[jardaholy]

For some reason, the robot may not follow the simple agreements about defensive signalling, which are described in the BBO manual.

---

The first lead was ♣10, which I had to overtake with ♣A, I returned ♦3 taken by ♦K. One more round of diamonds was played and the declarer than tool ♥ finesse, my partner playing ♥ eight as the second card in the trick.
This time, there was actually no impact on the results, which was 2NT just made and everybody, who arrived in this contract, got the same. Still, I was absolutely sure during playing the board that the declarer possessed just singleton heart. My experience is that the bots provide count pretty fair, just taking care of not to lose trick by spending too high card for that. It is a completely mystery for me, why it did that. Are there some other signals used by the bots, which are not described in the manual?

[smerriman]

The manual is completely wrong. You should never assume anything about the cards that GIB plays.

Even the part about signals when following to an opening lead are often incorrect. Partially correct enough that there was probably some code somewhere that was intended to do it originally, but provably buggy (especially the part I found about it playing at random from a doubleton, but even the other instances which were guaranteed to not have been simulation based).

Perhaps there was similar code that handled count signals originally, but the randomness and bugginess of GIB kills any potential advantage you could get by reading into it.

[pilowsky]

The manual specifically mentions two excellent books by Bird and Anthias. One dealing with suit contracts and the other with no trump contracts.
I bought both, and my understanding of GIB signalling and leads improved tremendously. I recommend them highly.
This is a supplier for the suit and the NT contract books.

https://www.amazon.c...d/dp/1554947693
https://www.amazon.c...&s=books&sr=1-2

You can read a sample of what's in the books here:
http://youth.worldbr...nd-taf-anthias/
You may be able to get them cheaper elsewhere.

The key point is that, like Chess computers, Bridge computers don't follow 'rules' the way humans do. They 'simulate' (think thinking) thousands of hands based on the auction and played cards.
This gives them a list of probabilities for the best card to play next. It's the same with chess computers. They then play the card that comes out top of the list.

Chess computers do much the same thing. Instead of thinking "I'll launch an attack on the Kingside" they run down all the possible variations (within the limits of their computing power and available time) and pick the best one.
Since Chess is played 'double-dummy' fast computers now have the edge over humans. Bridge computers still have to rely on probability.

If you are dissatisfied with the robots explanations, you can try calling the Director - but it never does me any good.
So I occasionally send a mean chat message to my GIB partner, but he never responds. It could be my Australian accent.

[pescetom]

smerriman, on 2021-February-07, 17:28, said:

The manual is completely wrong. You should never assume anything about the cards that GIB plays.

Even the part about signals when following to an opening lead are often incorrect. Partially correct enough that there was probably some code somewhere that was intended to do it originally, but provably buggy (especially the part I found about it playing at random from a doubleton, but even the other instances which were guaranteed to not have been simulation based).

Perhaps there was similar code that handled count signals originally, but the randomness and bugginess of GIB kills any potential advantage you could get by reading into it.

It's depressing that in 4 years BBO has done nothing to correct even the manual. Just quoting you would help.

[jardaholy]

smerriman, on 2021-February-07, 17:28, said:

The manual is completely wrong. You should never assume anything about the cards that GIB plays.

Even the part about signals when following to an opening lead are often incorrect. Partially correct enough that there was probably some code somewhere that was intended to do it originally, but provably buggy (especially the part I found about it playing at random from a doubleton, but even the other instances which were guaranteed to not have been simulation based).

Perhaps there was similar code that handled count signals originally, but the randomness and bugginess of GIB kills any potential advantage you could get by reading into it.

Thank you for the comment.I play daylongs daily since Covid-19 started and right now, I have got around 11 000 boards played. So, I have nice statistical data to play with (I was graduated from probability theory and mathematical statistics). I was a bit surprised with the case, I put here, because it seems to me that the bots, in fact, really follow two basic rules: 1) they signal HIGH card to the first lead with top honeur in hand 2)they really provide count later (small = odd), provided that there is no way, it could cost the trick. So, there are exceptions...

[jardaholy]

pilowsky, on 2021-February-08, 03:13, said:

The manual specifically mentions two excellent books by Bird and Anthias. One dealing with suit contracts and the other with no trump contracts.
I bought both, and my understanding of GIB signalling and leads improved tremendously. I recommend them highly.
This is a supplier for the suit and the NT contract books.

https://www.amazon.c...d/dp/1554947693
https://www.amazon.c...&s=books&sr=1-2

You can read a sample of what's in the books here:
http://youth.worldbr...nd-taf-anthias/
You may be able to get them cheaper elsewhere.

The key point is that, like Chess computers, Bridge computers don't follow 'rules' the way humans do. They 'simulate' (think thinking) thousands of hands based on the auction and played cards.
This gives them a list of probabilities for the best card to play next. It's the same with chess computers. They then play the card that comes out top of the list.

Chess computers do much the same thing. Instead of thinking "I'll launch an attack on the Kingside" they run down all the possible variations (within the limits of their computing power and available time) and pick the best one.
Since Chess is played 'double-dummy' fast computers now have the edge over humans. Bridge computers still have to rely on probability.

If you are dissatisfied with the robots explanations, you can try calling the Director - but it never does me any good.
So I occasionally send a mean chat message to my GIB partner, but he never responds. It could be my Australian accent.

Thank you for the comment. Since I was graduated from probability theory, I completely know, what you are talking about. My idea was (and perhaps still is) that the basic method, the bots use, is really Monte Carlo simulation (which can be very strong tool, as it is in Chess!), but they got implanted some
first priority extra rules (use pure Monte Carlo simulation for the first leads, for example, is nonsense, because no human player would play with them). I still think that some rules are implanted, but they are hidden (there may be also difference between simple and advanced bots). If this is the case, it is bad, because to know the rules could help to limit the randomness of results of daylong tournaments and to let better players to win with higher probability.

Thank you also for the references, I am thinking about purchasing them.

[pilowsky]

jardaholy, on 2021-February-08, 16:47, said:

Thank you for the comment. Since I was graduated from probability theory, I completely know, what you are talking about. My idea was (and perhaps still is) that the basic method, the bots use, is really Monte Carlo simulation (which can be very strong tool, as it is in Chess!), but they got implanted some
first priority extra rules (use pure Monte Carlo simulation for the first leads, for example, is nonsense, because no human player would play with them). I still think that some rules are implanted, but they are hidden (there may be also difference between simple and advanced bots). If this is the case, it is bad, because to know the rules could help to limit the randomness of results of daylong tournaments and to let better players to win with higher probability.

Thank you also for the references, I am thinking about purchasing them.

No worries.
Here's an example I just played where GIB gives me extra tricks by not unblocking. http://bit.ly/UnBlockGIB
Who knows what they were 'simulating' I hope that one never gets fixed.

[pilowsky]

Another simulation failure: After 1♣ - pass - 1♥ - pass - 3NT - and GIB leads a heart.
This is because - according to Bird and Anthias - this is a situation where GIB will prefer a passive lead in your bid suit.
The simulation suggests to GIB that the opener doesn't have Hearts.
I'm not sure if this is an accurate interpretation of what's happening but I noticed some of the more successful players using this sequence against GIB.
FWIW the good players at the CLub tended to make passive leads in NT contracts all the time as well.
I was relying on my KT ♥ as a stopper - then this classic GIB play occurred.
http://bit.ly/PassiveNTlead

---