Board 7
Both sides vulnerable
| ♠ A Q 4 ♥ 9 6 3 ♦ 10 8 6 3 ♣ A Q 5 |
I open with one diamond in first seat. Partner bids one heart and I rebid one notrump. Partner raises to two. Even though I have only 12 HCP, I do have three honor tricks, which is a maximum by some standards. Still, 4-3-3-3 is a minus, and I have no fitting card in partner's suit. So I suppose I should decline the invitation. I pass, and LHO leads the spade eight.
| NORTH Robot ♠ 9 5 3 ♥ K Q J 2 ♦ A 5 ♣ J 10 9 8 |
||
| SOUTH Phillip ♠ A Q 4 ♥ 9 6 3 ♦ 10 8 6 3 ♣ A Q 5 |
| West | North | East | South |
| Robot | Robot | Robot | Phillip |
| 1 ♦ | |||
| Pass | 1 ♥ | Pass | 1 NT |
| Pass | 2 NT | (All pass) |
I can take three clubs, two spades, two hearts, and a diamond. Eight tricks unless the opponents can take six first. The spade eight looks like top of a doubleton. So I must duck the first spade to maximize the chance that I can keep the opponents from establishing and running the suit. The diamond suit does represent a danger. But that's the suit I opened, so it may be hard for the opponents to attack that suit.
I play low from dummy, East plays the king, and I duck. East continues with the six of spades. I win with the queen (the card West knows I hold) and West drops the jack. The jack? Apparently West led low from jack-ten-eight rather than top of his sequence.
I play the six of hearts, concealing the three, to dummy's king. West plays the four; East, the seven. If the club king is onside, I've made this. I can take three clubs and establish another heart for my eighth trick. What if the club king is offside?
Let's say I ride the jack of clubs and it loses to the king. West switches to a diamond. I duck one round and win the second round. Since clubs are blocked, I can cash only two club tricks, reaching this position:
| NORTH Robot ♠ 9 ♥ Q J 2 ♦ -- ♣ 8 |
||
| SOUTH Phillip ♠ A ♥ 9 3 ♦ 10 8 ♣ -- |
I need three of the last five tricks. I lead a heart. The opponents win the heart ace, cash whatever diamonds they can, and I have the rest. If they can take only one more diamond trick, I've made my contract. If they can take two, I'm down one. If they can take three, I'm down two.
If I don't block the clubs, I can cash three club tricks, ensuring I can't go down more than one. Maybe floating the jack of clubs is a bad idea.
What happens if I lead a club to the queen? If the king is onside, it makes no difference. I can repeat the finesse later. If the queen loses to the king, I'm in good shape. The clubs are no longer blocked. I either make or go down one depending on what happens in diamonds.
The problem comes if the king is offside and they duck. If I assume the king is onside, then I might find myself in some precarious predicament later on. But I don't think my robot opponent is up to ducking. If the club queen holds, I think I can safely assume East has the king.
I wouldn't do this against opponents I trust. But I decide to play a club to the queen. East plays the deuce; West, the king. See? I told you he wouldn't duck.
West shifts to the four of diamonds. I duck in dummy. East takes the king, and I play the six to disguise whether West's four is low or high. East continues with the nine of diamonds--three--seven--ace. It appears West had QJ74 and East had K92.
If so I'm down. The defense has taken three tricks and can take three more when they get in. Maybe West started with five diamonds, QJ742, and East has the heart ace. That seems unlikely. Why would the defense set up the suit in the hand that has no entry? East probably has the long spade. If he had the heart ace and only one more diamond, he would have shifted back to spades after winning the diamond king.
Still, it doesn't hurt to try. If that's the layout, I don't need to knock out the heart ace early. I can cash all my tricks, reaching this position:
| NORTH Robot ♠ -- ♥ Q J 2 ♦ -- ♣ -- |
||
| SOUTH Phillip ♠ -- ♥ 9 ♦ 10 8 ♣ -- |
I need one more trick. If East has the heart ace and is out of diamonds, I can play a heart to the queen. East can win and cash a long spade, but he will have to concede a heart trick to dummy at the end.
I play a club to the ace and a club back to dummy. Both opponents follow. I cash the club ten, and East pitches the spade seven. That confirms my suspicion that the heart ace is on my left. East wouldn't be pitching his spade winner if he held the heart ace.
The ten-eight of diamonds may prove useful if I've misread the position. But my hearts are worthless. So I pitch the three of hearts. West pitches the heart eight.
