GIB accepts a claim it shouldn't

[xeno123]

From a recent robot tournament, I claimed making 3NT after 9 tricks had been played and GIB accepted the claim. Now it is indeed possible to make 3NT, but if you examine the four-card ending you will see that there are still a lot more ways to go wrong than there are to go right:

 

[hv= sn=xeno123&s=SAQ74HA9D754CAQ75&wn=Robot&w=ST9HQT854DJCKT864&nn=Robot&n=S863HKJ76DA63CJ93&en=Robot&e=SKJ52H32DKQT982C2&d=n&v=n&b=5&a=PP1N(notrump%20opener.%20Could%20have%205M.%20--%202-5%20%21C)P2C(Stayman%20--%2011-%20HCP%3B%2012-%20total%20points)2D(Overcall%20--%2011-%20HCP%3B%20twice%20rebiddable%20%21D%3B%2011-12%20total%20points)2S(2-5%20%21C%3B%202-5%20%21D%3B%202-3%20%21H%3B%204-5%20%21S%3B%2015-17%20HC)P2N(Invite%20to%203NT%2C%20may%20not%20have%204-card%20major%20--%209%20HCP)P3C(4-5%20%21C%3B%202-5%20%21D%3B%202-3%20%21H%3B%204-5%20%21S%3B%2015-17%20HC)P3N(4-%20%21H%3B%204-%20%21S%3B%209%20HCP%3B%2010+%20total%20points%3B%20stop%20in%20%21D)PPP&p=DJD3DQD4S2S4S9S3STS6S5SQHAH4H6H2H9HTHJH3C3C2CQCKHQHKD2D5S8SKSAC6C5CTCJSJ&c=9]400|300|[/hv]

 

In fact even double dummy I'm guessing most non-experts will have trouble seeing the correct line immediately.

 

(I had miscounted and thought both clubs in my hand were high when making the claim - I do indeed have two high clubs, but entry problems make it hard to enjoy both).

 

Am I missing something here?

[Stephen Tu]

Well the East bot has weirdly (for a human, not a bot) thrown the J of spades, so your S7 is good and the hand is rather trivial, you don't have to do the DA stepping stone squeeze thing. The hand is basically double dummy at that point if one has been paying attention to players showing out, so from GIB's perspective it makes no difference whether it throws SJ or not, so it randomly throws it since throwing a diamond is no better, this is one of the inherent design flaws. A bot can't really figure out that something that is trivial from a bot perspective might not be easy for a human player who might not have been paying careful attention or is forgetful, or is not good enough to visualize how the squeeze works. All it knows is: 1. these are the possible board layouts based on bidding & play so far 2. playing this card gives me x tricks on that layout 3. pick the card that maximizes average score on the possible layouts 4. If there is a tie pick something random from the tied cards.

 

If we are going to have the bots accept claims at all (I think they should in general, but prob should be disabled in tournaments?), then I think they are going to have to accept claims like these since we aren't anywhere near a point where a bot can have natural language facility to accept statements of lines of play from humans. We also have to accept that they will make plays like the SJ.

 

The natural way to improve the bot's defense (but probably not possible now given current computing power requirements and/or programming skill/manpower at BBO) is for the bot to make calculations assuming the declarer playing *single dummy* rather than double dummy as it does now, so it plays cards assuming declarer can get something wrong. But this is computationally really expensive and why Ginsberg didn't implement it the decade+ ago when he quit working on it. I don't know if computers are fast enough now, maybe still not. And still for this particular problem it wouldn't help since the hand becomes double dummy and it is not a question of a declarer guessing wrong if they know what they are doing. It's tough because for a computer cashing the thirteenth spade really isn't "easier" than stepping stone squeeze, while for non-advanced human player there is considerable difference, so it doesn't know to keep SJ and make declarer work harder.

 

It might be easier to get it to not play the ten of clubs when you lead up to the J9.

[johnu]

If we are going to have the bots accept claims at all (I think they should in general, but prob should be disabled in tournaments?), then I think they are going to have to accept claims like these since we aren't anywhere near a point where a bot can have natural language facility to accept statements of lines of play from humans. We also have to accept that they will make plays like the SJ.

 

Agree that GIB tournaments should have the claim disabled. Otherwise a player could repeatedly try to claim until they get to a situation where there is a double dummy solution available that they may not have the ability to duplicate.

[1eyedjack]

Another case of cure worse than disease, in my opinion. The ability to claim an obvious contract significantly enhances my personal enjoyment of robot tourneys.

[xeno123]

Well the East bot has weirdly (for a human, not a bot) thrown the J of spades, so your S7 is good

 

Ah - that's what I missed. So then this claim is not so bad after all. I had been somehow assuming it threw a diamond, when indeed you have to go through some contortions via the diamond ace to take three more tricks.

[johnu]

Another case of cure worse than disease, in my opinion. The ability to claim an obvious contract significantly enhances my personal enjoyment of robot tourneys.

 

I would rather play in the fairest game possible which adds to my personal enjoyment of robot tournaments. If I found out a player beat me by gaming the claim function in a paid tournament, I wouldn't play another tournament unless BBO changed the claim function.

