Join Bridge Winners
All comments by Richard Willey
1 2 3 4 ... 7 8 9 10 11 12 13 14 15 ... 192 193 194 195
You are ignoring the author of this comment. Click to temporarily show the comment.
As soon as you put regulations into effect, folks are going to start to optimize around them
July 25
Richard Willey edited this comment July 25
You are ignoring the author of this comment. Click to temporarily show the comment.
This is vaguely reminiscent of Magic Diamond (An attempt to modify the a forcing pass system called Carotti - or perhaps ocarrot - into something that could be played in more events)
July 25
You are ignoring the author of this comment. Click to temporarily show the comment.
I'm sure that this would be possible, however, I am also quite sure that the size of the dataset would shrink enormously
July 24
You are ignoring the author of this comment. Click to temporarily show the comment.
1. This is more a question about Bots in general. However, I was think of a standard MP tournament format.

2. The bots are playing out a complete hand. The “effectiveness” of a lead is reflected in the MP score that is realized. The bots that are cheating ARE playing double dummy.

3. Ideally, you'd have comprehensive records.
July 23
Richard Willey edited this comment July 23
You are ignoring the author of this comment. Click to temporarily show the comment.
I have a simple suggestion: Rather spinning our wheels repeating the same complaints that Nic isn't releasing his algorithm, why don't we try something a bit more constructive… Can all the folks who are pontificating come up with a methodology that we can use to validate his system…

I propose the following:

Generate a series of virtual tournaments in which bot are competing against one another.

One set of bots with play legitimately. They'll use simulation based methods to bid / play / defend. However, they won't have a wire. To make things interesting, we can have a variety of different bots competing at different skill levels by constraining the length of the simulation runs that they are using.

A second set of bots will be provided with partner's hand when they are defending (perhaps when they are bidding as well if Nic has tests that look at this). Here once again, we can play some games by tweaking the frequency with which a cheating bot has a wire during the tournament.

In both cases, we can do something similar to cross validation and run multiple permutations with the various cheating frequencies hand strengths for a given set of hands to make sure that there aren't any weird cross effects.

At the end of the day we should be able to draw some conclusions regarding how reliable Nic's system is…

So, this is my proposal.
Any suggestions how this can be improved upon?
Alternatively, do folks see any major flaws?
July 22
You are ignoring the author of this comment. Click to temporarily show the comment.
> Nics analysis will provoke lawsuits when taken to
> the expected conclusion. I don’t want that to happen.

Maybe, maybe not…
But that's something for Nic and the layers to work out

> Nor do I want organizations to adopt this without serious V&V discipline.

What does “adopt” mean? I'd be shocked if any zonal authority ever attempted to use this system to bring charges of cheating. (At least not publically). Even if they system worked perfectly, I doubt that you'd ever be able to win a case in court.

The value of this system is to make administrators aware that cheating was pervasive and is most likely widespread.

If you want to have any hope of controlling cheating then you need to switch to an electronic playing environment. Anything else is shoving your head into the sand…
July 22
You are ignoring the author of this comment. Click to temporarily show the comment.
> Data tools used in regulated businesses are serious and not quaint

Perhaps, but I think that your mistake is assuming that many businesses or groups within businesses are regulated…

I have a bunch of friends who are biostatisticians for pharma companies.
Their work is highly scrutinized

I have other friends who work as quants.
All their stuff is highly proprietary and doesn't get shared with the folks in the cubicle next door, let along publically
July 22
Richard Willey edited this comment July 22
You are ignoring the author of this comment. Click to temporarily show the comment.
I think that it is a mighty slender reed on which to make claims that insurance companies can realize significant cost savings.

I suspect that there are benefits to playing bridge, just as there are benefits to walking / cooking / reading / travel / what have you. (Note that your own claims are not specific to bridge, rather they refer to social interaction)

I am not convinced that the marginal benefit of playing bridge compared to any one of a variety of other activities is significant enough to matter, I would be skeptical about making any such claim, and if we can't make such a claim I don't really see the point…
July 22
You are ignoring the author of this comment. Click to temporarily show the comment.
As a rule, I am not a big fan of security through obscurity. As Adam mentions, the folks who design cryptographic algorithms have very much moved to a model in which these are publicly disclosed and critiqued (largely because its very very easy to make seemingly insignificant mistakes that completely compromise the resulting system so its good to have a lot of eyeballs looking at this stuff)

I personally would have preferred to see Nic disclose more details about the details of the various cheating detection algorithms. I appreciate his concern that the adversary might be able to take advantage of this information; however, ultimately I don't think that this will matter that much.

In his book, Nic notes that the adversary is adaptive. For example, once it became clear that how powerful video analysis of the naive pre 2015 systems proved to be, it would be relatively easy for the adversary to increase the complexity of their signally system. Simply put, the types of successes that we saw with video analysis / code break in 2016 probably won't be replicated in the future.

In much the same way, I suspect that the utility of Nic's software is going to be degraded as the adversary becomes more selective about when they choose to cheat. If folks are able to limit themselves to cheating on a small number of high leverage boards then I suspect that it will become much more difficult to make a compelling case. I suppose that one might be able to train the expert system to also look for such high leverage boards, however, even here you're going to run into trouble with the size of the resulting data set.
July 22
You are ignoring the author of this comment. Click to temporarily show the comment.
How quaint…

Seriously, the processes that you are suggestion are much more heavy weight that what actually gets used in the real world.
One can argue whether this is good or bad, however when it comes to data analysis what you are describing is the exception not the rule.

