[lycier]

Today I found a Gib blind spot on bidding, now see my traveller :




- From the first to eighth hand, all the bidding sequences are same.

My hand is classic.

---

Result : 4♥Ex-3
No double, no trouble. After double, I found that it is easier for Gibs to make some mistakes when responding cuebid opening suit to show 11hcp and 11-16TPs,forcing. Its mistakes are it is often difficult to stop correctly, often forcing to game with some wrong description, 4♥ is just a best example,

- Frome the ninth to 15th hand, all the mistakes are caused by south human players.
Here is a classic hand.

---

Result : 3♦N-4
Many people played rediculously.

- Here is last hand.

---

Result : 4♣N-5
The double says " 5-HCP", obviously this bid is undefined. This is another issue.

[manudude03]

In the first lot of hands, why isn't East bidding 2NT over 2S?

[1eyedjack]

And I don't think that East is strong enough for 2D opposite a protective X.. West has already borrowed one of his Aces. I would rank East options over X in decending order of preference 1N, 2H, 1H, 2N, 2D.
Actually I am a bit undecided about the ranking of the first 4 options. Just not about the 5th.

[iandayre]

This is a good example of how GIB bids one round at a time. I consider the cuebid a poor choice, since E can anticipate partner bidding Spades. Now you are at the 3 level, knowing partner likely has only 3 Hearts. Much better 2H the first time, which should end the auction. Yes even as it went, 4H is awful.

[Stephen Tu]

iandayre, on 2016-April-19, 12:18, said:

This is a good example of how GIB bids one round at a time.

1 round at a time is perfectly fine. It just needs the priorities of its rules ordered correctly. Like here have the priority for just jumping in a major higher than cue bidding when holding only one major, and not being strong enough to force to game.

It's probably really hard to program bidding trying to looking into future rounds of the auction given how many possibilities there are. Just get the current round right, you'll have something reasonable to bid on the next round. The rules just have to be carefully designed to account for that.

[lycier]

I am wondering why Stephen Tu reply is always so noticeable, I perfectly agree with him on the matter.

[johnu]

Stephen Tu, on 2016-April-19, 12:33, said:

1 round at a time is perfectly fine. It just needs the priorities of its rules ordered correctly. Like here have the priority for just jumping in a major higher than cue bidding when holding only one major, and not being strong enough to force to game.

It's probably really hard to program bidding trying to looking into future rounds of the auction given how many possibilities there are. Just get the current round right, you'll have something reasonable to bid on the next round. The rules just have to be carefully designed to account for that.

You have actually explained why 1 round at a time is not perfectly fine. By designing or redesigning bidding rules so you will not have an impossible problem on the next round, you are explicitly endorsing the viewpoint that 1 round at a time is not good enough.

[Stephen Tu]

johnu, on 2016-April-19, 15:26, said:

You have actually explained why 1 round at a time is not perfectly fine. By designing or redesigning bidding rules so you will not have an impossible problem on the next round, you are explicitly endorsing the viewpoint that 1 round at a time is not good enough.

The humans designing the rules and their priority in relation to another have to keep multiple rounds of bidding in mind and adhere to principles that have been tested based on decades of experience. Then they boil this down into a set of instructions that tell the computer how to pick a bid on *one single round*, the current round.

The computer, however, does *not* need to think about future rounds. If it is presented a well designed rule set, it doesn't have to anticipate future rounds and select on based on avoiding problems. It can simply follow the human well-designed one-round at a time rule set and land on something reasonable. Here, if this sequence were fixed, it would just simply find the rule for jumping in a major matching a higher priority rule than cue bidding, and chooses that. It would have no idea *why* it's a bad idea to cue bid, and be choosing that rule based on anticipating being stuck if partner bids 2S. It doesn't need to know at all about future rounds. All it has to know is that humans think that 2H is the best bid on this round, it doesn't need to know reasons or have a plan for the next round.

On the next round it will simply look into the rule set for the next round of bidding.

[johnu]

Call it what you want.

Well designed rules will implicitly consider the next round(s) of bidding. The computer doesn't have to think about the next round of bidding because the current round of bidding should already have given it consideration.

Example: With 15 HCP, a 5 card major and semi-balanced distribution, many players open 1NT instead of 1 of a major. The reason is to solve a rebid problem if responder bids 1NT. If the computer is programmed to open these hands 1NT, the computer doesn't have to think about subsequent rounds of bidding before making the bid, it just looks up the opening in the bidding library. The next round of bidding has already been considered.