 

Clicking through a few tricks when you are just playing winners with nothing to think about doesn't bother me at all.

[1eyedjack]

Then we agree to differ.

[Bbradley62]

What is the criteria for GIB to accept a claim? Is it "tricks can be cashed off the top, no finesse, squeeze, end play or split needed"?

[1eyedjack]

Obviously I am not qualified to answer that - hopefully someone else can. I will just say that I am incredibly impressed with this feature. I have never yet encountered it misjudging its response to a claim. Even where the claim is for fewer than all of the tricks, which is much more complicated to analyse.

 

If there is a problem with the claim function it is that GIB assumes that you human have the same level of concentration as GIB and will retain a complete recollection of every card played in the hand to date, down to specific spots. That is not a major problem in my experience.It is what caused this thread, but play it out and you end up with the same result.

[iandayre]

What is the criteria for GIB to accept a claim? Is it "tricks can be cashed off the top, no finesse, squeeze, end play or split needed"?

 

I can't say with certainty, but from my experience I believe that you are correct.

[iandayre]

Obviously I am not qualified to answer that - hopefully someone else can. I will just say that I am incredibly impressed with this feature. I have never yet encountered it misjudging its response to a claim. Even where the claim is for fewer than all of the tricks, which is much more complicated to analyse.

 

If there is a problem with the claim function it is that GIB assumes that you human have the same level of concentration as GIB and will retain a complete recollection of every card played in the hand to date, down to specific spots. That is not a major problem in my experience.It is what caused this thread, but play it out and you end up with the same result.

 

The other problem is that it allows you to claim the maximum amount of tricks you can be assured of taking - or any amount less. Including Zero even if you hold all high trumps or winners. This is not permitted in live bridge where you cannot concede tricks it is impossible to lose. I would like to see this changed, but I don't want to restrict accurate claims, so I agree with Jack on that point.

[johnu]

What is the criteria for GIB to accept a claim? Is it "tricks can be cashed off the top, no finesse, squeeze, end play or split needed"?

I can't say with certainty, but from my experience I believe that you are correct.

 

This can definitely be gamed by unscrupulous players.

[1eyedjack]

This can definitely be gamed by unscrupulous players.

That possibility is not in question. When I suggested that the cure was (and I repeat in my opinion, and for that matter still so) worse than the disease, I think that acknowledgement of the disease is implicit. Repeating or reinforcing its existence does not add to its severity.

 

These issues are not dependent on the answer to BBradley62's question or of the accuracy of his suggested possible answer. I am not sure how important is that question anyway. For me it is sufficient that the algorithm is accurate. I do not need to understand the guts of it.

 

I don't have to speculate on the enhanced enjoyment that claiming brings. The ability to claim against robots is a relatively recent introduction, before which we had years of experience of having to click through 13 top tricks. I don't have to guess at how much more pleasurable the game is now.

 

Add to all that, whatever your objections it remains a level playing field. There is nothing to prevent any other human mirroring the strategy of the "unscrupulous" player.

[johnu]

Add to all that, whatever your objections it remains a level playing field. There is nothing to prevent any other human mirroring the strategy of the "unscrupulous" player.

 

QFP. You're right. Nothing prevents a player from cheating online, except for honesty and character.

[1eyedjack]

Oh spare us the hyperbole. No doubt you regard it as cheating to take advantage of the knowledge that the human has the best hand.

 

BTW, BBO must have the capability, had it the will, to log the number and frequency of rejected claims made in robot tourneys, including tied to the identity of the claimant.

 

If you think that the possibility translates to regular abuse in fact, which I doubt,

and that declarer has a choice of lines of play in such cases, which will in reality be only a fraction of rejected claims,

and that the rejection of claim diverts him from the line to which he was committed before claiming, which will only be a fraction of the remainder,

and that he actually net gains from that diversion, after deducting some inevitable resulting losses, which will in turn be a fraction of a fraction of a fraction

 

then it might be interesting if a survey by BBO supported that hypothesis.

[johnu]

Oh spare us the hyperbole.

 

QFP

[johnu]

No doubt you regard it as cheating to take advantage of the knowledge that the human has the best hand.

 

The "fact" that the tournament conditions say that the human has at least the most HCP, or that the human declares the hand (in human declares tournaments) are common knowledge to every participant who enters the tournament. Why would taking advantage of that be cheating?

 

Suppose BBO added an undo feature where GIB opponents automatically gave an undo. While intended for misclicks, you use the undo in case you guessed the wrong order of play. Would you consider that to be ethical or just part of the game?

[johnu]

If you think that the possibility translates to regular abuse in fact, which I doubt,

and that declarer has a choice of lines of play in such cases, which will in reality be only a fraction of rejected claims,

and that the rejection of claim diverts him from the line to which he was committed before claiming, which will only be a fraction of the remainder,

and that he actually net gains from that diversion, after deducting some inevitable resulting losses, which will in turn be a fraction of a fraction of a fraction

 

then it might be interesting if a survey by BBO supported that hypothesis.

