9

Ambiguous Information and False Confirmation

While the CIA believes al-Libi fabricated information, the CIA cannot determine whether, or what portions of, the original statements or the later recants are true or false.

—CIA OFFICER, Report of the Select Committee on Intelligence on Postwar Findings about Iraq’s WMD Programs and Links to Terrorism

The two equilibria discussed in this chapter occur under leading questioning only. In the ambiguous information, selective torture equilibrium, both the Cooperative and Innocent Detainee types play “information” while the Interrogator plays “no torture” after “information” and “torture” after “no information.” The false confirmation, selective torture equilibrium is similar to the ambiguous information equilibrium, except that in this case only the Innocent Detainee is sufficiently confident that the Interrogator is Pragmatic and chooses “information,” falsely confirming what the Interrogator wants to hear. The Interrogator is satisfied with the confirmation and thus does not torture after “information.” The Interrogator tortures after “no information,” however, so the Cooperative and Resistant Detainees are tortured in this equilibrium. We consider the features of each before discussing the substantive interpretation, a real-world case, and then returning to the model.

AMBIGUOUS INFORMATION

The Cooperative and Innocent Detainees play “information” because their belief that the Interrogator is Pragmatic is greater than their (individual and independent) thresholds and , respectively. The Interrogator chooses not to torture because the Detainee’s confirmation has satisfied her threshold (). She tortures after “no information” because there is no chance an Innocent Detainee played this move. Only a Resistant Detainee does not provide confirmation in this equilibrium.

The information is ambiguous because the Interrogator has asked a leading question and has received the confirmation she wanted, but does not know whether it came from a Cooperative Detainee who could provide real confirmation or from an Innocent Detainee who would confirm anything so as to avoid torture. Since both Detainee types confirm whatever the Interrogator says, their responses provide no new information. Everything was the same as it was, based on the accuracy of the question. If the assumption embedded in the question was accurate, confirmation cements it (unwittingly in the case of the Innocent Detainee); if it was inaccurate, confirmation cements that false information.¹

Equilibrium Features

Figure 9.1 depicts the ambiguous information, selective torture equilibrium, with the darker- and lighter-shaded areas representing the moves “information” by the Cooperative and Innocent Detainees, respectively. The dotted area covers the portion of the parameter space supporting torture. In the surprise torture equilibrium, there were two information hiding thresholds, the Interrogator’s actual belief () and the belief the Detainee thinks is the Interrogator’s (). In this equilibrium there is only one because there is no possibility for misunderstanding under leading questioning and so the parameter capturing possible misunderstanding (u) drops out of the payoffs and the resulting equation for the threshold.

Figure 9.1 Ambiguous Information, Selective Torture Equilibrium

There are, however, two versions of the information revelation threshold, one for the Cooperative Detainee () and one for the Innocent Detainee (). Finally, note that there is no Innocent Detainee recognition threshold ( or ). The reason for this is that, in this equilibrium, the probability the Detainee is Innocent after observing “no information” is zero; only the Resistant Detainee chooses “no information” in this equilibrium. As a result, this equilibrium occupies the two-dimensional plane on the face of Figure 9.1 rather than a rectangular solid.

INFORMATION REVELATION THRESHOLDS, AND

Since both the (knowledgeable) Cooperative and (nonknowledgeable) Innocent Detainees provide information under leading questioning in this equilibrium, each has an information revelation threshold. As we noted in Chapter 7, there were no solid grounds to constrain these independent thresholds and they could lie anywhere along the vertical q axis. I have set the Innocent Detainee’s threshold below the Cooperative Detainee’s threshold only to distinguish them. Nothing depends on this ordering, however, and it could be that the situation is reversed or that the thresholds are equal. The point for our purposes is that the equilibrium space is the region north of the lowest threshold.

As they increase, the equilibrium space shrinks. As the Detainees become more suspicious that the Interrogator will torture them anyway, even if they provide exactly the confirmation sought by the Interrogator, they become less likely to provide that confirmation. Conversely, the more the Interrogator is able to convince the Detainees that she will not torture if only the Detainee supplies her with the answer she wants to hear, lowering the thresholds, the more the equilibrium space grows.

