The main logic of torture is to inflict so much pain that the victim reveals all his information to make the pain stop. Incentives for truth-telling in this situation are eerily similar to those in the bank stress tests.
All banks want to report that they are healthy. To distinguish the lying sick banks from the healthy ones there has to be some verifiable information. Healthy ones have this information (e.g. they passed the stress test) and the sick ones do not. The healthy banks have to have the incentive to reveal the information. This is all too clear for the healthy banks: by revealing their results they can avoid bank runs, get liquid, start lending etc.
In the torture analogy, a healthy bank is an informed terrorist with real information of an attack and a sick bank is someone, say an uninformed terrorist, with no information. The assumption of the pro-torture people is that the informed terrorist will have the incentive to report his information to avoid pain. But an uninformed terrorist has the same incentive. To tell one from the other, the informed terrorist’s information has to be verifiable. For example, there has to be “chatter” on Al Qaeda websites that can be used to cross-check the veracity of the torture victim’s confession. If this information is out there anyway, one might ask why torture was necessary in the first place. Presumably, the information is vague or ambiguous. The torture victim’s information brings some clarity.
This process seems heavily error-prone. False confessions may also cross-check by accident. The information is so noisy that a lead that is very weak may be thought to be strong. The more noise there is, the more the victim’s report is uninformative cheap talk – it contains no true information as informed and uninformed terrorists all give information, false and true and impossible to distinguish.
There is a second problem. Once a healthy bank releases the stress test information, the game is over – the market has the information and responds correspondingly. A torture victim faces further torture. There is no way for the interrogator to commit to stop torturing. If the victim knows this beforehand, the victim lies in the first place as there is no way to stop the torture. If the victim finds this out between episodes of water-boarding, the victim might start lying to contradict their earlier true confessions.
The efficacy of torture relies on verifiability of information and the ability of the torturer to commit to stop if good information is revealed. Both properties are hard if not impossible to satisfy in practice.
6 comments
Comments feed for this article
May 7, 2009 at 12:42 pm
oops
isn’t it possible that the pro-torture people believe the torture was applied to those that we knew were higher ups in the organization or that they believe it should only be applied to those that are higher up in the organization?
just because we failed to stop 9/11 deos not mean we had no idea about the hiearchy or who was in it. if the head of the bin laden unit of the cia (under clinton) is to be believed, he had a hierarchy chart on the wall.
May 7, 2009 at 5:03 pm
Etl World News | Assorted links
[…] 1. Torture and the bank stress tests: more related than you might think. […]
May 9, 2009 at 1:38 pm
bob
This is not the way to argue against torture.
Unfortunately, there are well understood solutions to both problems.
The first problem is solved by normal statistics, as dealing with error is one of its major focuses. In some sense this is arguing for more torture of more resources in more systematic ways; a larger sample size of resources with more rigorous procedures should be able to isolate any valuable trends within the data with a reasonable rate of error. It is also of course possible to design games that pair off untrustworthy resources against one another to produce higher quality information.
The second problem is also solvable. Basically the argument is that a resource does not have any incentive to release information. However even if one cannot convince the resource that the torture will stop, surely one can either find or create incentives that are indeed believable. And, not to sidestep the original problem, one can show that the value of a resource is negative due to maintenance costs once the information has been extracted. One can then create the incentive of self-destruction within the resource and offer that as a convincing and elegant solution that demonstrates mutual incentives for stopping.
May 13, 2009 at 11:00 am
Mother Theresa, Mahatma Gandhi and Waterboarding « Cheap Talk
[…] the Captain is on board with my earlier post. And Hitchens, while a difficult man to agree with on all topics, has some new information to […]
May 20, 2009 at 12:21 am
Torture: Incentive Compatibility « Cheap Talk
[…] in Uncategorized | Tags: game theory, incentives, torture, vapor mill, war | by jeff Sandeep has previously blogged about the problems with torture as a mechanism for extracting information from the unwilling. As […]
January 20, 2010 at 3:13 am
The Limits of “Enhanced Interrogation Techniques” « Cheap Talk
[…] by numerous commitment problems. We have blogged about these before. Here is Sandeep’s first post on torture which got this whole project […]