[1eyedjack]

iandayre, on 2016-January-21, 18:06, said:

It had nothing to do with programming, it had to do with that fact, as I said originally, that the descriptions in GIB, in many cases, are not in keeping with sound bridge principles. This strongly indicates to me that the original programmers were not experts. If they were, in my opinion they did a poor job.

I think perhaps that the flaw in your logic is an expectation that because an explanation is provided for a bid, that that explanation must have resulted from a considered human bridge player/programmer, in order that their expertise may then be called into question. In many cases, and I suspect in most of the more stupid ones, the explanations come as a generic response to the fact that the sequence has NOT been specifically programmed in. That leads me to discount the conclusion that the programmers were not experts, but it still leaves open your final fall-back conclusion that if they were experts then they did a poor job. That can only be measured against a defined expectation of what would have been a "good" job having regard to the difficulty of the problem, which is way beyond my grasp but I am perfectly happy to accept that it is a monumental task if I am so informed.

[lycier]

Please don't be excited, don't be angry. I think the bridge must be happy. To be honest, most of us are fanatic fans of Gib on the BBO and we had some specious comments on Gibs in the past. This is very normal, after all, most of us don't understand programming, nor is a bridge expert, perhaps most of us can give BBO neither an reasonable opinion nor any useful advice. So we look forward to get the bridge experts of Gib programming to help, I think the best candidate is Stenphen Tu.

[Vampyr]

1eyedjack, on 2016-January-17, 23:01, said:

I would have thought that it was sufficient that it requires the same skill set.

OK, pretend that I said "requires the same skill set" in my earlier post.

I don't play on BBO or with any kind of robot, so I don't know how much playing ŵith a robot involves "gaming" the robot. I know that some people mention opening 1NT on very unsuitable hands. And there was a thread recently in which GIB was shown to play a hand hilariously badly. Are there bidding methods that will prevent your GIB partner from playing a hand worse than it would do at another table? There are definitely, as shown in the OP, calls that must be avoided so that GIB doesn't go berserk. So some of the skills are certainly different, but if the situations when these skills are needed occur infrequently, then skill at bridge becomes important.

I do find playing most of one's bridge with a robot partner and opponents a bit sad, but that is irrelevant to the discussion.

[helene_t]

Vampyr, on 2016-January-22, 03:38, said:

I don't play on BBO or with any kind of robot, so I don't know how much playing ŵith a robot involves "gaming" the robot. I know that some people mention opening 1NT on very unsuitable hands. And there was a thread recently in which GIB was shown to play a hand hilariously badly.

As for the hilariously badly play I wouldn't worry about it. We all sometimes play badly and masterpoints are awared for play against human opps who play a lot worse than GIB does.

As for the special tacticts involved in playing against robots, there is of course some truth in it. Just like if I played most of my bridge against a particular pair who had some habbits I was aware of (never leading trumps, never upgrade or whatever) I would learn to take advantage of that.

[Vampyr]

helene_t, on 2016-January-22, 04:15, said:

As for the special tacticts involved in playing against robots, there is of course some truth in it. Just like if I played most of my bridge against a particular pair who had some habbits I was aware of (never leading trumps, never upgrade or whatever) I would learn to take advantage of that.

Quite. But this doesn't happen and playing with robots does.

[iandayre]

Stephen Tu, on 2016-January-21, 22:57, said:

It's just that your opinion is coming from someone with admittedly no background in computer science, who has no programming experience. You don't have a solid grasp of just how many possible auctions there are in bridge, and how hard it is to construct enough rules for a stupid machine to try to cover every one that comes up in practice. It's like criticizing a dog trainer that when they are done you don't have a talking dog with human intelligence like Brian on the animated Family Guy show. Unrealistic expectations for even expert bridge players who are also expert programmers.

I think professional computer programmers, like me, are more understanding of how difficult it is to train a computer for a game with as many possible sequences as bridge.

Go look at Truscott's bidding dictionary. See how many pages he has to cover, and that's only for first 4 bids of uncontested sequences. Imagine writing rules to cover all of those, and all the contested ones, and the ones later in the auction. And ordering the rules so the computer picks the best bid consistently. It's not so easy!!

Stephen, I also respect you a lot. I openly acknowledge my lack of programming knowledge, and by no means do I suggest that it is easy or can be done overnight. My problem is with the DESCRIPTIONS, haven't I said that multiple times? If the descriptions are poor, perfect programming performance would still result in poor bridge bidding. I believe that you or I or many other experienced players could write better descriptions for many GIB auctions, and then it would be up to the programmers to implement them.

[helene_t]

iandayre, on 2016-January-22, 11:42, said:

My problem is with the DESCRIPTIONS, haven't I said that multiple times? If the descriptions are poor, perfect programming performance would still result in poor bridge bidding.

The programmers main task is to generate the descriptions, i.e. assign meaning to the alternative calls - the rest is relatively straight-forward.

