Not the greatest defense

[gartinmale]

http://tinyurl.com/m3bd595 has the full hand.

 

Forgive the bidding, I was screwing around in a non-ACBL tournament. Regardless, GIB probably shouldn't have let this through (it even got the ending wrong since I had previously bid 2!d).

[1eyedjack]

There are a HUGE number of threads in this subforum in which GIB defender gets greedy and, rather than cash a setting trick, tries something else in the hope of maximising undertricks, only as a result letting the contract through. It may well be more prevalent in MP events than in IMP. I would certainly hope so.

 

I think that most of the time that this happens (not always) it results from the human player departing from system in the bidding, sometimes by a wide margin. I know that this has happened to me on a number of occasions. Had the human bid honestly (ie held the hand that his bidding showed), it might be reasonable to expect the cashing trick not to disappear in most cases.

 

I don't see an easy solution to this, but among the possibilities might be

 

1) Build into GIB a degree of scepticism regarding the trust that it places in other bids around the table, perhaps playing higher trust (100%?) in partner's bid but slightly lower trust in opponents.

2) Assign a higher priority to making/beating contracts over creating additional undertricks etc, regardliess of how much trust is placed in others' bids. May I suggest that this priority should be given greater weight the higher the level of the conttract. At slam level, given an opportunity to take it immediately one off, GIB should NEVER (in my view) delay taking that trick in favour of additional undertricks, even at MP scoring.

[Antrax]

I really think improving GIB's ability to deal with psyches is one of the last areas people should invest their time in.

[1eyedjack]

My knee-jerk reaction was to agree with you whole-heartedly. My more considered reaction having completed the commute home was to agree but with somewhat less vigour.

 

By my reckoning the priority to assign to a fix is a function of

1) the frequency with which the problem presents

2) the severity of the effect when presented

3) the ease of implementation of a solution

4) available resources

5) whether the proposed fix might solve multiple issues

 

the first 3 of the above are not absolutes but are measurements relative to equivalent factors for other issues vying for priority.

 

The possibilities are complicated by the fact that several alternative fixes may be viable for a given issue with varying degrees of effectiveness usually inversely proportionate to ease of implementation.

 

For example, a bespoke defence against psychs would be daunting project to undertake and the prospect probably feeds a knee-jerk antipathy to addressing this problem. By contrast, a sticky-tape solution of simply expanding a human opponent’s range of points and distribution by a point or a card per suit might be less effective than a bespoke defence but relatively trivial to implement for modest rewards. Personally I rank the frequency quite high, the severity quite high, and a cheap a cheerful fix of this nature not rocket science, although if I am assured that even this would be a complex task then I shall accept that.

 

The second measure that I suggested would I think reap dividends even where no-one has psyched. The fact that it might prove particularly effective also against psychs would just be an added bonus.

[Antrax]

You have to wonder what happens to regular bidding when you do that, though. My guess is that adding more possible hands to the simulation will worsen GIB overall, since it'll become more likely to find a biased sample in the time given.

It would be interesting if BBO staff analyzed a sample of hands to see how many first-round human actions GIB disagrees with, to get a measure of how frequent psyches are. It's not 100% because humans hand-hog so they might bid NT or spades over raising hearts etc, but I can't think of another easy way to try and figure out how much psyching really goes on.

[1eyedjack]

Sounds like a good idea.

 

It is certainly a drawback to my proposal, that the net benefit, if any, would be hard to predict in numerical terms, and probably could only be evaluated by trial and error, then measuring the result. It would be hard to persuade the developers to devote resources when the benefit is speculative.

 

In one of my wilder dreams I had a vision of the BBO server logging system "departures" by humans, tied to the identity of the transgressor. In that way a specific human individual's tendency to psych (and perhaps the types of psych) can be made known to GIB (perhaps only the opposing GIBs). It would require some storage space but not I think a lot. It would not have to store entire hand records, just a count of (eg) number of 1N opening bids with a singleton, or number of lies in response to Stayman and so on. Only a few bytes of space per parameter, which gets updated and overwritten as more current information becomes available.