The Problem with Equating Travel Routes and Terrorism: 34 Dead Civilians
A few weeks back, Seton Hall published a report showing that since the DC Circuit reversed the habeas petition of Mohammed al-Adahi, “the practice of careful judicial fact-finding was replaced by judicial deference to the government’s allegations. Now the government wins every petition.” The report traced a number of factors that, before al-Adahi, judges examined with some skepticism, but after, fairly regularly accepted as evidence that a detainee was a member of al Qaeda.
Among those factors were staying in certain guest houses and traveling a particular route that–the government effectively claimed–meant you were a terrorist. Thus, it no longer mattered whether you had fought for al Qaeda. In the absence of more direct evidence, the government argued that where you traveled was one piece of evidence that you should be detained as a terrorist.
Tellingly, while the government has a declaration they routinely submit in Gitmo cases on the significance of guest houses to al Qaeda, they have not (as far as I know) ever submitted a similar declaration providing evidence for a tie between travel routes and al Qaeda membership (the closest they have is a report on Tora Bora which seems to argue “if you were in this vicinity you must have been in Tora Bora and, Osama bin Laden!”). In fact, that’s part of what infuriated David Tatel in the Latif case–the way the majority opinion simply accepted the government’s evidence about Latif’s travel back to Pakistan–where hundreds of innocent of Arabs were picked up at the time–as corroboration for the error-ridden report the government submitted as its main proof that Latif could be detained.
Latif left Kabul in November 2001 and then traveled through Jalalabad before eventually arriving at the Pakistani border where Pakistani authorities detained him. According to the government, this path mirrors that of Taliban soldiers retreating from Kabul. Although not contending that this evidence is dispositive, the government argues that because Latif’s admitted route is consistent with that of Taliban soldiers and with information in the Report, it is a helpful piece in the puzzle, bolstering its claim that the Report’s inculpatory statements are accurate.
Fair enough, but how helpful? If this route is commonly used by innocent civilians, then the evidence is not that helpful at all. To understand why, consider a simple hypothetical. Suppose the government were to argue in a drug case that the defendant drove north from Miami along I-95, “a known drug route.” Familiar with I-95, we would surely respond that many thousands of non-drug traffickers take that route as well. Given what we know about our own society, the I-95 inference would be too weak even to mention. Cf Almerfedi, 2011 WL 2277607, at *4 n.7 (noting that some conduct such as possessing an AK-47 is so “commonplace in Afghanistan [that it] does not meaningfully distinguish an al Qaeda associate from an innocent civilian”). On the other hand, if the alleged drug trafficker had driven along an infrequently traveled country road, then a contention that that road was “a known drug route” would carry more weight. The burden of proof is on the government to demonstrate whether travel on a particular route to the Pakistani border, when considered in context, is mqre like the lonely country road and thus worthy of consideration when it comes to distinguishing between enemy combatants and innocent civilians.
I raise all this not just to point you to the Seton Hall report, which is well worth your time. But because today, SCOTUS will decide whether or not to accept two cases–Latif and Uthman–in which these issues are central (we won’t find out whether they’ll take the cases until Monday).
And because of this WSJ report, showing the tragic result of assuming that travel patterns must be indicative of terror ties: 34 dead civilians, targeted by Turkish warplanes after a US drone spotted a caravan of Kurdish smugglers using a route frequented by PKK guerrillas.
Above and out of sight, a U.S. Predator drone loitered. It was on a routine patrol when U.S. personnel monitoring its video feeds spotted the caravan just inside Iraq and moving toward the Turkish border, according to U.S. officials and the Pentagon’s assessment of the fatal strike.
U.S. military officers at the Fusion Cell in Ankara couldn’t tell whether the men, bundled in heavy jackets, were civilians or guerrilla fighters. But their location in an area frequented by guerrilla fighters raised suspicions. The Americans alerted their Turkish counterparts.
[snip]
Then Turkish warplanes appeared. “It was like a lightning bolt,” Mr. Encu said. “I saw a bright light and the force of the explosion threw me to the ground…When I turned my head I could see bodies on fire and some were missing their heads.”
The strikes lasted for about 40 minutes, survivors said. Of the 34 men killed, 11 were members of Mr. Encu’s extended family. It was the largest number of Kurdish civilians killed in a single attack in Turkey’s long conflict with the region’s militants.
[snip]
The killings sparked clashes between hundreds of stone-throwing protesters and the police in Kurdish parts of Turkey. In the town of Uludere, Mayor Fehmi Yaman charged that the attack marked the latest in a series of government efforts to intimidate the local population, much of which supports Kurdish militancy.
