Torture effectiveness - mistaking semantics for syntax

There are some very weird semantics associated with the question 'Does Torture Work?', to wit

1. "Yes It Does. I Have Proof" - The Dick (Cheney)
2. "No It Doesn't. People Will Say Anything" - Most Everyone Else
3. "It Works Sometimes" - The Remaining Few

Who is correct? Whose theories reign supreme?
Getting back to semantics, the issue ends up being, "What exactly do we mean by 'Work'"? and "Will you get the answer you need"?

 I suspect pretty much everyone agrees that

1. If you have the correct person, and
2. If you know the correct question, and
3. If you torture the crap out of the person
4. Then, you will get the correct answer. 

Simple no?

The underlying problem is, of course, how one separates the wheat from the chaff. The above assumptions also result in the following problems

Problem A (aka "The Szell Issue'): If you don't have the correct person, but you torture the crap out of them, you will get answers (unfortunately, not the ones you want). e.g. while Szell is laying on with a dental drill
    Szell: Is it safe?

    Babe: Yes, it's safe, it's very safe, it's so safe you wouldn't believe it.

    Szell: Is it safe?

    Babe: No. It's not safe, it's... very dangerous, be careful. 

(For the irretrievably slow of thought, the point being that since Babe has no idea what Szell is asking, and all he wants is for the damn drilling to stop, he'll say anything)

Problem B (aka "Poor Mister Rogers"): If you have the correct person, and you ask the wrong questions, and you torture the crap out of them, you will still get answers (again, unfortunately, not the ones you want).  e.g. CIA Guy waterboarding away like mad

   CIA guy: Who is the 20th highjacker?

   Bad guy: Atta!

   CIA guy: Who is the 21st?

   Bad guy: Habib!

   CIA guy: And the 22nd?

   Bad guy:  Mr. Rogers!  Just make it stop!

(Point being, of course, there is no Habib or Roger.  Or worse, Habib is the Bad guy's - innocent - father, and Roger is his - equally innocent - next door neighbor.  The Bad guy will say anything to make the waterboarding stop.  And Habib/MrRogers are next in line for waterboarding...)

The bottom line to all the above is that you just can't safely assume that torture will work.  You will always get answers.  Unfortunately, Selection Bias will result in cherry-picking results that you know worked, and omitting the rest.  

(Selection Bias - If I make the claim that I can reliable flip a coin and get 10 heads in a row, pretty much everyone would scoff.  However, if I tape myself flipping a coin 10 times, and repeat the experiment, i should get 10 heads around once every 1024 times (do the math).  Selection Bias is me pointing at the one successful attempt and saying "See!  I can do it!  I told you!")

The sad thing is that we're probably never going to know how many people we waterboarded, walled, and wiped the floors with, who gave us irrelevant, inconclusive, and idiotic information.  Its all part of the price we pay for our freedoms, right?