GIB ruffs partner's winner

[lakshmanok]

Watch as GIB ruffs partner's winner (trick 11). GIB is playing *after* partner, so there is no excuse:

handviewer link

[semeai]

GIB was always getting one of the last two. The play at trick 11 (ruff or no ruff) didn't matter.

 

I'm sure GIB knew/simulated/whatever this and just chose one of the two cards that works (i.e. anything but the top trump) to play at trick 11.

[lakshmanok]

True ... I missed that robot's trump spot was lower than North's.

[Stephen Tu]

 

[beaten to the punch by semeai]

[Bbradley62]

Of course, there have been hands posted where GIB simulated that "it doesn't matter" when actually it did matter and GIB's sims just didn't cover all the possibilities. It would be nice if GIB would make the "normal" play when "it doesn't matter" to reduce his chance of occassionally getting screwed.

[Stephen Tu]

Of course, there have been hands posted where GIB simulated that "it doesn't matter" when actually it did matter and GIB's sims just didn't cover all the possibilities. It would be nice if GIB would make the "normal" play when "it doesn't matter" to reduce his chance of occassionally getting screwed.

 

As has been pointed out before, GIB has no idea what a "normal play" is. It's basically looking at all the plays and picking the best one on the sample deal set (best DD play if basic bot, best single-dummy if advanced), the size of the set being constrained by thinking time and config parameters. If the sample deal set is too small (doesn't capture bad break or other special cases to cater to) or not representative (bidding DB flaw), it will draw the wrong conclusion about what is "best". If it doesn't matter, GIB has no idea that ruffing partner's winner is less or more normal than discarding. It thinks both plays are equivalent, so it just picks one at random. There are no rules that say this is "normal" or not, the only rule is maximize your score expectation given what you know, searching among the set of legal plays.

 

Asking for GIB to make "normal plays" is basically asking for it to use human-rule type thinking, which GIB and other programs have shown to be inferior to the brute force search approach. It's simply too hard to write down rules of how to play, there are so many of them, often need to be broken, often need to be re-ordered as to what task to attack first. Think about how many things an advanced player looks at to decide when to start drawing trumps and how many of them to draw. Think how massively more involved that decision process is vs. that of a raw beginner who would often simply start drawing trump as soon as getting the lead and draw all of them. Multiply that by everything else you have to do on a bridge hand, and think about how you would write down a set of rules to follow so you could pull someone off the street who'd never played a hand of bridge in his life, give him that rule set, and then expect him to blindly follow that rule set and declare well.

 

Instead of saying it would be nice if GIB makes "normal plays", simply clamor for increased thinking time especially on the basic bots. Maybe some intermediate cost option between basic & advanced. Or ask for bidding upgrade patches for the older standalone BBO client so that we can use the thinking time on our own PCs instead of relying on the server farm.