Perversely, then, the more a detainee accepts the interrogator’s logic of torture—no information results in torture, but pleasing the interrogator means avoiding it—the greater the chances are for ambiguous information. In other words:

Implication 9.1 (Torture’s Logic and Ambiguous Information). Everything else being equal, the more the logic of torture works as its proponents envision, the greater the likelihood of ambiguous information elicited by the interrogator under leading questioning.

We can also think about the threshold and the logic of torture mathematically. Given that, all else being equal, the thresholds and go down as the torture costs k imposed on them increase, it follows that the space supporting ambiguous information increases. We have, that is, confirmed the common sense intuition:

Implication 9.2 (Brutality and Ambiguous Information). Everything else being equal, as the brutality of torture increases, the greater the likelihood of ambiguous information elicited by the interrogator under leading questioning.

INFORMATION HIDING THRESHOLD,

“Hiding” information may seem contradictory under leading questioning as compared to objective questioning. In the latter, the interrogator must decide at one point that she is satisfied with the information provided by the detainee, not knowing what the total possible amount and quality are, and thus stops torturing. In the former, the interrogator tells the detainee exactly what she wants to hear, so the “total possible amounts” (if not the quality) of information is what the interrogator decides to ask, not what the detainee actually knows.

The interrogator must, however, still be satisfied with what she has heard before she stops torturing. Her satisfaction will depend both on what the detainee is willing to confirm and, crucially, on what she wants confirmed. The higher her demands, the more she wants confirmed, the higher her threshold.

This can also be seen in the individual parameters constituting the threshold: . First, as we have said repeatedly, the Interrogator’s prior probability the Detainee is Cooperative p_C is likely to be high, whereas the prior probability the Detainee is Innocent p_I is likely to be low. Of all the costs in , the one of most concern to interrogators and their employers is r, the reputation or credibility costs from not using torture on uncooperative detainees. As we have said, the torture cost c is likely to be very low, as is the “unnecessary” torture cost a (keeping in mind ). Combined with a low probability the Detainee is Innocent, we know from Proposition 7.4 that approaches one-half.

Even so, recalling Proposition 7.4, the upper limit to remains under one-half. This pushes the equilibrium region to the left side of the parameter space, covering more area on the face, as captured in Figure 9.1. This means more area for ambiguous information and reduces the area covered by torture (to the left of the threshold). As the threshold approaches one-half, ambiguous information decreases, but the space occupied by torture increases. In other words, the more the Interrogator demands in terms of confirmation (the higher the threshold), the more likely the use of torture, even under leading questioning, suggesting the following implication:

Implication 9.3 (Torture Persists under Leading Questioning). Leading questioning will not eliminate torture; the higher the interrogator’s demand for confirmation, the more torture.

FALSE CONFIRMATION, SELECTIVE TORTURE

The false confirmation, selective torture equilibrium is the other equilibrium which occurs under leading questioning only. This time only the Innocent Detainee plays “information,” falsely confirming what the Interrogator wants to hear. The Cooperative Detainee believes that the Interrogator is Sadistic and not Pragmatic () or suspects that the Interrogator is Pragmatic, but will believe that the Detainee is still withholding information and so will torture anyway ( and ) and as a result chooses “no information.” (Remember that although the Cooperative and Innocent Detainees have the same preference ordering, their beliefs can differ.) The Interrogator has received the confirmation she wanted with “information” and so does not torture. The Interrogator does, however, torture the Cooperative and Resistant Detainee types after “no information.”

Since the Detainee providing the confirmation in this equilibrium is known to be Innocent, the confirmation provided by the Detainee cannot be objectively correct, accurate confirmation. At best, if the questions asked are accurate, the confirmation is unwittingly true. If the questions are inaccurate or misleading or based on erroneous information, then the confirmation will be as well. In either case, the confirmation provides no new information.

