onsdag 10. september 2008

Bob Woodward's reports on secret assasinations in Iraq

Yesterday I heard about Bob Woodward's exposure of a secret killing program in Iraq. To describe it he used phrases like "Manhattan Project" and "The top secret operations, he said, will 'some day in history ... be described to people's amazement.'"

This made me think a bit about what this was all about. The thoughts below are of course pure guesswork (I don't have a security clearance, hopefully I never will). "Some day in history" I will either be proven right or wrong, but today this is just an intellectual exercise. How do I think the US has organized it's secret killing program in Iraq?

  • The US is selectively targeting central people in the terrorist networks. This means that they have some way of targeting the central people in these networks, which in turn means that they have some kind of metric to sort the Iraqi population by that puts the most central terrorists at the top.

  • A candidate procedure for creating such a metric springs to mind: If they can perform traffic analysis of messages being passed around in the terrorist networks, then they can figure out who is the most central. My guess is that they are hoovering all kinds of networks (phone, cellular, internet, whatever) for clues about who communicates with whom, then they create a big whopping matrix of who-talks-to-who links (nonsymmetrical or symmetrical depending on your preference). Then they run an eigenvalue analysis (a-la pagerank or T-rank). Seed the analysis with data about known terrorists, and see who comes out on top. (Undoubtly there will be a bit more to it than this, like e.g. building a tree structure out of the "strongest change in rank" connections and using that to find the most central people in subnetworks, but I am sure the mathematicians at the NSA and their subcontractors have had loads of fun on this problem and have found some concoction of transforms that does the trick nicely). Voila: Target list. Join (that's a technical term, see the link) that target list with a list of probable locations and you have a target list that can be used operationally (i.e. be used for selecting when to blow up a car or a house).

  • They then send out death squads (or "targeted killing teams" or whatever euphemism they are using) and assasinate the people on the target list. If the measured terrorist activity drops (fewer bombs, kidnappings or whatever), then they assume that the list is working.

  • Of course it's not quite this simple. The algorithmic approach sketched out above will generate a lot of targeting opportunities (a lot of person/location/time tuples), so some more logic is necessary. This probably means that somewhere in the loop some poor officer is presented with a series of choices: We have a probable location of target X in location Y. The cost (measured in e.g. collateral damage) of hitting location Y now is C, but if that number is lower than the estimated value of the target (say "D") then the hit is deemed cost effective, and a decision can be made quickly. "Hitting the target" (i.e. killing whoever happens to be in the target zone) can then be done using anything from a Hellfire missile (launched by unmanned predator UAVs) via snipers on rooftops to a bunch of Blackwater operatives running into a house with blazing (and/or silenced) guns, and anything in between.

  • It is by no means the first time the US is involved in this type of operation. The Phoenix Program during the Vietnam Conflict was in many ways similar. They killed a whole lot of people they believed were the enemy. They probably killed a great deal of enemy personnel, but also a whole lot of people who were not "Enemy" of the US in any particular way, and anyway they lost the whole war. One possibly big difference between the Phoenix program and this program (which surely has a codename that will one day be made public) is the kind of data mining that goes into the targeting. Another difference is the ability to put remote controlled munitions on targets without having to necessarily have to pick up individuals to kill them. Without a proper foresic team on the spot I would guess that it is quite hard to differentiate between a terrorist bomb blast and the detonation of a Hellfire warhead, or between stray bullets and sniper fire, so I guess that a lot of the asssasinations can be camuflaged as something other than what they are.

  • Update (14. aug.) Bruce Schneier just blogged his opinion on Woodward's statement. He believes the main issue here is improved tracking capabilities on UAVs. I agree this might be a factor, but unless you know where to point the sensors in the first, an improved camera (or whatever) won't help very much.
  • Update (Oct. 21) . Washington Post reported yesterday that DARPA is looking very seriously at real time video reconnissance. I don't know how easy or hard it would be to input data from this kind of analysis into a traffic analysis software package. If we assume it is easy, there really is not much of a difference between shaking hands and placing a phone call, if you consider both events from the perspective of a traffic analysis sofware package. They are both events, they take place between people. If you can track individuals (and that is a big "if") and possibly even link them to actual terrorist acts, then you have very useful input for tracking social networks of terrorists.
Well, that's my thoughts on the subject. In a few years we'll se if this morning's tea leaves gave a good or bad reading ;)

Just for the record. My personal opinion on this matter is that extra judicial killings are all bad. Criminals, including terrorists, should be caught and tried in a court of law and if found guilty put in prison. If that takes forever then so be it.

Update: (12. mar. 09) I ust read that Seymor Hersh has hinted that he knows something about a semi-secret assasination squad called reporting directly to vice president Cheney run by the Joint Special Operations Command (JSOC). This could be part of the picture I paint above.

Ingen kommentarer: