[shyams]

So I am wasting time playing one of those Robot BBO tournaments when this happens.

---

The robots are "Advanced GIB" variety. After West's 2♠ bid, North bot's pass is out-of-tempo for a robot.

Seated South, I could reasonably infer that North has a trump stack and is running a simulation to re-evaluate its position. Technically, my double is based on UI but this was a random fun game so the outcome did not matter and I doubled anyways.

If this were some serious game (e.g. multi-table duplicate MPs, the kind BBO hosts for "Stars and Masters"), would South's double be deemed 'illegal' (i.e. subject to rollback or such)?

[jillybean]

Yes, the score may be adjusted and I would give you a PP. LOL

However, it’s a strange question as we are being asked to apply the LoDB to a game that isn’t bridge.

[barmar]

This is an interesting question, due to a technical issue.

Ostensibly you'd assume that everyone is on an even footing -- all players should receive the same UI, so no one has an advantage.

But to reduce load on the bot servers, BBO makes use of caching. If the board has already been played, and the same sequence of bids/plays have occurred, the robot's action at that point can be looked up quickly. So the second player in this situation shouldn't see the hesitation from the simulation.

This means that only the first player in this situation gets the UI, and has an advantage over the rest.

But there's also no TD in robot tournaments, so there's no way to get an adjustment.

BBO has some human tournaments that allow players to hire a robot partner. These have TDs that can make adjustments, but I'm not sure if any have done so based on UI from the bots.

You need to be careful about taking inferences from its hesitations. There are many places in its bidding database where simulations are required or prohibited (the latter is mostly for responding to asking bids like Stayman and Blackwood). So the reason for the hesitation may not be as obvious as you think.

The funny thing is that one of its regular hesitations is when you take a losing finesse. GIB always thinks before winning the trick. I suppose this is because occasionally it's right to duck the first time, and it needs to do a simulation to decide. But I don't notice this in most situations where it actually does duck.

[shyams]

Thank you for a thoughtful reply, barmar.

1. I was unaware that the servers use caching and consequently only the first player gets the UI freebie. Now, the deal occurred in an inconsequential tournament with nothing at stake (I was playing a free Total Points 25-min race). It is highly probable that I was the first to play the mentioned deal --- I pass a lot of normal-strength hands in order to avoid wasting time playing part-scores.

However, this feature/bug might be of interest for adjudicating BBO's prestigious monthly tournament for Stars.

2. There are inbuilt tempo adjustments during play that I'm sure you know about. These prevent human declarers from gaining an advantage and improving finesse odds. Perhaps something similar can be incorporated into complex bidding sequence (just a suggestion; I'm not sure if that's even feasible).

barmar, on 2025-April-30, 12:43, said:

The funny thing is that one of its regular hesitations is when you take a losing finesse. GIB always thinks before winning the trick. I suppose this is because occasionally it's right to duck the first time, and it needs to do a simulation to decide. But I don't notice this in most situations where it actually does duck.

3. I have noticed the hesitation when a bot wins after I finesse (i.e. I take a losing finesse). I can say from personal experience that it is impossible to notice when a bot ducks a losing finesse
... e.g. Dummy holds ♥KQ10x. I play low to dummy's ♥K and East plays low. To the human eye, the tempo is identical whether East holds ♥Axx OR holds ♥Jxx.
... In such situations, the weakness is that often (not always) West rises with the ♥A when I repeat the finesse (i.e. ♥x to ♥Q10x still unsure of ♥A or ♥J).
... Bots often fail to infer thatit's still a guess for me, the declarer. However, occasionally bots successfully duck a second time and defeat me when I play the ♥10.

[axman]

barmar, on 2025-April-30, 12:43, said:

This is an interesting question, due to a technical issue.

Ostensibly you'd assume that everyone is on an even footing -- all players should receive the same UI, so no one has an advantage.

But to reduce load on the bot servers, BBO makes use of caching. If the board has already been played, and the same sequence of bids/plays have occurred, the robot's action at that point can be looked up quickly. So the second player in this situation shouldn't see the hesitation from the simulation.

This means that only the first player in this situation gets the UI, and has an advantage over the rest.

But there's also no TD in robot tournaments, so there's no way to get an adjustment.

BBO has some human tournaments that allow players to hire a robot partner. These have TDs that can make adjustments, but I'm not sure if any have done so based on UI from the bots.

You need to be careful about taking inferences from its hesitations. There are many places in its bidding database where simulations are required or prohibited (the latter is mostly for responding to asking bids like Stayman and Blackwood). So the reason for the hesitation may not be as obvious as you think.

The funny thing is that one of its regular hesitations is when you take a losing finesse. GIB always thinks before winning the trick. I suppose this is because occasionally it's right to duck the first time, and it needs to do a simulation to decide. But I don't notice this in most situations where it actually does duck.

>> This means that only the first player in this situation gets the UI, and has an advantage over the rest.


As I understand, a bot conducts a random simulation. Thus, it is conceivable that playing a board 1000 times the bot will see the same sequence 40 odd times. And it is reasonable the conclusion of the simulation will be different at least once. So, even though such hand may well obtain the same result the route ought to be different.

In order to be a duplicate contest it necessarily must duplicate conditions and accordingly, because bots act on random simulations they must be their own random simulations (distinct from reusing simulations by different bots). Otherwise the server policy is a grift of the customer not getting the duplicate he paid for (independent robot simulations).

[smerriman]

axman, on 2025-May-01, 03:53, said:

As I understand, a bot conducts a random simulation. Thus, it is conceivable that playing a board 1000 times the bot will see the same sequence 40 odd times. And it is reasonable the conclusion of the simulation will be different at least once. So, even though such hand may well obtain the same result the route ought to be different.

In order to be a duplicate contest it necessarily must duplicate conditions and accordingly, because bots act on random simulations they must be their own random simulations (distinct from reusing simulations by different bots). Otherwise the server policy is a grift of the customer not getting the duplicate he paid for (independent robot simulations).

Robot tournaments are intentionally seeded so that if humans make identical choices, the robots will use the same random numbers in simulations and thus come up with identical bids and plays. Otherwise, people get upset when they got a bad score because a robot at another table acted differently (even though this is precisely what happens in human tournaments).

[barmar]

smerriman, on 2025-May-01, 13:54, said:

Robot tournaments are intentionally seeded so that if humans make identical choices, the robots will use the same random numbers in simulations and thus come up with identical bids and plays. Otherwise, people get upset when they got a bad score because a robot at another table acted differently (even though this is precisely what happens in human tournaments).

This can also be considered a fault of human tournaments -- your result ideally shouldn't depend on the random factor of which opponents you happened to encounter.

Of course, in most real world tourneys you can't avoid this to some extent -- you might run into Precision opponents on a hand that's well suited to their methods. But we'd prefer that our results mostly be dependent on things that we can control.

[pescetom]

Even if BBO does not want to do the necessary work (or colloborate with WBF on future rules about such) to hide human tempo, it would seem trivial to at least hide robot tempo: impose a reasonable minimum time (say 1.5 seconds) plus occasional extra random delay.

This would help to minimize adverse reactions of human players to robot-robot pairs too: they are understandably disconcerted by the instant calls and explanations of opponents.