[gwnn]

On some correspondence chess websites, there is a rule against using chess engines (computer programs) to choose your moves. Since there are tens of thousands of users, they have an automated system that will find users with unusually high % of correlation with computer moves and ban people without recourse if their rate is consistently high. I wonder if something like this could work in bridge as well?

As a simple example, suppose we want to decide whether 3rd hand is sending illicit information to the opening leader. All we would need to do is make a prediction function as to which suit 3rd hand would want to signal for and see how often 1st hand finds that lead. There are obvious caveats but there must be a % over and above which you are looking quite suspect, right? For example, suppose we decide that a random chance of finding partner's suit is 33% and a good expert can improve on that all the way to 66%. What would we think if someone found partner's suit 90% of the time? And upon further monitoring, it kept being really high?

Caveats (so no one needs to point them out):
1. Obviously there are people who jist like leading passively. That will cause them to hit partner more often than aggressive leaders. This should be taken into account and there's probably a wide range between honest players wrt style, even if they have the same "aggregate skill" in opening leads.
2. Auction. Yes I know that there are overcalls and only unbid suits and lead-directing doubles or even lead-directing passes. This should be an issue at some point, perhaps already in the automation process or perhaps only as a second round of screening.
3. A high percentage of "lucky leads" is not full proof of anything but it is something to check out further. I'm just thinking out loud about a first filter. It would produce a lot of false positives and perhaps also false negatives too.
4. I know what "standard deviation" and that if you investigate enough people some people will have a lot of success even if they just lead using two coin tosses. Thanks.

Similar but more complex methods could be used to find people who have a "wire" when playing a hand or finding double-dummy slams, but probably it would need a lot of work to make it anything feasible. Part of the reason is that computers suck at single-dummy play and high correlation with GIB actions signals nothing useful except perhaps that you need to get your head examined. Still, I wonder if there is some useful information to gather about possible cheating single-dummy if we compare to double-dummy success. Again I am not saying that the top 10 declarers are all cheaters because they make more tricks than the rest of declarers. But suppose all bridge hands were about a 60% line and a 40% line (with no overlap) and all you needed to do is identify which is which. Then beginners would hover around 50% success rate, wannabe/unlucky experts perhaps even close to 40%, while real experts at 58-60%. (I know bridge is not like this! It's just a toy example. I know that I have a lot of "guesses" in my declarer problems that for experts is a certainty or near certainty. And they can induce me into solving these issues for them, and... ) So if someone consistently made 70-80% of these toy bridge contracts, we should be rightfully worried about them. Like I said, I don't know how much sense this second example makes. Possibly none at all.

So anyway what do you think of such a crude or less than crude filter to check for cheating?

[Zelandakh]

gwnn, on 2015-September-17, 06:29, said:

So anyway what do you think of such a crude or less than crude filter to check for cheating?

I think it is considerably worse than separating partners physically and using 2 tables. I also think it is entirely impractical - take, for example, a pair that has an undetectable way of cheating on OL using the existing set-up. On leads in the 40-50% range they always "guess" correctly. On leads below 40% they are clever and make the normal play. So all of their lucky leads are replicated at several other tables. Do you think this method would be sufficient to convict the cheating pair?

If instead we make physical cheating impossible and scan for electronic devices (using detectors or whatever) I predict we will have a much higher success rate. Prevention here is much simpler than detection.

[gwnn]

Oh I mean I agree with you completely that electronic playing environments would be the best. No question at all about that. I'm just wondering if something like this would be workable in the context of non-electronic environments. Very possibly not. And I am sure it will not be implemented, or if it will, it will not be because of this post.

[hrothgar]

gwnn, on 2015-September-17, 08:27, said:

Oh I mean I agree with you completely that electronic playing environments would be the best. No question at all about that. I'm just wondering if something like this would be workable in the context of non-electronic environments. Very possibly not. And I am sure it will not be implemented, or if it will, it will not be because of this post.

First they ignore you, then they laugh at you, then they fight you, then you win.

[Cyberyeti]

gwnn, on 2015-September-17, 06:29, said:

On some correspondence chess websites, there is a rule against using chess engines (computer programs) to choose your moves. Since there are tens of thousands of users, they have an automated system that will find users with unusually high % of correlation with computer moves and ban people without recourse if their rate is consistently high. I wonder if something like this could work in bridge as well?

Unsure of how this works, if I play a 40 move draw, 37 of which were a book opening (perfectly possible in Catalan or Lopez) and the rest follow another game as those moves seem best, do I get accused of being a cheat ?

