Economist Roger Koppl has a paper, “Epistemic Systems”, in the journal Episteme: Journal of Social Epistemology, 2005, 2(2): 91-106 in which he discusses the logic of torture. Koppl argues that torture, as a method for extracting information, is effective only when two conditions hold.
First, the investigators must be able to recognize the truth when they hear it. Second, they must be able to credibly commit to stop torture once the truth is spoken.But torture will not yield reliable information because these two conditions do not hold. Thus torture is useless for intelligence gathering simply because governments cannot get around this basic problem. The government has commitment problems as far as stopping torture goes, in short, they cannot make a credible commitment to stop torture once the victim tells the truth. Torture victims know this perfectly well and thus will say anything and everything except what the torturers want to know, that is, they will hide the truth.
Two problems prevent governments from making a promise that is believable to the victim to stop torture once victim tells the truth. First, obviously they are using use torture because they don’t know the truth already. If they did they wouldn't need the information from the victim. But this means that they can't recognize the truth if the victim tells it. Second, even if they know they’ve got the truth, the victim is afraid they will keep torturing him anyway. Thus torture doesn't "work".
In a recent posting at the ThinkMarkets blog Koppl notes that,
If torture “worked,” we would have to consider seriously the arguments of its defenders, who tend to employ a kind of lifeboat ethics. If torture does not “work,” however, there can be no ethical defense of it.Update: The Inquiring Mind inquires into Tortured Reasoning.