflection made a lot of sense (for example, I have learned a lot about opening leads by playing against GIB). I don't think any of us really understands this game well enough to say that we *know* we are right in what is basically a judgment situation.

But even if we are willing to assume that GIB's RDBL on the hand in question really is as insane as it looks to us, that doesn't mean that its inaccurate explanation is a problem. There are several possible legitimate explanations of why GIB chose to RDBL (even though it "knew" it was overbidding).

1) GIB's judgment suggested that the hand in question is really worth 5 points

2) GIB's judgment suggested that by pretending that it had more points than it did, the opponents would be likely to run from 1NT and it would thereby escape a penalty

3) GIB's runout system includes no way to describe a hand like the one it had. It was forced to pick the bid that best described its hand and it judged that RDBL was the best it could do.

OK I admit I personally think 1) is absurd and I don't think much of 2) either (but maybe GIB knows more about how its GIB opps behave than I do). 3) is entirely possible (I don't know how GIB's bidding database describes the other possible runout actions). If 3) is true all it means is that some work is required in this area.

But in none of these cases is GIB breaking the rules of bridge. It is explaining how it understands the RDBL - the fact that it does not have its bid on this particular hand does not come into play.

Essentially GIB psyched, but psychs are a legal part of the game.

You are correct that fielding a psych is not a legal part of the game, but I don't think that this concept really applies in a form of the game in which 2 GIBs are playing against a human and a GIB. If the 2-GIB partnership fields a psych, that might be cause for concern (as one possible explanation was that GIB was programmed to "cheat" - I can assure you this is not the case). If a human fields a GIB psych, then I think all we can conclude is that the human was either lucky or that he has a lot of experience playing with and against GIBs. Nothing nefarious could be going on - after all we are talking about playing with a robot.

Playing in a game with 3 robots is different from "normal bridge" in some respects and I can understand how some people might not enjoy it. However, the robot game also presents all sorts of new challenges that I for one find to be very stimulating.

Hopefully GIB will eventually improve, but in the mean time I think all we can do is know what to expect and try to enjoy the game for what it is. After all, there are plenty of great things about playing with GIB - for example, it never yells at you

Thanks for your support. Sorry if my previous post was an overreaction or if I misunderstood what you were trying to say. Hopefully this post has helped to make you feel better about the whole thing.

Fred Gitelman
Bridge Base Inc.
www.bridgebase.com

[Gerben42]

Quote

Furthermore, I think you will find that certain aspects of GIB's card play are stronger than those of any human bridge player who has ever lived (seriously).

True, GIB does not miscount, he is never tired, never misses a discard, etc. I think in one of his papers Matt counted the number of mistakes for GIB and a World Class player. In the first few tricks GIB makes more mistakes than the WC player but this soon levels out up to the point where a few tricks from the end GIB makes zero mistakes, surely less than anyone alive.

Quote

Matt Ginsberg accomplishment in creating GIB is nothing short of remarkable and he has been most generous with respect to the terms under which he is allowing our members to use his program.

He was the first to construct a good Bridge playing program and invented new strategies so attack the intangible problem of a Bridge hand. There are now newer and stronger programs than GIB, but even if someone manages to create Chtonic, we have to pay debt to Matt Ginsberg who laid the groundwork.

[Poky]

fred, on May 26 2006, 06:48 AM, said:

1) GIB's judgment suggested that the hand in question is really worth 5 points

2) GIB's judgment suggested that by pretending that it had more points than it did, the opponents would be likely to run from 1NT and it would thereby escape a penalty

3) GIB's runout system includes no way to describe a hand like the one it had. It was forced to pick the bid that best described its hand and it judged that RDBL was the best it could do.

What I noticed is, GIB always uses XX as one-suited runout after 1NT-X, no matter of HCP. If so, it seams only the explanation was wrong. "Any 5+carder suit" would be enough.

[barmar]

I'm not sure, but I think GIB's runout system is the same as its NT overcall system: Cappalletti/Hamilton. Redouble is a single-suited hand (like a Capp double), and opener is supposed to bid 2♣, which it will pass or correct, and other bids are two-suiters.

I agree that the description given seems wrong. If GIB thought he had enough for you to make your 1NT, he'd do what most of us would do: pass and let you make 1NTX. No one, not even a silly robot like GIB, makes penalty redoubles at such a low level, because the opponents can easily run.

Fred, I don't think you should take these posts as serious criticisms. I've decided that dealing with GIB's quirks is part of the fun of the money games. But it's also fun to post things here, like "you think that's strange, look what GIB did to me today!" There's no human partner whose feelings could be hurt (unless we all consider Matt Ginsberg our "partner"), so we can all laugh about our misfortunes.

Has anyone figured out what GIB is looking for when you've agreed on a major, perhaps done a little cue bidding, and then it bids 5 of the agreed suit? Obviously it's some kind of slam invitation, but the explanation doesn't give any hint of what it's looking for. It usually seems to be looking for good trumps, but even when I have AKxx or KQJx it never seems to be enough for the slam, because there are also quick losers on the side.

On the other hand, it's often pretty good when it just leaps to slam. The other night I opened 2NT, and with 6=5=2=0 and 10 HCP it transfered and jumped to 7♠, which made (everything fit together nicely, no bad breaks, although my ♣AQJxx was close to useless). It was early in the MBT match and I was sorely tempted to stop playing and sit on my 2200; but I paid for 25 minutes and I wasn't just going to stare at the screen for 20 of them (my score eventually went down a bit, but I still managed to win).

[TrialBid]

I'm inclined to think barmar has it right, and that the action GIB expected me to take was to bid 2♣ over the Redbl--in other words, it "believed" it was making a forcing call.

Perhaps that suggests a simpler idea that could be added to Full Disclosure--as normally expected to be construed, the bid it is making is:

1- Forcing (and to what denomination and level, if implied)
2- Nonforcing-invitational
3- Nonforcing-signoff
and, additionally, for Dbls and Redbls
4- Penalty-oriented
5- Takeout
6- Action or card-showing
7- Rescue

It does make bridge sense to me that GIB, expecting a 2♣ bid from me, would then bid 2♠ as a rescue and might even steal the hand.

Barmar's other general idea, however, that penalty Redbls aren't made at low levels, I think comes out of a world that is playing too much Matchpoints. At total points, when an opponent has Dbl'd unluckily, there is just too much money to be made when there are 28HCP or so and no great fit. To have to settle for +480 instead of 1100 doesn't make sense to me.

[fred]

I am not sure if you are suggesting otherwise, but GIB is not really integrated with Full Disclosure.

GIB is able to supply text-only descriptions of bids. The display mechanism in BBO shows these descriptions in an FD-like window but FD is not involved in this.

Fred Gitelman
Bridge Base Inc.
www.bridgebase.com