FWIW, in my job before this one, I was the product manager for MATLAB's Statistics Toolbox. I spent lots of time working with end users in Finance, Automotive, and Defense. These folks were working on very large, very expensive projects. And, for the most part, projects like this one don't use formal software design methodologies.

As I said, this isn't always a good thing. See what happened with the Boeing SuperMax or all those collateralized debt obligations. However, this is the way the world actually work.

And, FWIW, black box algorithms get used and deployed all the time. (The exceptions tend to be in areas like banking where there are legal requirements that credit scores and lending models must be interpretable)
July 22
Richard Willey edited this comment July 22
You are ignoring the author of this comment. Click to temporarily show the comment.
During the BoG meeting on Sunday there folks made multiple references to a set of recommendations that some outside group had made regarding the ACBL's strategic directions.

Few quick questions

1. Who comprised this group?
2. Were they hired or did they work for free?
3. If they were hired, what was the selection criteria, and how much were they paid?
4. Do any ACBL executives or Board members have personal or professional relationships with the members of this group?

The major recommendation seems to have been that the ACBL needs to create a strategic plan… (Not very controversial) Will said group be compensated for helping with this work? If so, how much?
July 22
You are ignoring the author of this comment. Click to temporarily show the comment.
1. There is an enormous difference between software design methodologies that are used for formal software release and what happens with Exploratory Data Analysis. Suggesting that folks use the former for the later is laughable.

2. Its very hard to predict what might happen wrt to legal challenges, however, I would be shocked if anyone would ever use this type of system for enforcement type decisions. I wouldn't be surprised to see bellicose claims that it should be used this way, but I expect that saner heads will prevail. (I think that the value of this analysis is indicating the magnitude of the problem and not trying to prosecute any given pair for cheating)

3. “That’s hard for big data to identify except when event B has only one unique cause A”

This is a very interesting claim.
I'd love to know what this is actually based on…

(I have to work with latent variables and confounding effects quite often)
July 22
Richard Willey edited this comment July 22
You are ignoring the author of this comment. Click to temporarily show the comment.
> Perhaps ACBL should conduct their own study on the
> incidence of dementia among serious bridge players.

1. No one in their right mind pays attention to studies funded by groups with a stake in the outcome. Yes, there are examples in which pharma companies and device manufacturers fund medical trials, but the amount of oversight and expense involved in these is enormous.

2. The ACBL isn't competent to administer simple IT projects and you want them to hire folks to run something much more complicated.

3. I understand that you want people to believe that playing bridge helps ward off dementia. However, the data that you just cited is equally consistent with the hypothesis that “people with dementia don't traipse down to the bridge club”
July 22
You are ignoring the author of this comment. Click to temporarily show the comment.
> It's not even clear to me which concept is new

Is the concept new? Perhaps not…

Is doing all the work to collate / curate the data
Actually write code rather than talking about it
Put this into presentable format and publish a book?

Thats all very very new
July 22
You are ignoring the author of this comment. Click to temporarily show the comment.
1. The fact that Nic is not sharing his algorithm's / methods publically does not mean that he is not providing more information to concerned parties. (By which I mean folks administrating the WBF and the like)

2. It is possible to validate the accuracy of black box algorithms

3. I personally think that the value of Nic's software is going to degrade over time, not because the specific detection routines will leak, but rather because the adversary will adjust the frequency with which they cheat. As such I would have preferred that he had either provided more specific or delayed publication until he was willing to do so. However, this isn't my decision to make
July 22
You are ignoring the author of this comment. Click to temporarily show the comment.
If these savings were real, we should be targeting the insurance companies directly and soliciting funds rather than using the AARP as a stalking horse

I suspect that either the AARP or the insurance companies would want a hell of a lot more evidence than is available…
July 22
You are ignoring the author of this comment. Click to temporarily show the comment.
I have long been pretty skeptical regarding these types of incentive schemes and I didn't find the numbers that were baked into the proposal particularly compelling.

Part of my job at work involves evaluating different product proposals / engineering designs and providing recommendations regarding whether our company should make a go / no go decision.

We generally ask some pointed questions to the product managers who are making these sorts of pitches. Two of the most useful are:

1. How do you measure success? Suppose that your program gets put in place. How do we know that this is making a material difference?

2. What constitutes failure? How can we determine that this system is not working? Under what circumstances should we pull the plug?

I recognize that these sorts of things can be difficult to measure, however, if people can't provide good answers to these questions we tend to be highly skeptical about providing them with significant amounts of resources.

In the case of the incentive scheme, I don't actually believe that promises of future dollars from the ACBL will actually change people's incentives to recruit members. They'll certainly fill out forms and collect fees for work that they were already doing. I just don't believe that you'd see a tangible increase in recruitment.
July 21
You are ignoring the author of this comment. Click to temporarily show the comment.
@Nic

If you could send me a CVS file with the the data that you used to generate the scatter plot and the density graph, I can generate a couple tiff files and and see if folks like these better.

I don't need player names or anything like that, just the X/Y values
July 21
You are ignoring the author of this comment. Click to temporarily show the comment.
How eminently reasonable.

How dare you post such a suggestion…
July 21
You are ignoring the author of this comment. Click to temporarily show the comment.
thanks for the timely response
July 21
1 2 3 4 ... 7 8 9 10 11 12 13 14 15 ... 192 193 194 195
.

Bottom Home Top