[rfaronson]

Your AI is broken. When it decides that the proper thing to do is to ruff a suit, but there is a prior ruff ahead of it and it cannot overruff, then it will underuff. That often costs a needless trick.

As I was the bridge designer and programmer for The Sierra Network and Hoyle Classic Bridge, I know a fair about about bridge AI. I've tried reporting bugs through your bug report staff and the last one told me to take it to the forums. It's obvious you have little interest in improving your bot play, but some of these bugs are egregious enough to make me try.

Your random number generator is also clearly off.

Your bidding often incorrectly describes the bid. If a description says 4+ cards in a suit, If your bidding says 7+ points in response to my bid, it can NOT count short suit points for a stiff or a void in my opened suit. If the hand is too weak to bid over my first named suit or my second named suit, it is FAR to weak to bid a new suit when I rebid one of my suits in passout position, forcing me up another level because my void won't play well opposite the AI's weak six card suit. For that matter, if I bid two a second suit instead of reopening with a double, I CANNOT HAVE four card support for the AI's newly introduced suit in an auction like:

1H - 2C - P - P
2D - 3C - P - P
3H - P -- 3S

I've seen the AI bid 3S on fewer than 4 HCP with a suit when there aren't enough cards in my hand to have sufficient support given the ten red cards I am already showing.

And these are just off the top of my head. I have a log file of many significant bugs that I'm not bothering to report because I have not received a single reply indicating BBO cares to fix these bugs.

Yours truly,
Richard Aronson

[johnu]

Your mistake is in thinking that GIB is an AI program. GIB is a collection of If/then/else and various loops. An AI program would "learn" from previous hands. GIB will make the same mistake 10,000 times in a row if a programmer does not program something differently. GIB does use a good double dummy analyzer, but double dummy simulations are only as good as the parameters used to deal the other hands. Frequently the parameters are at odds with reality on the ground.

If a bid is not specifically defined, the results can be totally random.

[pilowsky]

He's right, Although GIB stands for Ginsberg's Intelligent Bridgeplayer. It should really be Ginsberg Intelligent Bridgeplayer. Ginsberg is intelligent, not the program. Tragically, the Intelligence stuck with the programmer, none of it passed on to the Robots. Mind you, the people on the Forum are very friendly, intelligent and helpful - or some combination of the above. Some of them actually know something about Bridge as well.

[smerriman]

rfaronson, on 2020-July-13, 18:36, said:

Your AI is broken. When it decides that the proper thing to do is to ruff a suit, but there is a prior ruff ahead of it and it cannot overruff, then it will underuff. That often costs a needless trick.

I doubt it; GIB will often throw away tricks, but the way GIB is programmed isn't remotely close to what you suggest; it will underruff if it calculates that, based on double dummy simulations of the current situation, it is equivalent (or better) to playing anything else. It falls down when it's simulating too few hands (or making assumptions based on the bidding, which are often too strict).

rfaronson, on 2020-July-13, 18:36, said:

Your random number generator is also clearly off.

Many have made this claim; not a single person has even been able to provide any evidence to demonstrate it. I've done numerous analyses of large sets of hands and every single time they lie within normal bounds. Stating it without evidence is pointless and most of the time is based on a lack of statistical knowledge.

rfaronson, on 2020-July-13, 18:36, said:

Your bidding often incorrectly describes the bid. If a description says 4+ cards in a suit, If your bidding says 7+ points in response to my bid, it can NOT count short suit points for a stiff or a void in my opened suit.

It may be bad, and it is, but it's not 'incorrect' if GIB has been programmed that way, and it has.

rfaronson, on 2020-July-13, 18:36, said:

.. because I have not received a single reply indicating BBO cares to fix these bugs.

They don't.

[johnu]

smerriman, on 2020-July-14, 02:17, said:

It may be bad, and it is, but it's not 'incorrect' if GIB has been programmed that way, and it has.

Yikes, a bug is a bug is a bug. Maybe it was an oversight, maybe it was a lack of bridge knowledge, maybe it was carelessness or not enough time to do bug fix, maybe QA didn't pinpoint a problem, but that's a serious bug and is not correct.

smerriman, on 2020-July-14, 02:17, said:

They don't.

On rare occasions, if you post in the GIB Robot Discussion forum you may get a response. Sometimes they say they will look at it without promising to fix anything, and sometimes they flat out say they can't/won't fix a specific bug.

[FelicityR]

If it's any consolation at least the robots haven't taken over our favourite game - yet

I remember world champion Gary Kasparov - my son was into chess at the time - losing his first chess game against a computer in the mid 1990s. Then another computer beat him fairly and squarely in a series of matches as I recall. I thought at the time 'Bridge will be next.' How wrong I was.

I realise it must be annoying for those players playing with and against robots, but bridge essentially is a social game. Long may it continue.

[hrothgar]

FelicityR, on 2020-July-14, 03:29, said:

If it's any consolation at least the robots haven't taken over our favourite game - yet

I remember world champion Gary Kasparov - my son was into chess at the time - losing his first chess game against a computer in the mid 1990s. Then another computer beat him fairly and squarely in a series of matches as I recall. I thought at the time 'Bridge will be next.' How wrong I was.