Equilibrium Features

Figure 9.2 presents the equilibrium, showing the Innocent Detainee’s information revelation threshold. The darker-shaded area above is the region covered by the Innocent Detainee’s confirmation (choice of “information”). The area covered by dots represents the torture of the Cooperative and Resistant Detainees.

Figure 9.2 False Confirmation, Selective Torture Equilibrium

INFORMATION REVELATION THRESHOLD,

We have discussed the effects of the Cooperative and Innocent Detainee thresholds under leading questioning above, in the context of the ambiguous information, selective torture equilibrium. The only difference here is that the Cooperative Detainee’s belief falls below or is above it, but his belief that the Interrogator will be satisfied with his confirmation falls below his threshold . In short, a Cooperative Detainee’s suspicion and mistrust of the Interrogator helps sustain this equilibrium. If he was not suspicious and distrustful, he would choose information, collapsing the equilibrium.

As we said above, the Innocent Detainee does trust the Interrogator and is willing to falsely confirm what she asks of him. Indeed, the more he trusts her (the lower the threshold), the greater the area supporting false information. In other words, we have the exact same implication (9.1) for false information as we did for ambiguous information:

Implication 9.4 (Torture’s Logic and False Information). Everything else being equal, the more the logic of torture works as its proponents envision, the greater the likelihood of false information via confirmation of inaccurate leading questioning.

Indeed, there is the same parallel with the ambiguous information equilibrium in terms of torture’s severity as well. Thus, we also confirm the common-sense intuition that more brutality leads to more false information:

Implication 9.5 (Brutality and False Information). All else being equal, increasing the brutality of torture of an innocent detainee under leading questioning increases the likelihood of false information.

Substantive Interpretation

The substantive interpretation of these equilibria is that the Interrogator has asked a leading question and has received the answer she wanted to hear. In the first equilibrium she does not know whether the confirmation provided is coming from a Cooperative Detainee with knowledge or falsely confirmed by an Innocent Detainee trying to escape torture. In the second case she is certain that the Detainee is Innocent, but gets the confirmation she seeks. Note that this equilibrium also captures the case when an Interrogator has asked a leading question which is false, a Cooperative Detainee willing to provide information has refused to confirm the falsely premised question, and the Detainee is tortured for it!

Rather than thinking about Detainees as being purely Cooperative or Innocent, one way of thinking about this is to imagine their knowledge or ignorance with respect to different types of information. So, for example, a Detainee may have information on locations (safe houses, weapons caches, training camps) but not names or plots. Such a Detainee would be knowledgeable (either Cooperative or Resistant) with respect to the former, but Innocent with respect to the latter. Asking leading questions might result in ambiguous information since some of the information would be correct, but other information would not be. This sort of case would provide a real-world example of these equilibria. Thus, for a case in the real world to match them, it must meet the following conditions:

1. The detainee was subjected to leading and not objective questioning.

2. The detainee confirmed the interrogator’s leading questions.

3. The ultimate value of some of the information was false and other was unclear.

4. The detainee was not tortured afterward.

Real-World Case: Ibn al-Shaykh al-Libi

Ali Mohammed al-Fakheri, better known by his nom de guerre, Ibn al-Shaykh al-Libi, provides just such a case.²

Born in Libya, al-Libi fought with the mujahideen against the Soviets in Afghanistan and by the 9/11 attacks had become the internal emir, or chief, of the Khalden jihadist training camp.³ He fled following the U.S. invasion and was picked up in Pakistan sometime at the end of 2001. After about two weeks he was turned over to the Americans and taken to an interrogation and detention facility at Bagram airbase outside Kabul in Afghanistan.