 

Good idea, I'm sure BBO posters will line up to be counted as cheaters.

[iandayre]

It should be remembered that anyone can kibitz a player in the Robot tournaments. One would expect that anyone who tried this trick would play slowly, others would finish before them. I and others often look in on other players after we finish, and it would be easily spotted if anyone tried claiming excessively.

 

You'd be amazed at how many players never claim.

[johnu]

It should be remembered that anyone can kibitz a player in the Robot tournaments. One would expect that anyone who tried this trick would play slowly, others would finish before them. I and others often look in on other players after we finish, and it would be easily spotted if anyone tried claiming excessively.

 

You'd be amazed at how many players never claim.

 

They could only try the claiming gambit when nobody is watching, probably the first half or 2/3 or so of the tournament. I'm assuming super fast players like Leo aren't going to hang around after they finish in 10-15 minutes so it's mostly a few players who finish closer to the time limit and are waiting around for the results. From what I remember, not that many stick around to kibitz.

[1eyedjack]

The permutations of possible formats of a robot tournament (or "conditions of contest" if you choose to call it that) without disturbing a level playing field are virtually limitless. You could, if you wished, expose all 52 cards at the conclusion of the auction and force everyone to play every hand double dummy. I suspect that it would not be very popular, but it would be feasible. Or you could, as you suggest, permit a player to test whether a finesse works before committing to it. Which would be almost the same result but with the disadvantage of delay.

 

And that I think is the point: A level playing field is a requirement, and is actually hard to disturb, but two formats each with a level playing field are unlikely to be equally desirable. Personally I would not enjoy the format that you propose and much prefer the existing format, just as I prefer the existing format over one that denies claims, although that is a closer preference.

 

BBO is unlikely to hit on a format that pleases everyone. There is a groundswell of players who dislike the "best hand south" format. Where different formats are attractive to different but substantial subsets of the population, they can cater for both by offering a variety of formats. I have not done any survey but I suspect that allowing undos would not have sufficient popularity to make the cut.

 

I suppose that they could run a trial where tournaments that are otherwise identical are run in parallel, one where claims are permitted and one where they are not, and then measure their respective popularity. I am not normally a betting person but I would have a flutter on the result of that.

[1eyedjack]

There is I think a potential improvement to the Undo rule. It would not be perfect, but does not need to be perfect to be an improvement.

 

If the Undo were allowed and submitted before LHO acts (whether for change of mind or, as the original laws intend, to correct a misclick) and auto-granted in that case, then it might make for a more popular format without causing too much offense to the purists.

 

The main problem with that is the speed with which robot acts, so that the interval currently available to submit the request is impractically short.

 

A possible solution to that would be to introduce a delay between human action and LHO robot action. I suspect that would be unpopular to a substantial population, even of those who otherwise would have no objection in principle.

 

A possible solution to THAT would be to provide the user with a configuration option allowing the user to pre-set the length of delay (including zero as an option). But that would have the disadvantage of adding complexity to the user interface for a marginal improvement to a small population, so that will never happen.

 

So I guess that we will be stuck with the no undo rule.

[cloa513]

GIB in free play often rejects a claim even though you can see its solid- DD calculator failing to finish in time? Another reason to doubt the official line of how GIB plays.

[1eyedjack]

GIB in free play often rejects a claim even though you can see its solid- DD calculator failing to finish in time? Another reason to doubt the official line of how GIB plays.

Have you an example? I have never encountered this. Mind you, if there is a flaw in its claim evaluation it is far better that it rejects a claim that it should accept, than the reverse.

 

But it should not (and does not) routinely accept a claim just because its DD calculator shows that it makes.

 

Say you had

[hv=pc=n&s=sakqt987hakdakcak&w=s43h9876dt876ct98&n=s65hqj42dj954c765&e=sj2ht53dq32cqj432&d=s&v=0&b=11&a=7nppp&p=CtC5C4Ca]399|300|Non Spade lead[/hv]

 

If you try to claim 13 on conclusion of trick 1, Gib will correctly reject, despite that it knows it will make. Not that the rejection aids South in any way.

[cloa513]

Have you an example? I have never encountered this. Mind you, if there is a flaw in its claim evaluation it is far better that it rejects a claim that it should accept, than the reverse.

 

But it should not (and does not) routinely accept a claim just because its DD calculator shows that it makes.

 

Say you had

[hv=pc=n&s=sakqt987hakdakcak&w=s43h9876dt876ct98&n=s65hqj42dj954c765&e=sj2ht53dq32cqj432&d=s&v=0&b=11&a=7nppp&p=CtC5C4Ca]399|300|Non Spade lead[/hv]

 

If you try to claim 13 on conclusion of trick 1, Gib will correctly reject, despite that it knows it will make. Not that the rejection aids South in any way.

Why shouldn't it accept a claim that the declarer makes the rest of the tricks? I have had it refuse on trick 4 or 5 and then one more trick and it finally accepts the inevitable. The simplest solution to pushing claim all the time is simply allow it to be tried once and then reject all further tries but GIB has to accept all DD-based claims that succeed.