Game Theory and Human Behavior, Part II

Remember the post Game Theory and Human Behavior, Part I? What, you don't? Okay, so I posted it back in February. Obviously it was meant as the first of a series of posts. You might want to read that one again before reading this post, because this one uses some of the concepts I explored in the first post.

The practice of torture is another signaling game. This one is a bit more complicated than the good employee signaling game or signaling in contract bridge, but the premise is the same. Let's call the person doing the torturing the investigator, and the person being tortured the detainee. (I refer to them both as male, since they always are on television and in the movies.) The idea is that the investigator seeks truthful information from the detainee. The investigator doesn't have the same knowledge space as the detainee, although the two may overlap. (If the investigator knew what the detainee knew, he wouldn't have to talk to the detainee at all!)

The detainee, on the other hand, seeks to provide the investigator with only enough information to satisfy him and stop the torture. So the detainee wants to give the investigator the very minimum amount of information that would satisfy his curiosity.

Economist Roger Koppl performed a thorough evaluation of the efficacy of torture in a 2005 paper entitled "Epistemic Systems." In this paper, he determines that torture works only if the following two conditions are met:

1. The investigator must be able to recognize the truth when he hears it.
2. The investigator must be able to credibly commit to stopping the torture upon hearing the truth.

If the investigator cannot recognize the truth when he hears it, then he won't find the truth regardless of the method of questioning. If the detainee doesn't believe that the torture will stop upon telling the truth, then he has no incentive to tell the truth.

Here's the problem: These two conditions are both necessary for torture to work, but they almost never occur together in real-life situations.

Consider the first condition. The more the investigator knows about the knowledge space of the detainee, the better he is able to recognize the truth. But how can he recognize the truth when he hears it, if he isn't privy to everything within the knowledge space of the detainee? This is how the detainee can get away with providing the least useful information out of the set of relevant information that he knows, or even worse, information that sounds truthful to the investigator but turns out to be false upon verification.

The second condition is similarly difficult to satisfy. Consider the case of a Chilean rebel, who broke down and confessed the names of the nuns and priests who had sheltered her. Her devout interrogators didn't believe that these holy men and women were involved, and continued to torture her.

Even if torture were effective on actual "bad guys," the consequences of using it on people who are actually innocent bystanders are the things nightmares are made of. In Discover Your Inner Economist, Tyler Cowen discusses the nightmare situation of being mistaken for a spy when you are nothing but a clueless American on vacation. How can you signal to your captors that you are not a spy? Anything that you do, from protesting your innocence, to providing them with information that sounds promising but turns out to be false, is indistinguishable from what a real spy would do in that situation.

There's always the situation in spy movies and political thriller TV shows such as 24, in which the entire population of New York City is in imminent danger from a ticking time bomb, which was set by a fanatical terrorist of some sort. The hero has to beat the information about the location of the bomb out of the bad guy. Usually it's a renegade cop played by somebody like Bruce Willis -- a guy who goes against the rules and saves the day, much to the chagrin and grudging gratitude of his superiors.

While these sorts of movies and shows are entertaining, they are not reality-based. Dr. Jean Maria Arrigo, a social psychologist with a mathematics background, refutes the ticking time bomb scenario in her papers "A Utilitarian Argument Against Torture Interrogation of Terrorists" (Science and Engineering Ethics (2004) 10, pp. 543-572) and "Torture, Terrorism and the State: a Refutation of the Ticking-Bomb Argument" (Journal of Applied Philosophy (2006) 23(3), pp. 355-373), co-authored with Vittorio Bufacchi. I focus upon this second paper.

Bufacchi and Arrigo first formalize the scenario. The dominant elements in the story are that first, a great number of lives are at stake; second, there is a time at which the catastrophic event will occur if it is not prevented; and third, a person privy to knowledge that will allow authorities to intervene and prevent the catastrophic event has been captured. Under these circumstances, some philosophers and legal experts believe that torture may be justified. A naive utilitarian argument might go something like, "we're causing one person some pain so to prevent that person from causing millions of others even more pain."

But this is an argument based on fiction. First, it assumes that torture is necessary for the captive to give up his information, when the opposite is more likely to be true. Consider the case described in Bufacchi and Arrigo's paper (p. 359):

Five foreign terrorists were captured by the local [counterterrorist police team]. All were found under arms with explosives and maps of targets....The question of how many [terrorist] cells were to be sent to the country to other targets was of interest. The first three terrorists were not even questioned, only shot. The next two were asked the question separately. One shot was heard. The officer said to the last terrorist, ‘Do you also want to remain silent?’ The guy began to lay out the entire operation, the training the cells had received, where they were to meet, where the weapons depots were located, and the route that the terrorists were to take to exfiltrate the country....The other cells were picked up along with in-country support personnel.

How was the fifth man chosen, you may wonder?

The group was searched and then fed and given tea as per the Shariat law of the Koran. Those who refused to eat or drink and made intense hostile eye contact were selected as the first three. The body posture decided who went fourth. The youngest, who ate the bread, drank tea, and thanked his captors was determined to be the least experienced. His AK rifle was not even clean...and he did not appear committed to the jihad.

So, they simply profiled the men they captured, picked the one who was most likely to confess first, and just asked him. No torture was necessary to obtain the information they needed. Once he was free from the influence of the more senior terrorists, and felt well-respected by his captors' adherence to local custom, he readily gave the police all the information they needed. Like the majority of "ticking-bomb success stories, the efficacy of torture interrogation is demonstrated only if the case is framed on that premise."

There are several premises upon which the argument for forward-looking interrogational torture is based. I will use Bufacchi and Arrigo's outline of the argument to explain this (Notation: P = premise, C = conclusion).

• P1: The terrorist has been captured.
• P2: Torturing the terrorist will make him reveal crucial information about the catastrophic event.
• C1: The terrorist must be tortured.
• C2: The crucial information will be revealed.
• C3: The catastrophe will be averted, saving the lives of many innocent people.

Unfortunately, the conclusions C1, C2, and C3 do not follow from premises P1 and P2. There are some hidden premises that must hold true before C1-3 can follow, and none of these premises are legitimate "from an empirical point of view" (p. 360).

P1 has an invisible, companion premise:

• P1A: This terrorist possesses crucial information about the catastrophic event.

Remember the overlapping knowledge spaces of the torturer and detainee? Well, they don't completely overlap; if they did, we wouldn't need to extract information from the detainee. So this premise is an assumption, and we could end up torturing some innocent bystander who was indistinguishable (based on our knowledge space) from a real terrorist.

P2 is problematic, too, as I have discussed above. But even if we assume that Koppl's conditions are both met, there are other problems. Many American POW's in Vietnam (e.g., John McCain) withstood a great deal of torture without confessing anything. Similarly, most victims of the Inquisition confessed to no crimes. Countless torture victims in the Algerian Civil War went to their graves with their secrets undiscovered. Terrorists are more than likely trained in the art of withstanding torture, making it even harder to draw any information out of them with this means. Finally, terrorists could be provided with false information to "confess" in order to deliberately throw their captors off the trail -- this tactic was recommended by Sun Tzu more than 2500 years ago!

And even if torture did yield information from the terrorist, what is to say that it will happen in a timely fashion? After all, if the time bomb is going to go off in 24 hours, anyone sufficiently devoted to the cause would use every ounce of their fortitude to resist for that length of time. Furthermore, torture interrogation is not a "quick coercion" -- it is a lengthy "degradation of the subject's resistance" over the course of months, not minutes (p. 361)!

According to Koppl, "It is a fair, but approximate, summary to say that, as a means of extracting information, torture works best when it is needed the least."

I couldn't have said it better myself.