A pair of FBI agents were the first to interrogate him. One, Russell Fincher, was a devout Christian and he quickly established a rapport with al-Libi, praying with him and talking about Jesus and Mohammed. In more than 80 hours of interviewing al-Libi, Fincher and his colleague got al-Libi talking—after having read him his Miranda rights. According to Fincher and Jack Cloonan, who read Fincher’s reports back in the FBI’s New York field office, al-Libi provided information on Zacarias Moussaoui, the so-called 20th hijacker, and Richard Reid, the “shoe bomber.”⁴ The Justice Department was building cases against both men at the time. There was even some hope al-Libi would cut a deal and work with the prosecution. He provided details about the operation of the jihadist training camps in Afghanistan. He told his FBI interrogators over warm coffee in a cold office about a plot to bomb the U.S. embassy in Aden, Yemen. Although it was in its final stages, the plot was foiled. Asked about Al Qaeda’s connections with foreign governments, however, he made no mention of Iraq.

Despite the information he allegedly provided, the CIA thought he was holding out and demanded to take over the interrogation from the FBI. It was al-Libi, then, two months before the capture of Abu Zubaydah, who became the first political football in the contest between the FBI and the CIA as to who would lead the interrogations of terrorist suspects. The CIA’s victory in that game and a harbinger of what was to come was communicated to al-Libi in dramatic fashion. A CIA officer who would later receive a reprimand from the Agency for threatening a different (and blindfolded) detainee with an unloaded gun and a power drill, interrupted a session with Fincher and told al-Libi he was going to be sent to Egypt.⁵ The officer promised he would “find [al-Libi’s] mother and … fuck her” (Isikoff and Corn 2007, p. 121).

Placed on the cold concrete floor with his socks, shoes, and gloves taken away, it took al-Libi only 15 minutes to decide he

would fabricate any information the interrogators wanted in order to gain better treatment and avoid being handed over to [Egypt]. According to al-Libi, after his decision to fabricate information for debriefers, he “lied about being a member of al-Qa’ida. Although he considered himself close to, but not a member of, al-Qa’ida, he knew enough about the senior members, organization and operations to claim to be a member.” “Once al-Libi started fabricating information,” he claimed, “his treatment improved and he experienced no further physical pressures from the Americans” (United States Senate 2006, pp. 79–80)).

The CIA did not stay satisfied for long, however, and soon guards tied al-Libi to a stretcher, duct-taped his feet, hands, and mouth, hooded him, and tossed him in the back of a pickup. The truck was driven straight into the bowels of a cargo plane which took off for Cairo.

The Egyptians made it clear what was in store for him if he did not provide information on future operations. Al-Libi claims not to have known about any other future attacks but said he began to make them up to avoid torture. His interrogators pushed his creativity to the limits, he said, when they demanded to know about Al Qaeda’s connections with Iraq. Unsuccessful in telling them what they wanted, they put him in a small box for about 17 hours. When he still failed to say something they wanted to hear, they knocked him down and punched him for a quarter of an hour.

He then concocted a story about three Al Qaeda members going to Iraq to learn about nuclear weapons. He said he supplied the names of actual Al Qaeda members in order to make his lies more believable. Believable they were, and he was rewarded with food. Some days later, however, the interrogators wanted information on connections between Al Qaeda and anthrax and biological weapons training in Iraq. Al-Libi claims he was once again unable to craft a story because he didn’t even understand the term “biological” and didn’t know what a biological weapon even was. The beatings ensued once again (United States Senate 2006, pp. 79–82)).

Receiving these reports back from the Egyptians, the Defense Intelligence Agency in February 2002 wrote up a report, a “Defense Intelligence Terrorism Summary,” that questioned the credibility of the claims made by al-Libi about Saddam Hussein’s support for training Al Qaeda in WMDs. Instead, the report stated: “[I]t is more likely this individual is intentionally misleading the debriefers. Ibn al-Shaykh has been undergoing debriefs for several weeks and may be describing scenarios to the debriefers that he knows will retain their interest” (United States Senate 2006, p. 77).

As is now well known, this did not prevent “President Bush, Vice President Dick Cheney, and other officials [from] repeatedly cit[ing] the information provided by Mr. Libi as ‘credible’ evidence that Iraq was training Al Qaeda members in the use of explosives and illicit weapons” (Jehl 2005a). Bush, for example, claimed “in a major speech in Cincinnati in October 2002 that ‘we’ve learned that Iraq has trained Al Qaeda members in bomb making and poisons and gases”’ (Jehl 2005a). Of course, we know now these claims were false.

