GIB is not better than the average BBO player

[Yu18772]

I am not an expert, but I am a fairly advanced player with a few top results at serious tournaments. I would gladly pay a few $ to play with a decent partner and against good defense, instead of blindly looking for one at times.

Also, I know GIB is not supposed to be perfect, but... it is also supposed to be better than the average player (according to bbo).

For example, today I tried it for free (not the first time, a couple of times I even deposited a few dollars to try it out).

Just from today - a bidding example, my partners holding (all NV, I am playing south, all other are GIBs):

 

QTx

xx

x

AKQTxxx

 

The bidding:

W N E S (me)

P 1♣ X 1♠

P 2♣ 2♦ P

P P!!! ( I dont know any player who would not bid 2s.....)

 

But the bidding is even not a major issue (one can get used to GIB settings after a while).

Throughout the several times I tried to use GIB , its defense was way worse than any common sensed beginners - GIB absolutely and thoroughly ignores human partner bids or defensive signals (I use standard, as written in its cc).

I can bid a suit and the robot decides to lead Qx from another suit against trump contract, or not lead Kx of my bid suit against NT contract etc...

I can discourage and it will not switch (even if there is another suit I bid), on every defensive hand it seemed that the robot is doing everything in its power to create tricks for the declarer .

We all make mistakes from time to time, but this is not a mistake and its not a fluke, its consistent. This is very frustrating and if GIB was human it would show lack of respect for the human partner.

In a way it shows lack of respect of GIB designers to the human partner, whats worse the repeating bbo statement that GIB is better than average starts to sound as bad promotion.

 

If GIB is indeed better than an average BBO player (I seriously doubt that), it is my impression that this comparison reflects quite poorly on the average BBO player rather than says anything positive about GIB.

If you have any settings I could try the "good" player version with flash bbo - I gladly would, until then I genuinely think that GIB's abilities are overstated, to say the least. I would expect a better and more honest assessment from bbo.

[1eyedjack]

I think that there is a case for protecting with 3C, ie concealing the Spade support. I perceive a danger of getting cut off from the Clubs, and even an 8 card Spade fit as trumps may not make up for that. Knowing that partner has Spades is of some value in upgrading the hand even if playing in Clubs as trumps. I realise that on some hands 2S could well work out the winner. I agree that it does not seem right to pass (although I also wonder who has the hearts).

[cloa513]

West probably has hearts and little or no HCP. East has Axx AKQ AKxxx xx

[hotShot]

GIB is a below average bidder and yes GIB ignores carding, but GIBs cardplay both as defender and declarer usually does not give away tricks. The error rate is close to world class players.

If you average that to a single grade, than GIB is significantly better than the average BBO player.

 

But the GIB does not "think" like a human, so GIBs errors often look "stupid" by human standards.

You probably know that GIB generates random deals based on the informations about the hands and searches for the best double dummy play in those deals.

Obviously the generated deals will get closer to the actually played deal the more information is available to GIB.

GIB bases its decision on different deals and the selected play could be "stupid" for this deal.

[helene_t]

Maybe the GIB simulations show that reopening will often help opps find a making 4♥?

[fred]

We have done some studies that make it clear that, despite its weaknesses, overall GIB is much more effective at bridge than the average ACBL Robot Duplicate player. It is not even close.

 

Since I am guessing that, in general, tournament regulars on BBO (especially ACBL tournament regulars) are stronger players than those who do not play in tournaments, the conclusion in the title of this thread is almost certainly (very) wrong.

 

Fred Gitelman

Bridge Base Inc.

www.bridgebase.com

[xxhong]

This comparison may not be fair. Gib can certainly be more effective than human players if both sides partner with gib. However, if you put one gib pair in a normal tournament, their performance may not really be above average, especially if you put gib in a tournament where most players know gib well. Basically, gib declares better than average player ( still, no way close to expert players because of some of the basic design problems). Gib's defense is just not comparable with average players because there aren't any defensive signals. For defense, often, it's not a matter of simulations. In latter rounds, it's often a sure thing after accurate signals. Gib's bidding is above average in some areas (after NT openings) and rather bad in other areas (slam bidding, competitive bidding, sequence after 2/1 and 2C and inverted minors). So I think generally, gib's bridge is comparable with average tournament players and can be weaker.

 

Finally, let me give you a deal to show how bad gib's defense is and how expensive it can be. (Against human players, I wouldn't redouble. Also, I know my human opp is aggressive from previous hands)

 

Here I sit south, the other human player sit west. N and E are gibs.

We have done some studies that make it clear that, despite its weaknesses, overall GIB is much more effective at bridge than the average ACBL Robot Duplicate player. It is not even close.

 

Since I am guessing that, in general, tournament regulars on BBO (especially ACBL tournament regulars) are stronger players than those who do not play in tournaments, the conclusion in the title of this thread is almost certainly (very) wrong.

 

Fred Gitelman

Bridge Base Inc.

www.bridgebase.com

[hv=pc=n&s=sa73hkq64da75cjt5&w=sj952h952d4caq632&n=sk84h8dkqj832c874&e=sqt6hajt73dt96ck9&d=w&v=b&b=4&a=ppp1cp1d1h1n2h3dp3ndpprppp&p=h2h8hah4ckc5c6c4hjhkh5c7dad4d2d6d7c2dkd9dqdtd5c3djh3s3s2d8c9cth9d3s6h6cqs4stsas5hqs9c8h7s7sjsksqs8htcj]399|300[/hv]

[fred]

Yes, of course GIB's current failure to understand signalling in many contexts often leads to brutally bad defense.

 

But you could say the same thing about (for example) the average human player's complete inability or unwillingness to count shape, points, and tricks.

