Considering the TSA and the Anti-Martyr Problem

Well, the TSA objective of making transportation safe is back on the front-burner. Now the TSA screening is a poor seive. It is a largely static target and is very costly, the largest cost of course is in the lost time that travellers endure in negotiating long security lines. Furthermore, it is likely that much of their efforts are counter-productive. For example, making box-cutters freely available and common on flights would make it harder, not easier, for a terrorist or terrorists to hijack a flight. The “rules” of engagement with those who would interfere with the operation and direction of airplane do not get time to negotiate or to “make demands” known like they might do in the 20th century. Once a person is identified as hostile (a prospective anti-martyr) that person is quickly neutralized by his fellow passengers. The age of passive passengers has past once the 9/11 event occurred.

However TSA has a purpose. It is visible and reactive. It can take the appearance of being the primary and front line defence in a strategy to identify and interdict prospective anti-martyrs. War and espionage (to which this anti-martyr interdiction campaign is related) is in part one of misdirection. To that end, the TSA screeners take a very public and obvious role. They (might) be the public and obvious strategy which is a counterfeit. If indeed the TSA plays such a role, we as the voting public will not know that for as soon as it is common and public knowledge that the TSA is a large noisy feint … then their will be an outcry to remove it and an alternate deception will be harder to enact.

From the public side of the anti-martyr interdiction campaign it is not clear whether the information problem, one of using network and other aspects of information theory and methods to sift the wealth of imperfect information to reduce the 6 billion suspects (people on the planet) to the correctly identify and interdict the hundred or so individuals who are likely earnestly planning and working to execute an effective plan. Is this a tractable problem? If it is, then only those working in the anti-martyr networks and those working against them would know of the existence and number of thwarted events and actors. If it is semi-tractable, i.e., individuals can be identified but some percentage are missed as is likely the case then what is most needed are more tools at our disposal legal, human and technological to enable identification of these individuals and networks.

In the past, I’ve argued that automated wiretapping methods which preserve our liberty via court and/or congressional supervision is possible. In brief allowing court supervised search engines access to voice and data networks to identify and locate conversations and data that is interesting. This data could then be presented to a human for review with any identification (source/destination) data not available for initial review. After a message or conversation is identified as being of interest a court warrant could be obtained to access the full record, which at that point only include what was known of the identity, location and time of the message. It is likely that today’s computing power is not sufficient to transcribe and analyse in real time the data flowing through our phone and network switches.

However this problem is likely not too different from the problem facing the detector arrays at CERN or Fermilab. At a modern high energy particle accelerator, particles collide at several points on the circle. At these points there are large detectors and calorimeters in place to analyse the collision events. These collisions produce many orders of magnitude more data than can be assimilated. Therefore the detector software has to filter early in the data processing procedure what might prove to be interesting event. What is “interesting” is driven of course by expectations, i.e., what is being sought.

By analogy the computational problem facing the analysis of voice/data switch data may likewise become tractable. If given entry points (data and voice) as well as key phrases, accents, and languages are part of the early data rejection/acceptance pipeline and not just a back end (post transcription) process then like the CERN/Fermilab detector situation the data overload may become more tractable.

The civil rights issue in place of using such a mechanism domestically is not the one, in my view, that is normally assumed. It seems to me the major objection to this mechanism is one of an invasion of privacy. The fact that a non-moral agent scans conversations is not a problem of an invasion of privacy. The problem is that in the hands of a government that desires to oppress, this tool makes insurrection impossible. That, not privacy, is the concern. Whether or not such a mechanism is in place today is not one to which we will likely be privy for some time, decades perhaps. That’s a judgement call that will be made by small Senate and executive committees. Which might be why remarks like, “never fail to make use of a crises” from our Administration and an external perception that the left regards the anti-martyr problem smaller than the right did … could possibly become a problem.