I play a spade to my ace, reaching the above position. Both opponents follow. I exit with a heart. To my surprise, it's East, not West, who wins the trick. East then cashes the diamond jack and concedes a heart to dummy. Making two.
| NORTH Robot ♠ 9 5 3 ♥ K Q J 2 ♦ A 5 ♣ J 10 9 8 |
||
| WEST Robot ♠ J 10 8 ♥ 10 8 4 ♦ Q 7 4 2 ♣ K 7 6 |
EAST Robot ♠ K 7 6 2 ♥ A 7 5 ♦ K J 9 ♣ 4 3 2 |
|
| SOUTH Phillip ♠ A Q 4 ♥ 9 6 3 ♦ 10 8 6 3 ♣ A Q 5 |
Plus 120 is worth 75%. The median result was passing it out. I briefly considered accepting partner's game try, and some thought this hand wasn't worth opening?
West played four, then seven from Q742 and the diamonds were blocked all along. It turns out I gave the defense a chance to beat me. Here was the position when I cashed the last club:
| NORTH Robot ♠ 9 ♥ Q J 2 ♦ -- ♣ 10 |
||
| WEST Robot ♠ 10 ♥ 10 8 ♦ Q 2 ♣ -- |
EAST Robot ♠ 7 2 ♥ A 5 ♦ J ♣ -- |
|
| SOUTH Phillip ♠ A ♥ 9 3 ♦ 10 8 ♣ -- |
If I knock out the heart ace at this point, I take eight tricks thanks to the diamond blockage. When, instead, I cashed the last club, all East had to do to beat me was to pitch a loser instead of a winner. If he pitches a heart, my hand is squeezed. To hold my diamond stopper, I have to let go of a heart and now have no way to cash the heart once I establish it.
I didn't visualize the diamond blockage. I was mesmerized by the defense's carding and assumed West had the queen-jack or queen-jack-deuce of diamonds left. I should have gotten the end position right. I lucked out that East defended poorly.
This deal is a good example of how humans and robots make different kinds of mistakes. My mistake was making an assumption about the layout and forgetting that my assumption might be wrong. That's a very human mistake and a difficult one to avoid.
Most of what we think about when we play a deal is reconstructing the opponents' hands from clues. Sometimes our deductions are virtually certain; sometimes they are more speculative. It's easy to forget which ones are speculative and back off those assumptions when necessary. But robots don't have that problem. They don't play detective in the first place. So a robot would not have a blind spot in the above position and would easily find the heart play.
East's mistake was assuming I knew what I was doing. That's a very robot mistake. He could see that, if I played correctly, it made no difference whether he pitched a spade or a heart, so he chose one at random. Giving your opponent a losing option when he has a choice is something a robot doesn't even "think" about. A human, on the other hand, would see that the long spade might come in handy but the low heart could not and would be unlikely to make this mistake.
It doesn't seem it should be that difficult to program a robot to take into account that humans might make a mistake.
ReplyDeleteI think it's actually quite difficult, given the decision made early on to use a double-dummy analyzer as the basic building block. You could perhaps tack on a subroutine to call whenever the robot thinks its play doesn't matter. The subroutine could then look for a play that gives the opponent the most losing options. But simply counting the losing options isn't optimal, since not all losing options are equally attractive. And the computer has no basis for deciding which ones are attractive and which ones aren't. The problem is that robots, by design, have no single-dummy logic to rely on. To take into account the possibility of opponents' mistakes properly would require redesigning the architecture from scratch.
DeleteThe BBO robots do, by the way, appear to have some algorithmic methods for catering to mistakes. For example, they tend to discard high cards rather than low cards when it doesn't matter. I think the programmers hope that you will believe your five of clubs is good when you see the opponents throw all their high clubs away, not noticing that they've carefully held onto the six. But that's simply following a pre-programmed rule. Working out what mistake an opponent might make is a different matter.
DeleteWhat is the difference between the top bridge programs (like Jack) and the BBO robots? Is it just the depth of the trees they use? Do they all use a double dummy analyzer as the "building block"?
ReplyDeleteI'm not sure, but I suspect the difference is in how many random deals they generate and how well they exclude "impossible" layouts. The general approach is the same: Generate random deals consistent with what you know and select the play that works most often on a double-dummy basis. The assumption that both sides are double-dummy is a critical feature and seems to work much better than one would expect, given the absurdity of the assumption. The deal that made me fully appreciate how the algorithm worked was one where I saw the robot go down in 7NT with 13 top tricks. This isn't the exact deal, but it's close enough to illustrate the point. Dummy had (void) AKQJ10xx K109 xxx. Declarer had Axxxx xx AJx AKQ. 13 top tricks. I led a spade. Note that declarer can discard any card from dummy except the diamond king and still make the contract double-dummy. It, presumably at random, chose a club. Now it had to guess the diamond queen. It misguessed and went down. It's easy to see how to add tweaks to the algorithm to avoid this. But if the methodology is flawed, adding tweaks won't help much. The same flaw will manifest itself in another way. What we really want to do is fix the flaw at the design level, and I have no idea how to do that.
Delete