 

Both GIBs and average human players have clear strengths and clear weaknesses with respect to one another, but in the contests on BBO in which GIB participates, the bottom line is that overall GIB is considerably more successful than the average human player. It should also be noted that in the tests I am referring to, the human has had the advantage of "knowing" that he/she was always dealt the best hand at the table.

 

Something that is easy to forget: almost all regular posters to BBO Forums are probably in the top 10% of the general BBO membership in terms of bridge skill. I strongly suspect that several of you are easily in the top 1%. You may be better than you realize (compared to average) and that may make GIB seem worse than you realize (compared to average).

 

Fred Gitelman

Bridge Base Inc.

www.bridgebase.com

[Yu18772]

I think that there is a case for protecting with 3C, ie concealing the Spade support. I perceive a danger of getting cut off from the Clubs, and even an 8 card Spade fit as trumps may not make up for that. Knowing that partner has Spades is of some value in upgrading the hand even if playing in Clubs as trumps. I realise that on some hands 2S could well work out the winner. I agree that it does not seem right to pass (although I also wonder who has the hearts).

 

I would say that E doesnt have 4 hearts (would double again) and has a bunch of high cards with 5/6+ diamonds and W probably also holds only 4, or if 5 it has with nothing at all in terms of table entrances. So I wouldn't be worried about them playing 4!h. May be 3!c will work I dont mind either, although I dont like concealing support for the highest ranking suit in what looks like partscore competition.

 

But as I said - my main frustration is not with bids.....this is something I can try to adjust to. Its the defense....which is 50% of the time......

[Yu18772]

Yes, of course GIB's current failure to understand signalling in many contexts often leads to brutally bad defense.

 

But you could say the same thing about (for example) the average human player's complete inability or unwillingness to count shape, points, and tricks.

 

True, but I to a human I can try to point out that there might be a better line, or switch a partner. I can't switch GIB, for another GIB......

 

Both GIBs and average human players have clear strengths and clear weaknesses with respect to one another, but in the contests on BBO in which GIB participates, the bottom line is that overall GIB is considerably more successful than the average human player. It should also be noted that in the tests I am referring to, the human has had the advantage of "knowing" that he/she was always dealt the best hand at the table.

 

You should also note that in these tests the human is playing with a partner which is hard to understand, and impossible to discuss with, while the two GIB opponents play with a perfect partner from that point of view. One could only compare the level of GIB in such cases, if the human partner is well adapted to GIB, and as i understand - the better players dont usually play with GIB....so you apriori compairing to weaker audience that has additional challenges. If you want to compare GIB to an average player, put GIB pairs into human tournaments. About the best hand - the human hand has the highest high card point count, which is hardly the only relevant measure for the quality of the hand, and it was my impression that the lack of points is often times well compensated by distribution.

 

Something that is easy to forget: almost all regular posters to BBO Forums are probably in the top 10% of the general BBO membership in terms of bridge skill. I strongly suspect that several of you are easily in the top 1%. You may be better than you realize (compared to average) and that may make GIB seem worse than you realize (compared to average).

 

Fred Gitelman

Bridge Base Inc.

www.bridgebase.com

 

Thank you, but as I said in my original post, this doesn't seem to be a compliment for GIB.

 

Yehudit Hasin

[bd71]

My matchpoint average against 3 GIB robots in ACBL robot duplicates: 52.6% over ~5,000 boards

 

My matchpoint average against 3 humans, usually in ACBL tourneys but occasionally in main bridge room: 52.5% over ~800 boards

[Bbradley62]

My matchpoint average against 3 GIB robots in ACBL robot duplicates: 52.6% over ~5,000 boards

 

My matchpoint average against 3 humans, usually in ACBL tourneys but occasionally in main bridge room: 52.5% over ~800 boards

Your matchpoint score in ACBL robot duplicates is not a measure of whether you beat the robots; it's a measure of whether you beat the humans who are sitting in your seat at other tables.

[booggie]

We have done some studies that make it clear that, despite its weaknesses, overall GIB is much more effective at bridge than the average ACBL Robot Duplicate player. It is not even close.

 

Since I am guessing that, in general, tournament regulars on BBO (especially ACBL tournament regulars) are stronger players than those who do not play in tournaments, the conclusion in the title of this thread is almost certainly (very) wrong.

 

Fred Gitelman

Bridge Base Inc.

www.bridgebase.com

 

I suggest that in every Robot Duplicate game you also enter one robot ("ROBO") as South. You might make him ineligible for masterpoints, but in each game ROBO will get his MP score. After 1000 or so games, ROBO's average masterpoint score would give a measure of how well the robots play as compared to the humans in those tournaments.

 

There are some biases in this test. The human Souths all know they have the best hand, but ROBO won't. But ROBO should understand the bidding of his North partner better than most human Souths do. I think the result would be very interesting. From my experience I remain skeptical that the robots play as well as the average ACBL Robot Duplicate player.

 

Art Quaife

[fuburules3]

I suggest that in every Robot Duplicate game you also enter one robot ("ROBO") as South. You might make him ineligible for masterpoints, but in each game ROBO will get his MP score. After 1000 or so games, ROBO's average masterpoint score would give a measure of how well the robots play as compared to the humans in those tournaments.

 

There are some biases in this test. The human Souths all know they have the best hand, but ROBO won't. But ROBO should understand the bidding of his North partner better than most human Souths do. I think the result would be very interesting. From my experience I remain skeptical that the robots play as well as the average ACBL Robot Duplicate player.

 

Art Quaife

 

I'm sure they have had the GIB play the hands from robo tourneys and then matchpointed the results. I think I heard somewhere that the GIB averages about 55% in such simulations--perhaps someone who knows can provide a little actual data.