[patroclo]

I have seen that in this edition there isn't jackbridge.Does someone know why ?

[Bbradley62]

Maybe he thinks he should give someone else a chance to win.

[helene_t]

Just noticed that the system restrictions are stricter than GCC. You can't play strong club for example. Apparently Jack played some SEF like system in 2013 but it used to play Moscito against other computers. Moscito is probably the system that is best implemented in Jack. Possibly the Jack team couldn't be bothered to update the World Championship compliant system when upgrading to version 6? Just speculating.

[Jinksy]

Giving a computer a Moscito-like program in a tournament seems like cheating, or at least very contrary to the spirit. You want the coders to be developing the best-playing AI, not have to spend half the development time rewriting its bidding understanding to deal with a specific competitor's differences.

[Zelandakh]

It seems to me that a relay-based system absolutely plays to the strengths of a computer. Why would a designer not want to do that? It is like entering a chess computer against human opponents and designing the opening repertoire to be as non-tactical as possible.

[helene_t]

Jinksy: Yes that's the reasoning. But explainging the SAYC 1♣ opening in strict logical terms will be quite daunting. Moscito openings can probably be explained in a briefer yet more accurate way.

Maybe the easiest solution is to allow all "natural" methods plus a limited set of well-defined conventions. Then everybody just has to program generic defence against "natural" methods and specific defences against the allowed conventions. But the defenders need to be able to make inference from the bidding wrt declarer's holding. If the discosure mechanism doesn't have a way to tell exactly which hand types you have denied when you rebid notrumps, exactly which hand types you can have when you respond 1NT etc., you give a huge advatange to programs that play a different style than opps assume.

I think the best solution might be along the line of allowing opps to query the meaning of your auction by means of "give me 100 random hands consistent with the auction" or "tell me whether East could have AJ94-KQx-K982-54".

If I were to organize this competition I would make it as an indy, everybody playing for example SEF. That is probably a system most commercial products would want to implement anyway, and making it an indy had the big advantage of avoiding the disclosure issue.

[Jinksy]

Zelandakh, on 2014-September-12, 04:00, said:

It seems to me that a relay-based system absolutely plays to the strengths of a computer. Why would a designer not want to do that? It is like entering a chess computer against human opponents and designing the opening repertoire to be as non-tactical as possible.

It obviously helps the program that uses it - probably far more so by being artificial in a way the other AIs aren't specifically programmed to deal with.

But I don't see why an AI winning that way is of much interest to anyone - it's essentially a bug exploit.

[hrothgar]

Jinksy, on 2014-September-12, 06:35, said:

It obviously helps the program that uses it - probably far more so by being artificial in a way the other AIs aren't specifically programmed to deal with.

But I don't see why an AI winning that way is of much interest to anyone - it's essentially a bug exploit.

There's more than one way to skin the cat.

You seem to be assuming that the the best way for computers to disclose information to one another is to use the same type of language that humans use to describe their agreements.

I'd think that it would be far easier all around to have the computer provide the opponents with a corpus of hands that is consistent with the bidding so far.

Rather than trying to explain a MOSCITO 1D opening as (4+ Hearts, 9-14 HCPs, 6+ slam points unless you have 10+ cards in the two longest suits), generate 1K hands consistent with a MOSCITO 1D opening and use these for the other computer to draw inference.

[Free]

hrothgar, on 2014-September-12, 07:53, said:

There's more than one way to skin the cat.

You seem to be assuming that the the best way for computers to disclose information to one another is to use the same type of language that humans use to describe their agreements.

I'd think that it would be far easier all around to have the computer provide the opponents with a corpus of hands that is consistent with the bidding so far.

Rather than trying to explain a MOSCITO 1D opening as (4+ Hearts, 9-14 HCPs, 6+ slam points unless you have 10+ cards in the two longest suits), generate 1K hands consistent with a MOSCITO 1D opening and use these for the other computer to draw inference.

Interesting view which has a lot of potential imo.

I think communication should go both ways so another computer can ask if a certain set of hands would also open 1♦. This way the other program can check if its analysis of the first 1K hands is correct. This way it can get an even more accurate picture of what its opponent is doing.

Also, in MOSCITO with a 4M-6m hand you can open 2 ways, so I guess just giving hands for a 1♦ opening might not be enough. Or maybe using the 2-way communication, the original program should state what action it would take with hands that don't fit the requirements.

[DinDIP]

helene_t, on 2014-September-12, 03:21, said:

Just noticed that the system restrictions are stricter than GCC. You can't play strong club for example. Apparently Jack played some SEF like system in 2013 but it used to play Moscito against other computers. Moscito is probably the system that is best implemented in Jack. Possibly the Jack team couldn't be bothered to update the World Championship compliant system when upgrading to version 6? Just speculating.

I haven't checked carefully every year but the boards I have seen indicate that Jack used a variation of Dutch STD in all WCs. In Jack 5 the various versions are listed as Jack, Jack Washington and Jack Verona. Haven't yet upgraded to Jack 6 so not sure if there are newer versions of its preferred system -- which would look very familiar to any human opponents: 5M, 15-17 NT, Walsh, Bergen, Flannery etc.

My understanding is that Moscito and other non-STD systems were banned after GIB's success using a variation of Moscito in 2001. One of the reasons was the additional coding required by other competitors; another was the testy personal relationship between Matt Ginsberg (developer of GIB) and Al Levy (the coordinator of the computer competitions for first the ACBL and then internationally) and between Matt Ginsberg and other programmers.

For some ideas on how humans could use computers to get better disclosure from opponents read Peter Winkler's entertaining and thought-provoking book "Bridge at the Enigma Club".

David

[patroclo]

On the web site of wbridge5. seems that the reason for the absence of jack is the high cost of the transfer in China