And since descriptions (i.e. meanings of calls) have to be generated for literally quadrillions of different auctions, you cannot just ask a bridge expert to provide them all. You have to try to make an algorithm that generates them on the basis of some generic principles which you have to try to formulate yourself. This is what is tremendously difficult, even for someone who is both a bridge expert and an expert in artificial intelligence. That is the whole problem.

[Stephen Tu]

What Helene said.

It is roughly the difference between:
1. Give expert a sequence, he can tell you what a particular bid ought to mean. But this is infeasible because you can't just write down every possible bridge sequence and ask him to define them all because it would take like longer than the age of the universe, and even if you could you couldn't really just store all the descriptions in one gigantic lookup table.
vs.
2. Have expert write a rule set down, such that no matter what the auction is, creates a flow chart such that a person who knows absolutely zero about bridge could follow the steps and arrive at a reasonable description in terms of strength and distribution limits of the 4 suits.

GIB is attempting to do 2, it has a set of probably thousands of rules that given a particular sequence, spits out some sort of definition. The problem is that the number of possible sequences is so huge, that no matter how many rules are written, inevitably some sequence is new and isn't covered, and it falls back on basic rule defaults that might make sense in a general way but look like nonsense in the context of the auction.

Humans are good at assigning meaning based on context, parsing novel auctions based on prior experience in analogous auctions. Computers have like zero ability to decide what constitutes an analogous situation. Everything has to be written out explicitly for them.

[lycier]

iandayre, on 2016-January-22, 11:42, said:

My problem is with the DESCRIPTIONS, haven't I said that multiple times? If the descriptions are poor, perfect programming performance would still result in poor bridge bidding. I believe that you or I or many other experienced players could write better descriptions for many GIB auctions, and then it would be up to the programmers to implement them.

Sorry,I never believe you could have ability to write better descriptions for many GIB auctions than the bridge experts of BBO Gib programming, I strongly think that not all the experts/experienced players could have ability to write better descriptions for Gib, just like civil engineering, not all the Engineering designers could be excellent building workers and not all the building workers could be good Engineering designers. You should know the programming of Gib robot is also a great feats of challenging to human intelligence. A great pity that you don't think so, what you said just meant you could have more ability to write better descriptions than the bridge experts of BBO Gib programming.

[lycier]

iandayre, on 2016-January-22, 11:42, said:

and then it would be up to the programmers to implement them.

Listen to the opinions of the experts, you can learn more knowledge, I think Stephen Tu's comments have more guiding significance.

[barmar]

One of the biggest problems in designing the GIB bidding rules is dealing with the fact that bids can have very different meanings depending on the context, and describing all the possible contexts is difficult. Competitive bidding increases the number of possibilities greatly, and trying to describe them in general terms, in the limited language of GIB's bidding rules, is daunting.

What happened in the case in the OP is that there's no specific rule for this type of competitive sequence. It doesn't have any general knowledge about sacrifices, like human players do. So it falls back to a default rule that in effect says "If you bid to level N, you expect to make it, so you have enough points, combined with what your partner has shown, to make that a reasonable contract". Making a contract at the 4 level requires about 25 HCP; since partner has shown 0, that must be how much you have.

The original programmer of GIB was indeed not a bridge expert, his expertise is in artificial intelligence. I also suspet he was more interested in the playing part of the game than the bidding -- that's the part that he published papers about in AI journals.

[johnu]

I've asked before but is BBO considering changing the bridge engine from GIB to one of the bridge programs that are being actively updated by the original programmers? Has BBO done any comparisons of GIB versus the best of the current programs? I know GIB was a former computer bridge world champion, but that was an awfully long time ago, and the competition has had many years to drastically improve their bidding and play.

[iandayre]

barmar, on 2016-January-22, 19:34, said:

One of the biggest problems in designing the GIB bidding rules is dealing with the fact that bids can have very different meanings depending on the context, and describing all the possible contexts is difficult. Competitive bidding increases the number of possibilities greatly, and trying to describe them in general terms, in the limited language of GIB's bidding rules, is daunting.

What happened in the case in the OP is that there's no specific rule for this type of competitive sequence. It doesn't have any general knowledge about sacrifices, like human players do. So it falls back to a default rule that in effect says "If you bid to level N, you expect to make it, so you have enough points, combined with what your partner has shown, to make that a reasonable contract". Making a contract at the 4 level requires about 25 HCP; since partner has shown 0, that must be how much you have.

The original programmer of GIB was indeed not a bridge expert, his expertise is in artificial intelligence. I also suspet he was more interested in the playing part of the game than the bidding -- that's the part that he published papers about in AI journals.

Well thank you Barmar. I guess I'm not as dumb as some of the others here seem to think I am. From now on I am not going to post poor GIB actions on this board, though I will continue to make robot reports, and to comment on others' hands. I will look forward with anticipation to future upgrades.

