[jm02130]

It's one thing for the robots to make bad bids. It's quite another for the explanation of what the bid shows to have no relationship whatsoever to the robot's actual holding. I've encountered that situation twice now in the last few days. Don't state that a bid promises support for a suit when in fact the robot bidder has a singleton.

[tritonium]

In what game are you playing where the stupid robot explains its bids?

[PhilG007]

 jm02130, on 2018-August-07, 12:40, said:

It's one thing for the robots to make bad bids. It's quite another for the explanation of what the bid shows to have no relationship whatsoever to the robot's actual holding. I've encountered that situation twice now in the last few days. Don't state that a bid promises support for a suit when in fact the robot bidder has a singleton.

If you think that you've been damaged as a result of a misexplanation,call the TD asap. If your complaint is upheld the TD will award an adjusted score. Of course this only applies when the opponents are
human.

[johnu]

 jm02130, on 2018-August-07, 12:40, said:

It's one thing for the robots to make bad bids. It's quite another for the explanation of what the bid shows to have no relationship whatsoever to the robot's actual holding. I've encountered that situation twice now in the last few days.

Only twice? You must not play much

[barmar]

Human players sometimes don't have what their bids ostensibly show, because there's no call that actually matches what they have. The same thing happens to the robots.

Advanced robots also use simulations in many situations to look for bids that deviate a little from what they hold, but look like they could produce a better result than the "book" bid.

[jm02130]

 tritonium, on 2018-August-08, 01:10, said:

In what game are you playing where the stupid robot explains its bids?

The ACBL IMPs game. When the robots start to do really dumb things I typically withdraw, so as to avoid distorting the scores for the rest of the field. I had a situation the other day in which a trump suit was agreed, I asked for key cards and when I discovered an ace was missing, bid 6, which was cold. The robot then decided to bid 7. There are serious captaincy issues with the robots.

[barmar]

 jm02130, on 2018-August-08, 09:56, said:

The ACBL IMPs game. When the robots start to do really dumb things I typically withdraw, so as to avoid distorting the scores for the rest of the field. I had a situation the other day in which a trump suit was agreed, I asked for key cards and when I discovered an ace was missing, bid 6, which was cold. The robot then decided to bid 7. There are serious captaincy issues with the robots.

Look at its explanations of YOUR bids. Probably one of them was explained as showing a really strong hand, like 25 HCP.

[wcneumann]

 barmar, on 2018-August-08, 08:46, said:

Human players sometimes don't have what their bids ostensibly show, because there's no call that actually matches what they have. The same thing happens to the robots.

Advanced robots also use simulations in many situations to look for bids that deviate a little from what they hold, but look like they could produce a better result than the "book" bid.

We could live with "a little." Unfortunately, there are no limits to how much the Robots can and will deviate from their explanations, i.e., I've seen deviations of up to 4 HCP and two cards in suits--even in the same explanation. I believe the simplest solution is to have explanations override the simulations when the HCPs, Total Points, and suit lengths differ by more than some number, like 3 in total. It may lead to less optimal bidding, but the explanations won't mislead the humans quite so badly.

BTW, I've also seen the Robot advertise 9 HCPs for 2NT bids over 1NT openings and only hold 8 HCPs. This has happened MANY TIMES. If the Robots are going to hold 8 or 9 HCP, then change the explanations if you believe in the simulations. Don't mislead the humans.

Just my 2 cents...

[smerriman]

 wcneumann, on 2018-August-08, 17:58, said:

BTW, I've also seen the Robot advertise 9 HCPs for 2NT bids over 1NT openings and only hold 8 HCPs. This has happened MANY TIMES. If the Robots are going to hold 8 or 9 HCP, then change the explanations if you believe in the simulations. Don't mislead the humans.

If basic GIB - with no simulations - often has 8 HCP when it says it should have 9 HCP, then I agree, the description is incorrect and should be updated.

If it is only advanced GIB which does this, then the description should not be updated (and it fact it might be logically impossible to update).

Why? Because the bid has been with the knowledge that you will assume he has 9 HCP and respond accordingly. If the bid is defined as 8-9 HCP, then that will result in you having to accept less often. That in turn will probably result in GIB no longer inviting with 8 HCP. You'll then complain that GIB always has 9 points despite advertising 8-9!

