[barmar]

Now that the Flash client allows playing with robots in the main club, I've gotten to see GIB declaring lots more than I did in the Best Hand Robot Rewards tourneys. Yesterday I saw something I couldn't believe. Here's a link to the hand:

http://tinyurl.com/dz8eaj

We were in a suit contract, and the opening lead was a side Ace. With JT in hand opposite K9x in dummy, GIB played the King! It's been over a day and I'm still trying to imagine how its simulations told it that this would gain a trick. Maybe a desperate expert declarer would do this to make the opponent think he had the QJTx and was unblocking. That wasn't going to fool this opponent, since it led from AQxx.

I think the contract was makable until it threw away this trick.

[jdonn]

He does it quite often. I could be wrong but I believe the sims don't incorporate logic, except for regarding the bidding, plays made on prior tricks, or proven things such as showing out of suits. In other words, he doesn't consider that he wouldn't underlead the ace there, so if his sims randomly show the ace to be onside more than the queen and he only needs one trick, the king it is.

Of course with K9x and JTx he would probably play low since it could be AQx onside.

[barmar]

I think you misunderstood. The defender didn't underlead the Ace, the defender led the Ace, and GIB played the King under it. It threw away a guaranteed trick (unless West has a singleton and will ruff the next trick), ensuring two losers in the suit even though the Ace had already been seen.

[arigreen]

Thanks barmar for for reporting this. I've seen this type of play occasionally but it's quite rare. I don't understand it yet either. Certainly we can construct hands where it's break-even to play the K under the A but I can't imagine how that play can gain. I will investigate this hand further.

In case of a low spade lead, Josh is right -- the GIB declarer makes no assumption that the A must be offside because LHO would have led it if he had had it. Still, most of the time GIB will play low.

[hotShot]

If the conditions under which dumping the K makes sense are rarely fulfilled, these hands should not have a significant weight after the simulations.

Since GIB operates under a time limit, complex deals (=> in the sense of time consuming simulations) might lead to very few simulations, giving an extreme weight to the simulated hands.

This is the likely cause of such odd behavior, and usually if you feed GIB with the same deal again, you will get a different result because you will simulate different random deals.


The other possible explanation would be, that GIBs simulations are not random enough, so that the simulated deals are to similar to each other.

[barmar]

It seems like the robots I'm playing with at regular tables play much faster than when I play in Robot Rewards tourneys, and they don't seem to play as well. Are the GIB settings different?

The simulation basis of GIB leads to some interesting results when you look at the comparisons. I played a hand yesterday where North opened 2♣, but at several other tables it opened 1♦ and reversed (I think it was something like 20ish HCP with 0454 shape). It's hard to get used to such non-determinism in a computer player.

[arigreen]

Yes barmar, you are correct. The flash client regular-table robots do play faster, and less well, than the tournament robots you are used to. One advantage of this is that we can support more robots playing at any one time. We are hopeful that such bone-headed plays as the one that spawned this thread will remain very rare, but if it they turn out to be common then we will have to increase the skill level (and decrease the speed) of these robots.

[barmar]

Here's another one.

http://tinyurl.com/caxgdv

At trick 5 I led ♠Q, declarer covered with the K, and North ducked with ATx! The only way that doesn't give up a trick is if I led Q without holding the J. And since there was also a finesse available in ♦, declarer's J could go away on that if it needed it.

[arigreen]

Not to mention the opening lead, which looks like it was chosen by the select-a-card-at-random algorithm.

I am in Houston now for the nationals and will be busy for the next 10 days, but I will try to figure out what is going on here as I have time.

[kfay]

barmar, on Mar 12 2009, 09:44 PM, said:

Here's another one.

http://tinyurl.com/caxgdv

At trick 5 I led ♠Q, declarer covered with the K, and North ducked with ATx! The only way that doesn't give up a trick is if I led Q without holding the J. And since there was also a finesse available in ♦, declarer's J could go away on that if it needed it.

These are really funny. Keep them coming!

[jdonn]

kfay, on Mar 13 2009, 12:37 AM, said:

These are really funny.  Keep them coming! 

Ok, just about 30 seconds ago:

Scoring: IMP

1♦-(P)-2♣-(2♠)
3N-(P)-4N-(P)
6N-(P)-P-(P)

---

2♠ Q♠ A♠ 5♠
4♠ 2♣ 3♠ K♠

2♣? Maybe I should just be thankful he didn't drop the king of spades at trick 1...

[bb79]

I was trying partnership bidding. My partner and opps are robots. West is dealer. I am south holding
void-KJ7-AK643-A9753.
the bidding is pass- 1D- (3S ) - 4S -(X)- pass - (pass) - XX - all pass.
Rdbl is alerted as 16+ points and forcing, but p passed
Partner has A-AQ53-T9872-QT2.