[dalmov]

The other day, I played a number of boards with three robots. I must admit that the robots bid precisely, and once a contract is established they play perfectly to fulfill it.
But, there seem to be a little problem with their defense, especially with their opening lead. In the following, I will show some examples where the robots chose a wrong opening lead that resulted in a fatal loss on their part.
With the first board, LHO-robot chose the ♠ 8 for opening lead, top of no thing, meaning “leading through strength”. Thanks to this, I had time to cash A then K of ♦ to discard the losing ♣. Thereafter, I had time to finesse for the Q of ♥ to make 13 tricks for a top score! The mistake with the robots is that they are probably not programmed to lead the “unbid suit”, which is ♣ in this case, and then, everybody will have an average score only.

---

With the second board, LHO-robot chose the opening lead of the ♣ 10 to RHO-robot’s ♣ Ace, which then returned a ♥. Thanks to this, I had time to manage to make the doubled contract of 3-♦ for another top score! LHO-robot seems not to be programmed to lead 3 rounds of ♦ first, then shifted to a ♣ and after that just waited for winning two more tricks: a ♥ and a ♣, a set of 2 tricks for a penalty of 300 points.

---

[Tramticket]

I might have made either off these leads. But I definitely wouldn't have bid the hands like that:

- Board 6: Did you really overcall at the three level on that ropey five-card jack-high suit? It was asking to be doubled and you were lucky to find partner with three-card support and some useful values.
- Board 7: North has a massive hand after South reverses and 4♥ doesn't begin to do justice to the hand. The jump to 6♥ also looks a bit premature.

[Stephen Tu]

The bots do have big problems on the defense, but these are mainly:

• no understanding of defensive signalling whatsoever
• not giving enough slack for a human bidder to have deviated from system being weaker/stronger than expected range or have unexpected shape.

The way the bots work generally is to deal out a bunch of random hands that match what it thinks the auction shows and the known previous cards played, find the best card(s) double dummy, then play the best card on average or choose randomly among tied winners. This generally works kind of OK, but the main flaws above kick in when:

• signalling would make one continuation much clearly better than another (humans understand this, and use signals to vastly constrain the possible remaining layouts, because "partner wouldn't signal that way" if holding certain hands)
• the human bidder doesn't have anything resembling what it thinks the bidding shows, which is either because the human bidder psyched, deviated slightly, or the bidding database is buggy on the auction in question (unfortunately the bidding DB has a ton of bugs, especially on competitive auctions)

As for opening leads though, generally it is OK if it understands the auction. (It may not on the first one, reverse sequences are rare with north/south jumping subsequently). Opening leads are a really inexact science, and margins are often pretty small. Humans use rules of thumb like "lead partner's suit" or "lead the unbid suit" or "lead from longest & strongest vs NT", but in reality these turn out very frequently the wrong thing to do. In theory the bot method, if it is modeling the hands for the auction correctly (frequently not because of the database bugs/holes), is going to be far more accurate than human rules of thumb on average. If we forced it to do human-like things such as always lead the unbid suit or always lead partner's suit, almost certainly it will do worse overall than its current method of statistically analyzing for the best lead. We should not want that. Mainly we want bidding database to be fixed so that the sample it is generating is representative of real-life.

On the second one, though I think the double by the bot is overly aggressive, leading a club and heart return was fine in theory, it is still setting the contract at that point and no better than the diamond. The mistake came later in the defense. Note leading diamonds doesn't set it 2; south still gets 2 diamonds, 3 hearts, 2 spades and a club, for down 1.

[dalmov]

dalmov, on 2018-July-09, 02:39, said:

The other day, I played a number of boards with three robots. I must admit that the robots bid precisely, and once a contract is established they play perfectly to fulfill it.
But, there seem to be a little problem with their defense, especially with their opening lead. In the following, I will show some examples where the robots chose a wrong opening lead that resulted in a fatal loss on their part.
With the first board, LHO-robot chose the ♠ 8 for opening lead, top of no thing, meaning “leading through strength”. Thanks to this, I had time to cash A then K of ♦ to discard the losing ♣. Thereafter, I had time to finesse for the Q of ♥ to make 13 tricks for a top score! The mistake with the robots is that they are probably not programmed to lead the “unbid suit”, which is ♣ in this case, and then, everybody will have an average score only.

---

With the second board, LHO-robot chose the opening lead of the ♣ 10 to RHO-robot’s ♣ Ace, which then returned a ♥. Thanks to this, I had time to manage to make the doubled contract of 3-♦ for another top score! LHO-robot seems not to be programmed to lead 3 rounds of ♦ first, then shifted to a ♣ and after that just waited for winning two more tricks: a ♥ and a ♣, a set of 2 tricks for a penalty of 300 points.

---

Reply to Tramticket: Thank you for the comment. I will reply only to your comment on Board 6.
First, the reason for me to overcall the opening bid of 3♣ by that J-high 5-card ♦ suit is the following. In theory, the pre-emptive 3♣ bidder could not have more than 9 HCPs. I had 11 HCPs. Then, by law of average, I could expect my partner to hold 10 plus HCPs. This means that our side had the balance of powers. Therefore, I wanted to compete. Now, what should I bid? I could not bid "double" because of the stiff in ♠ although the ♥ was decent and I loved to bid it, if partner also had 4 cards to support. I had 3 ♣ cards, RHO had 7 plus cards, which meant that my partner was very short there. Then, by law of elimination, my only biddable suit was ♦.

[Stephen Tu]

dalmov, on 2018-July-12, 03:29, said:

Reply to Tramticket: Thank you for the comment. I will reply only to your comment on Board 6.
First, the reason for me to overcall the opening bid of 3♣ by that J-high 5-card ♦ suit is the following. In theory, the pre-emptive 3♣ bidder could not have more than 9 HCPs. I had 11 HCPs. Then, by law of average, I could expect my partner to hold 10 plus HCPs. This means that our side had the balance of powers. Therefore, I wanted to compete.

This logic is faulty. There are several issues with your analysis:

• although your side rates to have the balance of HCP, this is only by a small margin, and only on average. Some percentage of the time LHO is going to have more than average share, and you can be in considerable trouble.
• Just because your side has the balance of HCP, doesn't mean you always want to bid over them. Some reasonable percentage of the time, your side has enough to set them defensively (5 tricks), but not enough offense to make (8 tricks or fewer, not 9). In particular the CK is better for defense than offense.
• bidding over preempts is not only about winning partial battles. It is also strongly about bidding games, since you are contracting for 9 tricks already, which is pretty close to game and because bridge is in general a lot about finding games since the game bonus is so valuable. Partner will expect a stronger hand than this, and often force you into a game that cannot make. If conversely partner assumes you can be this weak and passes too conservatively, you miss games when overcalling with a stronger hand.
• 3d is committal to diamonds. It makes it hard to find hearts if partner has a decent hand with only 4 hearts, because his 3H bid would show 5+. Conversely if he can make a takeout double, you can bid hearts with some confidence. Diamonds tends to lead to 3nt (if partner had stronger hand, same shape, auction would tend to go 3d-3s-3nt), and this will be hard to make since your diamonds need so much help.
• partner is there to protect if you want to compete. If he has his share of the deck, and club shortness, he can make a takeout double. If he doesn't have a takeout double, holding club length (clubs could be 1336 around table), or having weaker than expected HCP, there's a very strong chance you *don't* want to compete.