“The military knew these people were civilians. It was a deliberate attack,” he said. “The government has tried all means of suppression, which have failed, and now they tried this.”
The Turkish military initially said it ordered the strike because the convoy moved along a pathway frequently used as a staging point for attacks by the PKK.
[snip]
The killings threaten to spoil efforts to forge a Turkish-Kurdish consensus for a planned new constitution expected to partly address the issue of rights for the Kurdish minority.
Now, the US is hedging whether it told Turkey these Kurdish smugglers were PKK members because of their travel route. The Predator drones moved on, the government says; had they stuck around, maybe they could have confirmed these Kurds weren’t terrorists.
However convincing–or not–that hedge, the public Turkish explanation amounts to no more than travel route. They blasted a caravan of smugglers, killing almost all of them, because they were traveling on a route also used by PKK guerrillas. And with it, they blasted any credibility they had on wanting to engage their Kurdish minority, with potentially long-lasting consequences.
Obviously, the use of this travel-route-terrorism is different in the two cases. One delivers drone-assisted executions for “terrorism,” the other ratifies HUMINT-justified life imprisonment. But that’s why the Turkish example is so useful: because it provides a very graphic (and tragic) example of the costs of relying on such shoddy intelligence to target terrorists. With Gitmo detainees, we hide those costs down in Cuba or back in Yemen where detainee family members grow increasingly desperate for any justice from America. But the human and political costs are there, nevertheless.
Ah, yes. But as a society, we’re in love with the logic which creates this kind of intelligence nightmare. Google has a case against it right now by the Justice Department over hacking into personal networks from their slow-moving video cars while collecting ‘street view’ data for Google Earth and their Google Maps.
But the reviewing press is going gaga over Google Glasses, built off of what database? Bayesian inference and Bayesian filters run amok. And what, pray tell do you think gives the “Intelligence Community” the belief that they can invert “travel routes” and other data nexi to justify “military objectives” under Stephen Preston’s perversion of international law?
Every single global citizen should think long and hard about that before logging into a social network like Facebook or any other such data collector. It isn’t just that you’re giving up your data. You’re voting your consent for these methods, and these methods are killing real people.
And the costs keep piling up.
With compound interest on the unpaid bill, of course.
The longer that bill goes unpaid, the higher the price we will pay in the end.
Standard thinking in this area goes like this:
A contingency (between two events) exists when the conditional probability of the target event, given the conditioning event, is different from the baseline probability of the target event. When the difference is positive, the conditioning event can be taken as statistical evidence of the target event.
Thus, if the conditional probability of a person being a terrorist given that they are in a certain location is higher than the general (baseline) probability that a person is a terrorist (ignoring location), then a contingency between location and being a terrorist exists and the presence of that person in that location is statistical evidence of the person being a terrorist. The strength of said evidence depends on how much higher the conditional probability is when compared to baseline probability.
Now, here’s the problem with the above:
Statistical evidence is only that. The fact that there is a contingency (of some non-zero, positive amount) does not in any way imply a perfect relationship. Obviously, even if there is a (positive) contingency of being on a certain road and being a terrorist, that does not imply that everyone on that road is a terrorist. And most people know this, so this isn’t (yet) the problem.
The problem is that, to be blunt, human stink at on-line math involving multiple variables. In general, we can only track one continuous variable at a time and have a tendency to round all other variables to zero or one. Thus, if you are staring at some people on a road, wondering if they are terrorists, maybe even making an intelligent estimate of such, and someone points out that there’s a positive contingency between that road and being a terrorist, suddenly all that delicate math evaporates and you send in the bombers. You don’t carefully multiply the prior probability (based on dress, etc) by the conditioning factor; you simply decide: “well, that nails it! Attack.”
We’ve known this about people since the 1970s. Danny Kahneman got the Nobel for some recent work related to this. (Full disclosure: Danny K was one of the minion at my brother’s bris. tee hee) And, yet, my colleagues have failed miserably at getting the military to understand what this really means.
This seems relevant, just saw a link fly by on Twitter:
http://www.sltrib.com/sltrib/news/54126491-78/data-scanners-utah-dea.html.csp
@Jim White: Cause the place you wanna go looking for drug traffickers in HEAVILY Mormon counties (admittedly on the way to Las Vegas).
Someone has watched Butch Cassiday too many times.
One of the great perils for “pattern analysis”, the darling of IT specialists who would substitute statistics for on-the-ground intel, is the frequency of false positives. Concurrence and causality are not the same thing. The government needn’t care, however, so long as the courts and legislature refuse to question the executive’s policies, their consequences, and the secrecy that hides both.