In February 2003 al-Libi was returned to the Americans and transferred to Guantanamo. When questioned again by the CIA one year later, “Al-Libi claimed that he fabricated ‘all information regarding al-Qa’ida’s sending representatives to Iraq to try to obtain WMD assistance.’ Al-Libi claimed that to the best of his knowledge al-Qa’ida never sent any individuals into Iraq for any kind of support in chemical or biological weapons, as he had claimed previously” (United States Senate 2006, pp. 79–80)). As a result, the CIA reissued the intelligence reports with the recantation, in essence withdrawing the intelligence as untrustworthy (Isikoff and Hosenball 2005).

When asked why he backtracked on his earlier statements, al-Libi replied this way: “They were killing me. I had to tell them something” (Isikoff and Corn 2007, 124).

The ambiguity of the information received by al-Libi is captured nicely by the entry in History Commons, “an open-content participatory journalism” website akin to Wikipedia but focusing on events and containing supporting links to newspapers, documents, and other sources. It should be treated with caution and I have not relied on it for the account above, but an entry connected with al-Libi is apposite: “Provides Mix of Valid, False Information.”⁶

Of course, the key is how the CIA officials charged with evaluating intelligence from interrogations interpreted the information from al-Libi. As the epigraph to this chapter states, “a CIA officer explained that while CIA believes al-Libi fabricated information, the CIA cannot determine whether, or what portions of, the original statements or the later recants are true or false” (United States Senate 2006, p. 108). According to a “senior U.S. intelligence official, Al-Libi ‘changed his story, and we’re still in the process of trying to determine what’s right and what’s not right’ from his information. ‘He told us one thing at one time and another at another time.’ The official said that ‘the CIA did not know whether he was telling the truth’ about there being no connection with Iraq” (Priest 2004).

And finally there is former CIA Director George Tenet himself, who used the al-Libi “information” to support the Iraq war. Tenet put it this way in his memoir:

He clearly lied. We just don’t know when. Did he lie when he first said that al-Qa’ida members received training in Iraq or did he lie when he said they did not? In my mind, either case might still be true. The fact is, we don’t know which story is true, and since we don’t know, we can assume nothing (Tenet and Harlow 2007, pp. 353–354)).

BACK TO THE MODEL

It would appear that al-Libi should never have been a case study in this book at all, but rather in a book on how at least some terrorists immediately start singing to the FBI (Weiser 2014). It was only when the CIA demanded more, began asking leading questions, and threatened torture to get it that his case became relevant here.

In terms of the thresholds, was apparently rather low; al-Libi didn’t have to spend more than a quarter-hour in a cold cell before deciding to try and please his captors with whatever they wanted to hear in return for better treatment. His CIA and Egyptian interrogators, however, apparently had a higher threshold . Their threshold was high in two ways.

First, the interrogators had a high hurdle for what would satisfy them. They wanted to hear very specific “facts” confirming Al Qaeda’s pursuit of WMDs from Iraq and tortured him until they got them. Second, while al-Libi was willing to provide them with whatever they wanted to hear, his own ignorance of the subject matter made it difficult for him to give them what they wanted.

It is worthwhile pointing out that the RIT model excluded by assumption the possibility that a Detainee’s response to a leading question might fail to be understood by the Interrogator (that was why the u variable dropped out of the payoffs). We have in the al-Libi case a failure on the other side, not incorporated directly into the model because we gave the benefit of the doubt to torture proponents: a failure of the Detainee to understand an Interrogator’s question, even though it is a leading one. (Since it was not in the model, it was left out of Implication 9.3.) Presumably this would be a more common occurrence in objective questioning, but we see it even here. A failure to understand the question means that the Detainee effectively plays “no information” and is tortured until he does understand it—and falsely confirms it.