So, it would seem to me that default rule should be changed. Everyone knows it is often necessary to assume some values in partner's hand when considering competitive action. And even with my lack of programming knowledge, that sounds like none too easy of a task. But perhaps progress can be made. Best of luck.

[iandayre]

lycier, on 2016-January-22, 17:21, said:

Listen to the opinions of the experts, you can learn more knowledge, I think Stephen Tu's comments have more guiding significance.

Right Lycier. If on an arithmetic scale of bridge skill and achievement, if Fred Gitelman is a ten, I'm about an eight, and you're about a three.

[mgoetze]

iandayre, on 2016-January-22, 20:40, said:

Right Lycier. If on an arithmetic scale of bridge skill and achievement, if Fred Gitelman is a ten, I'm about an eight, and you're about a three.

LOL. Your bridge achievements must be pretty impressive, I wonder why you bother with GIB when you could get paid to play with a human instead.

[steve2005]

Stephen Tu, on 2016-January-21, 22:57, said:

I think professional computer programmers, like me, are more understanding of how difficult it is to train a computer for a game with as many possible sequences as bridge.

Programming bidding in competition is difficult. Looking at books produced on the subject shows the problem. Eric Crowhurst wrote two bidding books on Acol. The one that covers partner opening and opponents silent is relatively slim and covers most situations. His book on competition is huge, took him forever to produce and doesn't cover huge areas of bidding even in a general way. So, I think it is easy to see why programming a computer for bidding in competition would be a problem. With a computer program, you can't cover all hands you can only set rules to follow in hands you can't cover specifically.

That said, it would be nice to balance or sacrifice without being raised on mediocre hands without trick taking capability. But I have no idea if Gib's program can be easily made to accommodate this.

[Stephen Tu]

barmar, on 2016-January-22, 19:34, said:

I also suspet he was more interested in the playing part of the game than the bidding -- that's the part that he published papers about in AI journals.

GIB's bidding rules were originally taken from another program Meadowlark bridge, by another developer (Rod Ludwig). Don't know exactly what their business arrangement was. Ginsberg mainly worked on the play engine, though I think he did some work bringing in simulation into the bidding.

[Stephen Tu]

iandayre, on 2016-January-22, 20:38, said:

Well thank you Barmar. I guess I'm not as dumb as some of the others here seem to think I am. From now on I am not going to post poor GIB actions on this board, though I will continue to make robot reports, and to comment on others' hands. I will look forward with anticipation to future upgrades.

Just because GIB falls back on some default rule on some sequence doesn't mean the programmers were bridge idiots. It's just they didn't write enough rules, even though they worked on thousands of them already. They probably concentrated on constructive bidding first since it's a smaller problem and relatively easier. Once you start adding competitive bidding the number of possible sequences start to look infinite.

I still think you aren't understanding the issue. Even if you had Hamman as the bridge expert signing off on all the rules, it's hard to write enough of them and organize them so that a reasonable rule is triggered in every single possible auction. It's not a symptom of lack of bridge expertise, it's a symptom of the difficulty & scope of the problem.

[1eyedjack]

I don't have enough experience of other computer bridge programs to speak with authority, but the impression that I get is that they are streets ahead of GIB. Jack in particular.

So it is all very well whining about the quadbillions of lines of code required, but the elephant in the room is that the problem has already been proven to be solvable. No doubt there remain exceptions when Jack etc also go off the rails. Perhaps just not as frequently or horrendously.

If GIB's bidding logic is built on shakey foundations, and current efforts are focused on patching like whackamole the myriad problems further down the line that are directly attributable to the fragility of those foundations, then I do not have great confidence in the future, and I think it sad to see so many resources go to waste. Better to bite the bullet and start again from the bottom up with more sensible defaults. That would require dividing resources, because any new engine would, despite its potential, take time in getting up to the current GIB level warts and all, and in the meantime the userbase would still want to continue with the current warts and all GIB.

[lycier]

iandayre, on 2016-January-22, 20:40, said:

Right Lycier. If on an arithmetic scale of bridge skill and achievement, if Fred Gitelman is a ten, I'm about an eight, and you're about a three.

Incorrect Iandayre. If fred is a ten, I think I would be a 1 at most. Of course, I'm glad to see you can overvalue me.The bridge game is only my hobby, as a profession, it is a matter of Fred,Uday and Stephen Tu, don't know whether it is a matter of you.
Meanwhile, sad to know "your an eight" can't be admitted by everyone since perhaps you could be a -8, very cool...
Iandayre, no use to let me know you would have been an eight. In order to be able to show your talent, hurry up, let Uday know you as possible, he is a big boss of bbo, and then you can tell Uday you have more ability to make better robot than Fred,let you be a chief robot expert.Let's see whether robot you make is World Class or W.C.!

Thank you for your overvaluing me.