[jdonn]

Gib has K82 J63 QJT A852, 4th seat, both vul.

1♦ 2♣ P 2♦ (good!)
DBL P P 2♥ (huh?)

2♥ alerted as "3+C; 4+H; 11+ total points". Gib had neither 5 hearts, nor 4 hearts, nor any heart strength at all. It even had a club more than it had promised and a diamond stopper so could hardly have been stuck for a bid. Is it possible that gib simmed bidding a bad suit as working best because the defense would be off?

[arigreen]

GIB doesn't take defense into consideration when it simulates during the bidding. For various reasons, GIB simulated among 2♥, 2♠, and 3♣. In many of the simulated deals, all three actions led to the same contract (3♣). In some of the deals 2♥ precluded the opponents from competing further, allowing partner-GIB to declare 3♣ when LHO would have competed over an immediate 3♣. Not so surprisingly, GIB predicted that LHO was more likely to pass 2♥ (but compete over 3♣) when it held 4 hearts.

So, in a sense, GIB predicted that 2♥ would be a harmless psyche.

Why wasn't GIB concerned that partner might have passed or raised hearts? Because with 4 hearts, GIB assumed that partner would have bid 2♥ over the double instead of passing. Furthermore, GIB assumed that with a known fit in clubs, partner-GIB would always retreat to 3♣ rather than pass 2♥.

The best way for us to fix this sort of thing is to add more specific rules for what to do after a doubled cue-bid is passed back to GIB. As you can see, relying on simulations in these sorts of auctions can be problematic.

[pretender]

In many overcall auctions GIB shows a clear reluctance to bid NT when it has the stopper in the opponents' suit. I can think of many other hands, but in this given hand, GIB simulated from among 2H/S/3C but not 2NT? This makes it very hard to get to NT games in overcall situations or even to just rightside them to GIB if it won't bid them.

[jdonn]

So I had the right idea, but he decided it would hurt them in the bidding rather than the defense. Interesting.

[allfail]

What it meant is that GiB can psyche and it is fine, why do we have to fix this?

[jdonn]

allfail, on Mar 6 2010, 11:57 PM, said:

What it meant is that GiB can psyche and it is fine, why do we have to fix this?

Who says we do? But until we were told by the programmer that it was essentially a psych it could have been a bug instead for all we knew. Also it's nice to know if we are the overcaller here that gib expects us to bid a side major 100% of the time over the double.

[arigreen]

Mind you, GIB didn't know that it was psyching, just as GIB doesn't know when it's about to execute a strip squeeze -- it just takes the action that it thinks will work best. We do consider this 2♥ bid to be a bug, because GIB had a perfectly normal alternative that was not chosen.

[allfail]

Yeah I understand your point. It sort of depends on your goals though. As far as I can conceive there are many situations where GiB would have a bid which does not match its descriptions but works nevertheless. Sometimes there are no plausible choices sometimes there are other perfect bids. For me, it doesn't matter whether gib conscientiously psyches or it just thinks some bid works best, as long as the simulation was performed correctly (in my current understanding it is a big if however). It should be a strength of AI since we humans can't do such things in a short time and if the simulation is correct it should lead to a better result.

[Bbradley62]

It's my understanding that psyches are defined as random bids that mislead both the opponents and partner. If a partnership frequently makes the same psyche-like bid, that bid ceases to be a psyche and becomes part of their partnership understandings and must be explained as such. Therefore, if GIB will routinely make this bid on 3♥, it's not a psyche -- it's a part of GIB's bidding system and the explanation of the bid should be 4+♣ and 3+♥.

[jjbrr]

Bbradley62, on Mar 8 2010, 01:14 AM, said:

It's my understanding that psyches are defined as random bids that mislead both the opponents and partner. If a partnership frequently makes the same psyche-like bid, that bid ceases to be a psyche and becomes part of their partnership understandings and must be explained as such. Therefore, if GIB will routinely make this bid on 3♥, it's not a psyche -- it's a part of GIB's bidding system and the explanation of the bid should be 4+♣ and 3+♥.

I don't agree with your description of psyches as "random". I suppose they can be random, but in general I think they're thought out pretty carefully in practice.

[arigreen]

Bbradley62, on Mar 7 2010, 11:14 PM, said:

It's my understanding that psyches are defined as random bids that mislead both the opponents and partner.  If a partnership frequently makes the same psyche-like bid, that bid ceases to be a psyche and becomes part of their partnership understandings and must be explained as such.  Therefore, if GIB will routinely make this bid on 3♥, it's not a psyche -- it's a part of GIB's bidding system and the explanation of the bid should be 4+♣ and 3+♥.

According to GIB's system, 2♥ promises 4 hearts. All of us that play with GIB regularly know that GIB sometimes deviates from its system, but partner-GIB won't suspect that. In fact, partner-GIB would always prefer hearts to clubs here if it had 4 hearts. But since partner-GIB has implicitly denied 4 hearts by failing to bid 2♥ itself, this will never happen.

In any case, we consider this a bug for several reasons:

- 2♥ in this auction should probably show a 5-card suit.
- Bidding 2♥ on a hand like this confuses our users. A human partner won't always rebid 3♣ as GIB predicts it will, so the final contract is likely to be poor.
- 3♣ would have been a perfect description of GIB's hand.