Deinonychus antirrhopus

A Sloppy Defense of Dembski

Paul over at Exiled from Groggs has a defense, sort of, of Dembski's "Triad". The problem is the comment about the problem with Dembski's Triad is correct it is, IMO, a little sloppy. Further, it is this sloppiness that lets somebody like Paul wiggle out. Here is how I'd re-word the critique:

Prob(N|I) = x,

Prob(C|I) = y,

Prob(D|I) = z.

For those of you who've read my posts on this before you probably already know what the above means, but for those who don't, Prob means probability, N means necessity, C means chance, D means design, and most importantly I is our current set of scientific information.

The criticism that Dembski does not take into consideration the unknown means the following. That based solely on current scientific information Dembski's filter can set the above probabilities as,

Prob(N|I) = 0,

Prob(C|I) = 0,

Prob(D|I) = 1.

That is, once something is designed it is forever considered to be designed. The problem is, as the commenter is trying to point out, is that science keeps moving forward taking no hypothesis as being a fact (i.e. having a probability of 1 attached to it). When we stop doing that we leave the realm of science and enter the realm of dogma (political, religious or otherwise).

So, while the Triad maybe complete, the problem is that since our (scientific) information set is incomplete (and probably always will be) Dembski's argument is not at all helpful.

Realizing this possible objection Paul decides to put up another defense of Dembski. In this case, Paul points out the problem above,

The other way in which the challenge is presented is to point out that we don't know every regular process, so we can't exclude the possibility that there is a regular process that we don't know about which has been overlooked as an explanation.

However, from here he wanders off into pushing this problem back onto the opponents of ID. The logic here is as follows:

If you can't show Dembski is wrong, then he is right.

He even presents a challenge to ID opponents where they are to come up with a method for detecting design. Of course, this is just a diversion, IMO, since one can use the Bayesian method just fine to detect design. For example, one could have used the Bayesian method to detect design or not in the Caputo case (a famous Dembski example--to see this click on the extended Entry).

From this point on Paul goes off into an area that even Dembski has decried: pulling probabilities from one's third point of contact. But realizing even that doesn't help Paul looks at the idea of setting these probabilities via the democratic process. Yep, that's right the probability of flipping a fair coin is to be determined by a vote. Evidence schmevidence, who needs to bother with that crud when we have voters!

The Bible argues that people know that there is a God, and they repress this knowledge. There aren't many people, I suspect, who behave as though there definitely is no transcendent other.

There you go! Those of us opposed to ID are merely suffering from repressed memory syndrome (ironically another debunked psuedo-scienctific endeavor). Yes I know I'm becoming sarcastic, but I can't help it when I see such smug pronouncements about the existence of God and then watch as many of these ID proponents turn around and with a straight face say, "Oh, but the designer doesn't have to be the Christian God or even a God at all....honest."

The Caputo Case Example:

In the Caputo case, an election official in charge of which candidate is at the top of the ballot was brought up on charges that he was rigging elections. Having the top spot on the ballot is seen as giving that party an advantage. Thus, the law states that a random method must be used that gives each party an equal chance of getting the top spot. The history during Caputo's tenure overwhelmingly favored the Democrats (IIRC). To see how the Bayesian method could be used to determine if Caputo was cheating (i.e. the design hypothesis) or not we'd first have to postulate a prior probability that Caputo was either fair or a cheat. One place to start is with a "non-informative" prior. In this case, we set the probabilities for fair and cheat equal, or

Prob(Fair) = 0.5, Prob(Cheat) = 0.5.

Then we use Bayes theorem to update this.

I selected an example that fits with my recollection where there were 47 Democrats, 1 Republican and then 2 more Democrats in the top spot. Then for each observation we update Bayes theorem. Now, remember our priors is that Caputo is fair with probability of 0.5. But after observing the first round of ballots with Democrats at the top we see the probability that Caputo is fair falling to 0.338 (note I've given a 0.2 probability that the Republicans could get the top spot).

As we proceed through each observation this probability (that Caputo is fair falls until we get to the Republican observation. In that case we go from a probability of 1.83639E-14 to 4.59097E-13. But with the next two observations the probability starts its downard descent again until we end up with a probability of 1.19507E-13 that Caputo is fair.

Now for most people this would be incontrovertible evidence that Caputo is a cheater. However, the court was reluctant to conclude Caputo's guilt on mere statistical evidence (i.e. I think they would have been happier for example if a witness saw Caputo just write in the Democrats at the top instead of using some randomizing method).

Even if we change things so that the first observation is a Republican, the final result is the same; a probability of 1.19507E-13 that Caputo is fair. One could also argue that we need a higher prior probability on Caputo's fairness (some sort of presumption of innocence). Even setting the prior probability of fairness to 0.95 does not really help Caputo in that there are so many D entries in the end, the prorbability that Caputo is a cheater is still 2.27063E-12. While that is almost twice as large as the initial probability it is still a miniscule probability.

Now, Dembski has argued that this kind of test is insufficient. After all, the courts did not find Caputo guilty (although they did get as close to guilty as possible--i.e. they pretty much took away Caputo's oversight of who gets the top spot). But for most people this is sufficient to conclude that something was not right with Caputo's method. That the process was not as "random" as it was supposed to be. To get around this problem Dembski came up with the lower probability bound. See the main entry above for discussion of this concept.

The Incentive Problem and National Security

Veronique de Rugy argues that we are ill prepared for the next "9/11" given the way spending is being mishandled by the Department of Homeland Security. In reading the article one part in particular jumped out at me,

That’s not to say it isn’t prudent to prepare for an attack. But federalizing first-responder programs accentuates the incentive problems that already plague the political process. When such programs are a state responsibility, legislators have a strong incentive to accurately assess the risk and potential damages to their states. They have to decide whether to spend more on homeland security or on other accounts. When these programs are funded at the federal level, by contrast, a congressman from Wyoming has no incentive to admit that his state is not a likely target or that if it ever were a target, the damages would be limited. He has no incentive to turn down federal money and even less incentive to volunteer taxpayers’ dollars for other states.

Clearly there is an incentive problem here. If "homeland security" spending were more decentralized (at least in some regards) then it is likely that the spending would be more focused to where it is most needed. For example, suppose somebody sets off a dirty nuclear bomb in the middle of Wyoming, is it bad? Sure, but is it as bad as a dirty bomb set off in downtown San Francisco or New York? Should Wyomming spend as much on safe guarding against dirty bombs as California or New York? But when such funding is federalized then they politicians in Wyoming have little incentive to turn down such money and make sure it goes to places that need it more. In fact, they have an incentive to take the money. They can hire people to prevent dirty bombs from being set off in the middle of nowhere. This is basically pork barrel spending. It does damn little, but it generates good will for the politician. Further, since funds are finite, then this means that locales that are more at risk are now less well protected.

This is why it is bad to provide at least a minimum amount of funding for homeland security to each state. Or as Ms. de Rugy notes,

To make matters worse, Congress provides every state with a guaranteed minimum amount of grant money regardless of risk. As a result, rural, less-populated areas receive a disproportionate amount of money. Of the top 10 grant recipients, only the District of Columbia also appears on a list of the 10 places most at risk of attack.

You know, it would be one thing if we incurred a deficit to ensure there wasn't going to be another terrorist attack of a magnitude similar to or larger than 9/11. However, to incur a deficit and have it be so blatantly wasted is galling. Exactly how does a tip line on child pornography help improve national security? It may be a laudable program and such funding probably should be granted, but doing so in the name of national security is despicable, IMO.