I don't think this sort of thing would work as a blunt instrument, but as an aid to highlight people in need of investigation it could have some use.

[WesleyC]

Again, I'm surprised there isn't already an automated system for flagging suspicious actions at high level bridge events. Online poker sites have been using automated detection methods for many years and although they aren't conclusive, they have been instrumental in finding targets for in depth manual investigations.

I heard Ish talking about a metric that Boye came up with to gauge "opening lead accuracy". If the ACBL had been running this metric over the recent major tournaments and seen Fisher/Schwartz scoring orders of magnitude better than the other top pairs, a giant red flag would've been raised.

[gwnn]

Cyberyeti, on 2015-September-17, 09:40, said:

Unsure of how this works, if I play a 40 move draw, 37 of which were a book opening (perfectly possible in Catalan or Lopez) and the rest follow another game as those moves seem best, do I get accused of being a cheat ?

I don't think this sort of thing would work as a blunt instrument, but as an aid to highlight people in need of investigation it could have some use.

No that's not how it works, usually book moves are known already and anyway book moves are often not top choices of engines! The "optimal chess game" according to engines is some sort of 4 knights or an exchange french lol. And it's about patterns etc, not a single game.

[billw55]

gwnn, on 2015-September-17, 10:58, said:

No that's not how it works, usually book moves are known already and anyway book moves are often not top choices of engines! The "optimal chess game" according to engines is some sort of 4 knights or an exchange french lol. And it's about patterns etc, not a single game.

It depends what "optimal chess game" means. Since chess has no score, there are only three results: white wins, black wins, and draw. There are many more than three legal moves each turn, so perforce there are always multiple moves that lead to the same result. In this sense, many openings are equally optimal.

[gwnn]

billw55, on 2015-September-17, 12:04, said:

It depends what "optimal chess game" means. Since chess has no score, there are only three results: white wins, black wins, and draw. There are many more than three legal moves each turn, so perforce there are always multiple moves that lead to the same result. In this sense, many openings are equally optimal.

This is a threadjack of a threadjack and I'd rather reply to it in a different thread (and not now) if it is of interest to people. Suffice it to say, what you say is trivially true but a bit irrelevant wrt what I was saying (there are moves that draw easily and moves that require you infinite precision to hold on to a draw, for example - but I know you know that so why bring it up?). Could we concentrate on the main suggestion I made?

[gwnn]

Zelandakh, on 2015-September-17, 07:04, said:

I think it is considerably worse than separating partners physically and using 2 tables. I also think it is entirely impractical - take, for example, a pair that has an undetectable way of cheating on OL using the existing set-up. On leads in the 40-50% range they always "guess" correctly. On leads below 40% they are clever and make the normal play. So all of their lucky leads are replicated at several other tables. Do you think this method would be sufficient to convict the cheating pair?

If you have enough data, of course it would be enough for us to convict/zoom in on the pair. I think there are loads of 40-50% (or 40-60% to make it symmetric - please let's not get into what we mean by these percentages!) leads. If you assume they get right all of the 40-60% "guesses" and follow the field on the 0-39% and 61-100% leads, they will have a huge peak performance in these middling cases, and all we need to have is that there be enough of these middling cases to make their success obvious, which will happen with enough data. I am not suggesting to catch them based on the extravagancy of leads but based on very high % of success (and I am not saying that the top 10 opening leaders are automatically cheaters either, just that there is probably a limit somewhere).

[billw55]

gwnn, on 2015-September-17, 12:32, said:

This is a threadjack of a threadjack and I'd rather reply to it in a different thread (and not now) if it is of interest to people. Suffice it to say, what you say is trivially true but a bit irrelevant wrt what I was saying (there are moves that draw easily and moves that require you infinite precision to hold on to a draw, for example - but I know you know that so why bring it up?). Could we concentrate on the main suggestion I made?

by all means, gwnn

I think it is a good idea, as long as the limitations are understood. Obviously it should only be a flag for a closer look by real people. Perhaps three step process: an NBO could have a small committee that looks at cases flagged by the software, and then decides whether to recommend a more detailed investigation. Perhaps we could stipulate that some number of flags is necessary before inquiry.

Also importantly, we must be careful about harming reputations of noncheaters. There must be great care to keep the "flags" strictly confidential. This is difficult: leaks happen, theft and attacks happen, people duplicate the software independently, etc.

[steve2005]

On the chess front. I know of some sites which can tell if your running a chess engine on your computer, which is of course is forbidden unless you are declared as using a computer but I know of only one site which allows that. This doesn't stop you setting up a second computer and using a program on it for your moves.