[harleyf]

There's lots of threads talking about robot bidding here, but I have a question about robot play/defense. I'm working on my skills playing the daily robot tournament, getting my atrophied skills back into shape.

Ran across this hand this week:

---

This was a Just Declare, so the bidding isn't mine, hand was given to me and tried to deal with the mess.

The lead is the HK from west. Looking at my hands, I can see I can discard basically anything. I discard the D8 from dummy and the H2 from my hand. Opps take 5 hearts, down 4. That sucked, but at least it will be flat I say.

Nope. Score comes back, 6NT-4 = 20%. Played 100+ of times. WTF?

Reviewing the hand, very interesting. It's scored -12, -4, -3, -1, and makes 6. Down -12 is easy -- a few people saw this mess and just conceded all the tricks. Down 4 is me.

Down 3 -- discard a high heart at trick 1. This results in the robots not managing the suit well and end up taking the fourth heart with the heart Ace, never getting the last heart.

Down 1 -- discard the S9, followed by either another spade or a club. Robots switch to diamonds.

Makes - discard the club 3. Somehow, with this incredibly clever move, the robots switch suits at trick 2.

So, my question is whether any of this is predictable. Is there some logic I'm missing here? I realize they are just robots, and perhaps they just do dumb things.

[steve2005]

Robot defence would be much better if there was some signalling going on here.
Maybe 4♥ by East shows count but it is attitude that is needed.
Q♥ would be a shout for -4.
After which West should go Ace and another heart.

[smerriman]

 harleyf, on 2021-January-12, 19:58, said:

So, my question is whether any of this is predictable. Is there some logic I'm missing here? I realize they are just robots, and perhaps they just do dumb things.

This even surprises me, but the answer to your question is yes, this is predictable knowing how the robots work.

In this auction, South has promised exactly 4432 distribution with 12-14 high card points (with this strength, it rebids hearts with 5, and raises a minor with support). So there are extremely few possible hand layouts, meaning complete analysis is quite easy.

GIB West will simulate possible hands for South, and then pick the card which gives the best result on average. It will not make bridge-related inferences based on what cards you decide to discard (ie it won't assume you didn't throw away a trick).

Let's solely compare two lines of play - one where you discard a diamond on trick 1, and one where you discard a club.

In several cases, what you discarded makes no difference. For example, when it considers the case where the only points East holds is the Queen of spades, double dummy will tell it to lead a heart in both cases.

Or, if East holds the Ace of spades, double dummy will say it's completely irrelevant what it plays at trick 2 - it can switch or continue with hearts freely.

However, suppose East holds the King of diamonds - GIB will consider a layout like this:

---

If you discard a diamond, it will consider the king of hearts equally as good to play as any other suit - down 1. However, if you discard a club, it MUST play a diamond at trick 2 for down 2.

Or this layout, where East holds some quacks:

---

If you discard a diamond, double dummy will say 'play anything but the ace of hearts (even low heart is OK)'. If you discard a club, double dummy will say 'don't touch hearts at all'.

I've gone through every possible combination of the 2-4 HCP East can have. While many tell it to continue in hearts, the club discard increases the number of outcomes which tell it to switch.

If the hands which favor a switch come up more times than not, it will switch. This relies on the randomness of simulations; while the heart continuation is still more *likely* to come on out top, discarding a club will improve the odds in your favor.

Is this anything to do with real bridge or real bridge logic? Of course not; real bridge has signalling, and defenders not assuming declarer is an idiot and throwing away an extra trick for no reason. But it's GIB.

[johnu]

 smerriman, on 2021-January-12, 23:24, said:

This even surprises me, but the answer to your question is yes, this is predictable knowing how the robots work.

There is usually logic involved in GIB's catastrophic incomprehensible decisions. That logic is usually further defined as "Bad" logic, or "Broken" logic.

[smerriman]

 johnu, on 2021-January-13, 01:52, said:

There is usually logic involved in GIB's catastrophic incomprehensible decisions. That logic is usually further defined as "Bad" logic, or "Broken" logic.

That's why it surprised me; normally it is incomprehensible, while here it was entirely logical (and replicable / exploitable), even if that logic has a (pretty hard to correct) flaw.

[Evies Dad]

Thanks smerriman for that explanation.

So it is not broken logic at all. Gib is playing for the best long term MP score.

I always feel Gib gets a lot of unwarranted criticism in the play. Quite often it's defence can be brutal against you.

As humans we tend to notice when it goes badly wrong and don't make a fuss, or even notice at all where a switch gives it an extra trick.

When you can see you are going down the advice would be to try and trick Gib, much as you might try to trick human opps into a mistake.

Two side points.
At IMPs you would cash the AK and not risk the slam making.
In a typical MP human field there will always be people not bidding the slam. So that skews the odds towards safely setting the contract as well.

[johnu]

 steve2005, on 2021-January-12, 21:00, said:

Robot defence would be much better if there was some signalling going on here.
Maybe 4♥ by East shows count but it is attitude that is needed.
Q♥ would be a shout for -4.
After which West should go Ace and another heart.

There is signaling going on. Unfortunately, it is random signaling. The good thing is that you can count on partner not revoking.

[harleyf]

Thanks very much for the reply.

The thought of risking letting the opps make the slam / -2 vs a guaranteed -1, I guess I'd always just set the contract. Still too new at this to worry about IMP vs MP differences. But I see why this might be in the long run.

I'm not good enough to try to imagine hands yet - I'm still concenrtating on what I've got and how to play it. I'm going to have to stare at this and see if there's some general principle I can glean from it

[smerriman]

Just in case I wasn't clear from my initial post, a heart continuation is 100% the correct play in all forms of the game.

GIB is clearly making a horribly *wrong* play. But there is a reason why a program which has zero bridge sense and can only analyse double dummy will make mistakes; and I was just pointing out there why discarding a club will make such a robot more likely to make that mistake (even if it would only work rarely).

Any human player would make the correct play of a heart regardless of what you discard.