[berean]

I have been playing in Robot tournaments for a few years and have kept up with the GIB system. Last month I was having one of my best starts. I was coming in 1rst or 2nd 25% of the time in the robot games. Shortly after the middle of the month my ability to win, or place high up in these games was dramatically reduced. I felt that I was playing well and that something was not right. When I compared scores at the end of a session and went over the hands of the other players I noticed that when the bidding was the same by different players on the same hand, the GIB opening lead was not always the same. Since this intrigued as well as annoyed me, I decided to do some more analysis. I rented the advanced robot for a few days and analyzed the hands played at each table. The experiment study confirmed my initial findings. I then decided to contact BBO.

MY QUESTION: "Could you please explain to me why, when all bidding is equal, GIB’s bids and opening leads vary from table to table?"

BBO REPLY: "The robots are all programmed by the same software but also have the capability of choosing their line of play or defense, including the lead."

If you read the BBO reply carefully you will see that it not entirely correct. GIB robots do not have free will. They can only do what they are programmed to do, and a certain amount of randomness has been programmed into their selection of opening leads and play of the hand. The question that needs to be asked is why, and more importantly, is it fair to the player paying to play in what is supposed to be a game of skill, albeit with some level of chance.

Playing in Robot tournaments has been reduced from a game where a person with some level of skill and a basic understanding of the GIB system can strive to do well, to in what can be more accurately described as a random crap shoot. As of the beginning of this month I no longer play in Robot Tournaments. If anyone can confirm my observations I would like to hear from you.

[ArtK78]

This is far from an official reply, as I have nothing to do with BBO administration, let alone GIB.

If a robot concludes that there is a choice between two opening leads, it makes sense that it would choose one a certain percentage of the time and the other the remaining percentage of the time. If one is the winner and the other is the loser, then the luck of the opponent will depend on the choice of the robot.

[hrothgar]

Here's how I would describe things:

GIB, like almost all top bridge programs, uses Monte Carlo simulations to determine the best line of play.
As such, there is an inherently stochastic element to its decision making.
However, its important to note that GIB is not being random for the sake of being random.

With this said and done, arguably, it might make sense to synchronize the seeds for the RNGs that the GIBs use each round. This way, if you have precisely the same auction the GIBs will generate precisely the same lead (while still following their stochastic evaluation process)

In the abstract this doesn't sound too difficult.
It might even be a good thing.
However, I'll need to think about it some more.

[berean]

ArtK78, on 2012-February-08, 09:48, said:

This is far from an official reply, as I have nothing to do with BBO administration, let alone GIB.

If a robot concludes that there is a choice between two opening leads, it makes sense that it would choose one a certain percentage of the time and the other the remaining percentage of the time. If one is the winner and the other is the loser, then the luck of the opponent will depend on the choice of the robot.

If all things are equal then GIB actions should be the same. Let us say the auction goes 1NT-2C-2D-3NT with the two Robot opponents passing. I have observed that at different tables the GIB Robot opening lead can be different. One can only conclude that an element of randomness to GIB actions is being programmed into the Robots. Do you think that serves any useful purpose? Why should anyone one person or another be given a gift or a bottom on an auction like this because they received one random opening lead over another? Thanks for your reply.

[BunnyGo]

berean, on 2012-February-08, 10:18, said:

If all things are equal then GIB actions should be the same. Let us say the auction goes 1NT-2C-2D-3NT with the two Robot opponents passing. I have observed that at different tables the GIB Robot opening lead can be different. One can only conclude that an element of randomness to GIB actions is being programmed into the Robots. Do you think that serves any useful purpose? Why should anyone one person or another be given a gift or a bottom on an auction like this because they received one random opening lead over another? Thanks for your reply.

See hrothgar's comment. GIB partially works by running simulations (Monte Carlo) to try and figure out what would be best. To generate good simulations requires random number generators with a seed. When the seeds are the same, there should be the same result (if you could show this wasn't true, I'd be surprised). When the seeds are different, the two GIB programs may get slightly different simulations that cause it to choose differently between two or three close choices (it should rarely or never deviate from the top choices).

DISCLAIMER: none of the people responding so far work for BBO or are directly involved in programming GIB. We all simply know a little about programming and how GIB works.

[ArtK78]

Is it possible that the time spent in the auction, and not just the actions taken, also comprises part of the seed to generate the random numbers? In that case, identical auctions might generate different seeds and different results.

[barmar]

In the Main Bridge Club, there's quite a bit of random difference among the tables. First, there are 3 types of robots in use: 1) If you use the old BBO version, GIB runs on your PC and its thinking time is dependent on the settings you configure and your PC's CPU speed; 2) New version basic bots ($1/week rental) run on BBO's servers, but have very limited thinking time and use a simplified bidding algorithm; 3) New version advanced bots ($1/day rental) also run on BBO's servers, but they have more thinking time and a more advanced bidding process. In addition to this, the random seed is different at each table, so even the same type of

In robot tourneys there shouldn't be as much variation. These are all advanced bots running on the BBO servers, and they all use the same random seed. So if you have the exact same auction, you should normally expect identical play until the humans do something different. But even though these are all running the same program, there can still be occasional variations. Part of the GIB design is a limit on its thinking time, and computers are not always consistent on how much time they take to run the same code. However, I wouldn't generally expect this to affect opening leads, since it usually wouldn't have used up its time quota so early in the hand.

There have been no changes to the robot programming in recent months. The only changes we've made lately have been to the bidding rule database, most recently on December 15. See http://www.bridgebas...on-20-released/ for that announcement. If your success has changed in the past couple of weeks, I think it's just a run of bad luck, because we didn't do anything at our end.