Torture might seem to be an effective means to extract information when 1) there are two or more subjects, 2) the subjects can survey different subsets of the message space, 3) the only element in the intersection of these subsets is the truth, 4) the investigator knows that these subsets have in common only the truth, and 5) the disutility of revealing the truth is sufficiently low that capitulation is utility maximizing. Condition 2) implies that the subjects have not pre-coordinated their actions in the case of torture. Pre-coordination would generally imply agreeing to given lie or an unhelpful mix of lies and truth. When conditions 1) through 5) hold, one might imagine, the investigator can continue to torture his subjects until their answers coincide. Once he receives the common answer, he can infer that it is true and cease torturing his subjects. As in our previous cases, however, the investigator cannot credibly commit to stop torturing his subjects once the truth comes out.
Koppl, R. (2005). Epistemic Systems *. Episteme: Journal of Social Epistemology, 2(September), 1-44.