The Guantánamo Files and Future Violence Assessment

In Guantánamo Files, the New York Times coverage of Guantánamo from WikiLeaks documents, one piece in particular caught my attention: a discussion of the difficulty of judging detainees’ risk of future violence. Among other facts, the authors bring up two specific cases. Said Mohammed Alam Shah, who was detained in 2001, was deemed to pose no future threat to U.S. interests and sent back to Afghanistan in 2004, where he revealed his real name, Abdullah Mehsud, and went on to plot numerous high-profile attacks before detonating a suicide bomb in 2007. Murat Kurnaz, on the other hand, was judged “high risk.” He was eventually released in 2006, under pressure from Germany (his country of residence) and Turkey (his country of birth) and went on to write numerous critiques against the prison and the system – potentially embarrassing to the U.S., but hardly evidence of “high risk” activity.

So, given both the problematic nature of the evidence in the Guantánamo cases and the high stakes involved, how does one proceed? How do you decide if someone poses a risk or not?

Predicting future criminal violence: past behavior predicts future behavior

A review in the February issue of Current Directions in Psychological Science focuses on just this question: how do we assess the risk of future violence in a criminal population? The news, at least as far as Guantánamo Bay is concerned, is not very happy.

In a review of the various instruments currently used in violence risk assessment, the authors conclude that an unstructured, clinical approach, whereby a clinician selects, measures, and combines risk factors based on his personal experience and judgment, has the least empirical support of any method. Furthermore, there is little evidence that any of the leading validated instruments, that are based on a combination of actuarial, or statistical, patterns and clinical approaches, predicts violence better than any another. Indeed, most seem to point to general history and lifestyle patterns of the individual in question as the most important factors in predicting future violence. As John Douglas put it in his recollections of some of the FBI’s most notorious serial killers, “The best predictor of future behavior is past behavior.”

It all goes back to the old “thickness of the folder” finding. In a classic assessment of instruments that accurately judged the severity of mental patients’ histories, the single most important factor turned out to be the thickness of a patient’s folder – in other words, the sheer length and volume of clinical history. Independently, none of the questionnaires or assessment tools provided much greater accuracy.

But what if there’s no past behavior to go on?

This should give us pause. Unfortunately, in the case of Guantanamo, the folders, or the various measures other than clinical assessments, are generally unavailable. These are far from regular prisoners and don’t come with case histories on a platter. The available tools for forming judgment are limited. So, what to do?

It’s a frustrating question, and I don’t pretend to have an answer. What I will say is that your response will likely depend on your answer to another question: would you rather risk an Abdullah Mehsud going free, but also release a lot of Murat Kurnaz’s in the process; or, is the potential downside of a Mehsud so great that it is worth keeping the Kurnaz’s if it also means minimizing the risk of a Mehsud-type error? It’s signal detection theory (SDT)—how do we discern the ratio of signal, or actual hit (in this case someone who will pose a threat), to noise, or the fog that contributes to both false alarms (Kurnaz is high risk) and incorrect misses (Mehsud is not)—made even more sticky and unmanageable because of the countless political biases and moral conceptions (and preconceptions) that it inevitably raises.

So, what do you think? How can we put our most mindful minds to work on a question that affects not just stocks or food choices but very real political, and perhaps most importantly, human outcomes? How do we decide absent the possibility of an obviously artful choice?