While there are innumerous GIB bugs, I would rather be mislead to get to the right spot than bidding by the book - which is just what Basic robots do, so you could always use them. Unless you're talking about tournaments, in which case I think you'll be outnumbered by your opponents' preferences. Perhaps BBO could let you choose between the two.

[aleaxit]

Consistent explanations and robots deviating from them us one thing, and may be hard to fix. What's intolerable and should be easily fixable (I suspect it would be enough to just edit a bunch of text!) is when explanations are inconsistent, lacking logical coherence. This happens often.

For example, when (say) out of two possible heart raises, one is shown as being "11-13 total points", the other as "15+ total points" -- there being NO documented way to raise hearts with exactly 14 total points.

Even worse: in a cuebidding situation, on partner's 3H, 3S is explained as showing a spade FIT (spades having also been bid previously), while 4C is a cuebid showing the Ace of clubs and denying the Ace of spades (as if 3S was a cuebid being skipped -- which according to 3S's own explanation, it's not).

Then there are the cases where explanations, while not logically inconsistent, make no bridge sense. For example, the many cases where, say, 3D is explained as showing "4+ diamonds" when it would make zero bridge sense to bid it with just four cards; in fact the robots bidding it always happens to have 5+ (as makes bridge sense), but the explanation appears to be broader than the reality.

Not damaging but VERY annoying are explanations which just aren't "updated" as the bidding develops, such as a bid being explained as showing "strong rebiddable hearts ... 4+ hearts" -- the 4+ part should be snipped because the strong rebiddable part implies it, and more than it.

If the explanations are natural-language reflections of constraints actually programmed into the robots' bidding, the fix is less easy than just editing text, but even more important: the constraints should reflect every bid having a sensible meaning, and there being a legitimate bid for every possible hand. If, as I suspect, explanations and actual operating constraints are not automatic reflections of each other, then it should be just a job of text editing. One way or another, improving the quality of explanations removing all the issues I've mentioned here is probably my most-desired improvement to the robots!

[barmar]

We know there are lots of holes in the bidding database. It's not just text, it's a complex set of constraints and pattern matching. When we fill one hole, it's very easy to create another.

The periodic robot upgrades are mostly fixing problems like this, but there are still many to find and fix. All we can ask is that you keep sending robot reports (this is missing in the mobile app, but it will be in the next version, which will have all the features that are in the preview of the new web version) and posting in the GIB forum. Post specific hands, they're much easier for us to work with than broad complaints like this thread.

[jm02130]

We all realize that bridge is much more difficult to program than chess or backgammon. Just keep at it. Some bidding options would be nice too, instead of having to bid according to a single canned system. Bridge Baron will play precision or a weak NT system, why can't the bridge base robots? Yes, Bridge Baron plays these systems badly, but at least it tries.

[barmar]

 jm02130, on 2018-August-09, 12:03, said:

We all realize that bridge is much more difficult to program than chess or backgammon. Just keep at it. Some bidding options would be nice too, instead of having to bid according to a single canned system. Bridge Baron will play precision or a weak NT system, why can't the bridge base robots? Yes, Bridge Baron plays these systems badly, but at least it tries.

Mainly because it's all we can do to just fix the problems in the existing bidding database. Creating a new one for a different system would be a huge project.

[0 carbon]

 barmar, on 2018-August-10, 09:40, said:

Mainly because it's all we can do to just fix the problems in the existing bidding database. Creating a new one for a different system would be a huge project.

I think with your bidding database of millions of hands you might try some machine learning to get to a system with good alerts & results.

[barmar]

 0 carbon, on 2018-November-17, 23:18, said:

I think with your bidding database of millions of hands you might try some machine learning to get to a system with good alerts & results.

None of us have any experience in programming machine learning.

A few years ago we tried to use our archives to provide advice to the robots. We classified hand patterns and collected auctions, and created a database with the bids that we found at each stage in the auction. Then we had the robot discard bids that didn't appear in any of the saved auctions.

It actually had very minimal impact. One problem is that except for the first few bids, it's rare to have identical auctions with the same hand pattern, so there weren't many matches in the database. Even with several years worth of hands being mined.