I realise it must be annoying for those players playing with and against robots, but bridge essentially is a social game. Long may it continue.

The only reason why bridge hasn't been next is that relatively few AI researchers care about the game

[pilowsky]

AI is the shortest oxymoron

[smerriman]

johnu, on 2020-July-14, 02:59, said:

Yikes, a bug is a bug is a bug. Maybe it was an oversight, maybe it was a lack of bridge knowledge, maybe it was carelessness or not enough time to do bug fix, maybe QA didn't pinpoint a problem, but that's a serious bug and is not correct.

A bug is unintentional. It seems pretty clear to me that GIB's point-counting method was intentionally simplistic. Given its heavy reliance on point counts, it would probably have broken entirely if they got it to reevaluate based on the auction (and that was part of what Advanced GIB was meant to handle anyway).

johnu, on 2020-July-14, 02:59, said:

On rare occasions, if you post in the GIB Robot Discussion forum you may get a response. Sometimes they say they will look at it without promising to fix anything, and sometimes they flat out say they can't/won't fix a specific bug.

Extremely rare, and hasn't happened for a long time, and definitely not since the merger with FunBridge.

[pescetom]

smerriman, on 2020-July-14, 14:41, said:

Extremely rare, and hasn't happened for a long time, and definitely not since the merger with FunBridge.

I assume you are thinking specifically about fixes to GIB.
In other areas of the software I have received some negative replies (occasionally embarassing) from the developer, but also fixes or attempts at fixes sometimes even when no such promise was made.
On the whole I think BBO is paying more attention to us now than in the past and just as well too.

[smerriman]

pescetom, on 2020-July-14, 15:05, said:

I assume you are thinking specifically about fixes to GIB.

Correct. BBO in general is going well, but they seem to have given up on GIB entirely. One case in point was my bug report from 2016, where Fred himself said he'd make sure the programmers looked at it, over a year later Diana said she'd follow it up, and it's still occurring today.

jdonn - who was the primary point of contact on the forums and said back in 2017 he should and will start replying to forum threads more regularly - hasn't even visited the forum in over 2 years. (Maybe no longer working at BBO?)

However, I'm hoping that's in part due to a long term plan to move to Argine.

[johnu]

smerriman, on 2020-July-14, 14:41, said:

A bug is unintentional. It seems pretty clear to me that GIB's point-counting method was intentionally simplistic. Given its heavy reliance on point counts, it would probably have broken entirely if they got it to reevaluate based on the auction (and that was part of what Advanced GIB was meant to handle anyway).

It's poor/incomplete programming, especially after a quarter century of existence. Failing to reevaluate a hand based on a potential trump suit is something that should have been a high priority to fix. Much like some of the conventions that are implemented. Maybe the 1st bid in a conventional sequence is coded, maybe even the response to a conventional sequence is code, but then there is nothing coded and GIB will get "lost" in the bidding.

IMHO, any programmer would be embarrassed to have this coding (or lack or coding) in a fully mature product.

[armant2k]

I am writing only because of the curious Robot behavior that was starkly exhibited during this hand.
(url="http://https://tinyurl.com/y2mh2ugn")


We were playing against 2 robots, and my wife left me in my 3NT opener (26 HCP). With her long Clubs, I had 11 tricks cold after the opening lead wasn't a Spade.

I also had 6 Diamonds, ♦AKQ2 in my hand, ♦94 in Dummy. During the first 9 tricks the E Robot discarded 2 Diamonds, so rather than claim 12 tricks, I lead the ♦A to test the waters, although I still fully expected to lose the 13th trick. I was surprised to see W drop the ♦J and E drop the ♦T. Could E have started with 6 Diamonds and E a stiff ♦T?

Curious, I continued with the ♦K and found the E Robot was out of Diamonds. The W Robot followed with the ♦5, and put the ♦6 on my ♦Q. He kept the ♦3 to beat my ♦2 on the last trick.

(I wonder if E had instead originally held ♦T763, would he have held onto the ♦7 to win the final trick, just to make his robot partner buy him a beer? )

The GIB robots apparently are programmed to toss the highest of equivalent cards, and since they 'remember' every card played, the W Robot in this case 'knew' the ♦3 was equivalent to the T, 6, or 5. I have noticed in the course of many hands played versus the basic GIBs robots, it appears that the GIB program addresses "Restricted Choice" situations by (almost?) always discarding from the top.

[pilowsky]

I recently started playing high encourage. Now I almost reflexly discard the highest of equivalent cards if only to remind myself that all the cards in the suit are 'top'. Just today I recall discarding Ace from AQT after the J fell under the K. It was quite comforting to win the T over the 9 at the end.
My understanding of GIB is that they use RC to prevent really clever people from figuring out where the Honour is. This is useful if you are working out the German tank holding in WW2, but probably not that common amongst the average Bridge-player - by which of course I mean me. Not you.

[kontoleon]

Some times Robots throw tricks out of windows, with out any reason to do this.

E.G last 3 hands from robot was A3!s and small other card. (no trumb) he run the Ace discard from his partener, and keep playing 3s even if he know that opponent have one more!

Or just throw K on Kx even if ace played by opponent! But we love GIB errors