Viewing the al-Libi case through the lens of the model also captures something else of importance. It has been widely known for a while now that neoconservatives within the Bush administration led by Vice President Cheney were bent on invading Iraq no later than early 2002, whether or not Iraq had WMDs or links to Al Qaeda, as secret memos summarizing the Bush administration’s thinking and prepared for top UK policy makers illustrate (Ricketts 2002, Rycroft 2002). The problem was, the memos point out, that Bush administration hawks needed evidence linking Iraq to Al Qaeda’s pursuit of WMDs for justification. To this degree, the hawks could care less whether or not al-Libi was “innocent” of this knowledge. What they wanted was confirmation.

Both of the equilibria in this chapter express this idea in Figures 9.1 and 9.2. The equilibria are planes on the front surface of the parameter space and not three-dimensional solids because there is no chance the Detainee is Innocent after observing “no information.” The Interrogator believes a Detainee who confirms what she wants to hear is either Innocent for sure (in the false confirmation equilibrium) or either Innocent or Cooperative (in the ambiguous information equilibrium)—and she doesn’t care either way.

To this degree, the two “bad” information equilibria are special cases of bad information in terms of these Interrogator beliefs. It is important to note that this does not at all mean that this surface is the only region colonized by bad information in the model. We have seen this already, though we have not made much of it in our focus on the Cooperative and Innocent types of Detainees. In both these two equilibria and the surprise torture equilibrium in the last chapter, a Resistant Detainee may be providing false information in one interpretation of the move “not reveal information.” As we shall see below, however, this will be the case not just for the Resistant Detainee.

The argument thus far has shown that:

1. EITs are torture, and the effectiveness of interrogational torture is an open question. (Chapter 2)

2. The Bush program approximates closely the ideal model of interrogational torture and includes limits on torture; the Bush and ideal models provide benchmarks for comparison with the game theory models to come. (Chapter 3)

3. The Bush model generates strange, quixotic outcomes. (Chapter 4)

4. The Bush interrogational torture program is more realistically modeled as objective and leading question variants of an incomplete information game, with three types of detainees, two types of interrogators, and uncertainty about the amount and value of information provided. (Chapter 5)

5. By positing a set of Detainee strategies, calculating the Interrogator’s expected utility using Bayes’ Rule to identify her best response, and checking for incentives to deviate by any of the Detainee types, it is possible to derive a perfect Bayesian equilibrium in which a Detainee is tortured after providing information. (Chapter 6)

6. The RIT model generates nine perfect Bayesian equilibria, the formal and empirical characteristics of which generate important observations, propositions, and implications, including:

(a) The Interrogator’s thresholds for believing that a Detainee is Innocent after “no information” are less than one-half, with close to one-half and closer to 0.

(b) The Interrogator’s information hiding threshold under objective questioning is greater than or equal to one-half, whereas her information hiding threshold under leading questioning as well as the Detainee’s version of are a little less than one-half.

(c) Objective questioning (potentially) provides better information, but is necessarily accompanied by more torture, than leading questioning, which, however, provides less valuable information.

(d) All things being equal, interrogators are more likely to get less valuable information than highly valuable information. (Chapter 7)

7. Surprise torture of a Cooperative Detainee—even if he has provided all his information—is not only likely, but perversely more likely

(a) the more willing the Detainee is to divulge information and

(b) the more important the information is in terms of “unknown unknowns.” (Chapter 8)

8. The perversity under objective questioning persists under leading questioning, namely

(a) the more the Innocent Detainee believes Interrogator promises of no torture in exchange for confirmation, the more likely is ambiguous and false information.

(b) the more brutal the torture, the more likely is ambiguous and false information.

(c) the more important the confirmation is to the Interrogator and the more difficult it is for the Detainee to understand what is being asked, the greater the likelihood of torture.

(d) leading questioning will not eliminate torture; the higher the interrogator’s demand for confirmation, the more torture. (Chapter 9)

The next step in the argument is to examine and interpret the valuable information, selective torture equilibrium.