[1arrygiff]

On this deal Ben, North, made a 3-level Jacoby transfer with 4 pieces (showing 5), then raised (showing 6), announcing "Jacoby transfer - 5+ hearts".<br><br>This type of error should be easily preventable. My confidence in Ben took a nosedive.<br><br>

---

[smerriman]

1arrygiff, on 2025-April-10, 09:29, said:

This type of error should be easily preventable.

Out of interest, why would you say this? Based on my understanding, Ben is not told what any bids mean or what any of the descriptions say - the alerts come from GIB and are just provided to humans; Ben had to figure everything out from scratch from example auctions, and this type of delayed 2nt in interference isn't exactly common to have occurred regularly in training.. so I would have said this type of error is very hard to *avoid*.

At least, that's how it worked originally, then since it was getting Blackwood responses wrong, some very specific bids were recoded to come from GIB instead.

[1arrygiff]

smerriman, on 2025-April-10, 13:35, said:

Out of interest, why would you say this? Based on my understanding, Ben is not told what any bids mean or what any of the descriptions say - the alerts come from GIB and are just provided to humans; Ben had to figure everything out from scratch from example auctions, and this type of delayed 2nt in interference isn't exactly common to have occurred regularly in training.. so I would have said this type of error is very hard to *avoid*.

At least, that's how it worked originally, then since it was getting Blackwood responses wrong, some very specific bids were recoded to come from GIB instead.

Maybe "easily" was an overbid and you could be right. Showing a six card suit holding a 4-4-3-2 hand seems extreme. My thinking was that lacking a 5+ card major, a Jacoby transfer bid could be omitted from eligible calls programatically, forcing choice from remaining legal calls only, or, failing that, suit length be omitted from the explanation. These were expedients that occurred to me. I take it that's difficult in an AI-driven process.

[Huibertus]

I plaid the same hand, same 3♦ and 4♥ bid. I belief BEN actually plays Lebensohl here. 3♦ weak sign off that makes 4♥ understandable.

What is wrong is is the annotation of the bids, which indeed should be easy to fix, it's a database with bidding sequences and the meaning of the last bid after all.

Sadly these inconsistencies frequently occur. One I always hate is a 4♣/♦ bid explained as forcing only to see a pass right after you bid it.

[lorserker]

sorry about this.

smerriman explained well.
ben finds itself in an unfamiliar situation and improvises, inventing stuff on the spot.
there are no bidding rules programmed (probably we should add some)

you will find ben consistently bidding better than gib, but on occasion losing its mind like this.
we'll try to reduce the frequency of these brain farts as close to zero as we can.

[johnu]

lorserker, on 2025-April-11, 13:48, said:

there are no bidding rules programmed (probably we should add some)

I've wondered about how much AI should go into bridge bidding and play. Clearly for something like chess, go, backgammon, where there is no partner and there is no disclosure to the opponents except for the move you just made, no thought needs to go into disclosure or agreements/rules, other than basic laws of the game.

IMO, to be a good robot player for human partners and opponents, it should be able to play the popular conventions and bidding styles that are commonly used in various parts of the world. But rather than rely on random human players, those conventions and styles need to be based on specific sources that are integrated into a coherent system, because that's what human partners expect. Same for carding, leads and signals. Those should be selectable by the user.

[dsharpcctx]

I believe that the user should be entitled to accurate explanation of the bids made by the robot player. If the robot player is "just guessing," then that should be the explanation of the bid. To default to the GIB descriptions is misleading. If judgment is used, (which is okay,) then that should be noted in the description.

I really have no good understanding why BBO does not offer the option of the user selection the conventions of the robot partner from a list of known conventions. I downloaded a website a bridge playing program that plays reasonably well and offers a list of different conventions that it plays. I do not see why BBO could not do the same (other than possibly laziness.)

The same thing could be said regarding carding.

[smerriman]

dsharpcctx, on 2025-April-11, 21:08, said:

If the robot player is "just guessing," then that should be the explanation of the bid. To default to the GIB descriptions is misleading. If judgment is used, (which is okay,) then that should be noted in the description.

Oh, that would make things incredibly simple - Ben can then just describe every bid as 'just guessing'

[lorserker]

smerriman, on 2025-April-11, 22:22, said:

Oh, that would make things incredibly simple - Ben can then just describe every bid as 'just guessing'

yes and some guesses are better and coincide more with actual system than others