Leave a Reply

Your email address will not be published. Required fields are marked *

5 comments

  1. Boonton says:

    I think you put too much faith in database mining. Part of your problem is that you are only looking at one type of terrorist. Over on McArdle’s site, I presented the following taxonmy of terrorists:

    ****************************
    http://meganmcardle.theatlantic.com/archives/2009/12/oh_dear_3.php

    a. Local insurgents – fighting some local battle in places like Israel, Pakistan, Afghanistan, Iraq etc. using terrorist tactics to win on the ground with little interest beyond their limited battlefield.

    b. International organizations- Like Al Qaeda seeking to attack countries around the world either by using terrorists to travel to them or finding locals willing to mount attacks.

    c. “Spontaneous Jihadi Syndrome”- Like the Fort Hood attacker who either because of mental illness or radicalism seek to become terrorists. They may or may not make contact with overseas groups. Whats important about these types is that they may snap without any warning. There may be no phone calls or emails coming from ‘headquarters’ to them. They may simply decide to ‘take up the cause’ with no direction from outside their own minds.

    d. “Random crazies” – You may recall a little while after 9/11 some teenager took his father’s small plane and crashed it into an office building. No one was there at the time and he ended up accomplishing nothing more than killing himself and ruining someone’s corner office. Some terrorism will come from random crazies who want to blow things up, go out in a blaze of glory and so on. (See also the Waco incident….)

    e. Unexpected terrorists – While right wingers like to litter the comment boards demanding ‘profiling’ of Muslims boarding planes, the fact is we have two major non-Muslim terrorist attacks in recent history. The first was Tim McVeigh, a white guy who appeared to have little backing except by one or two buddies. The other was a doomsday cult in Japan that used its scientist members to create nerve gas and injur hundreds in the subway system. Go back a little more and for decades there was a terrorist conflict in Northern Ireland that had nothing to do with Muslims and Latin America has a low level war going with narco-terrorists as well as Marxist oriented terrorists who are running a business of kidnapping and ransoming people.
    *************************************

    At best what you describe addresses b and c types. Even c types, though, won’t be so easily detected unless they first make contact with known terrorists abroad or openly discuss their plans online or on the phone. The detectors at CERN or Fermilab get a major break in their datamining challenges, the particles aren’t trying to outsmart the scientists.

    By definition, the data mining you’re talking about will always be backwards looking. It assumes the terrorist this week will look like the terrorist of last week. Unfortunately this system is easily fooled by making the terrorist of this week appear different from the one last week. We have been lucky that Al Qaeda has not been all that bright in varying their profiles. They always seem to be young, single, Arab men who for some odd reason tend towards college degrees in engineering. But I wouldn’t hold out hope that terrorist organizations will remain dumb for very long. In Israel, for example, female suicide bombers were unheard of. The profile happy advocates were caught blind when females started being used for suicide bombings. Going forward, how hard would it really be for Al Qaeda to be more watchful for ‘flags’ by buying round trip tickets with credit cards, checking some luggage for long flights etc.

    Here’s the other reality, look at data mining and credit card companies. They mine data like crazy to see who will become a deadbeat. Yet for their clear financial incentive they still end up losing in many cases. Credit cards, though, can live with limited losses. Critics want the TSA to achieve 100% perfection with data mining and the fact is it isn’t going to happen. For no other reason than you are at best only capturing type b terrorists and even then they will adapt in order to dodge your profiles.

    Now take another peek at the supposedly silly and annoying systems the TSA has put into place. By having random pat downs, inspection of carry on bags and making everyone take off their shoes the TSA has dramatically limited the possible methods of getting serious weapons on board planes. This is why the terrorist had to stuff the explosives down his crotch. This raises the cost of attacking an airplane.

    1. Since you can’t get much explosive on board, you have to use higher tech explosives that generate more bang per ounce but at the same time you can’t use anything that sets off the metal dectors.

    2. Since you have to stuff it in your crotch, getting it out and setting it off is clumsy and offers a higher chance of detection.

    3. Since the inspections are either universal or random (for the more intense checks), there is no getting around profiles.

    Here are the facts, twenty years ago you could blow up a plane by not even getting on board. Check some baggage with a bomb but don’t get on, that’s no longer an option. Then you could blow up a plane if you were willing to die yourself by bringing a bomb on your carry on luggage. Now you have to find high tech explosives, be willing to die and even then you may only damage the plane.

  2. Mark says:

    Boonton,
    Of your categories, “a” is not related to domestic terror and the rest are susceptible to intelligence analysis or “data mining.” Your assumption seems to be that the parameters of the search of fixed in stone and not adaptable to intelligence gained or changing events is somewhat a problem. As for locating a McVeigh type actor, credit card companies while allegedly not being able to detect defaulters (perhaps they do, perhaps not … it’s likely that they do better than economists at predicting trends. 😀 ) … but those same companies do a constant and often successful effort at combating fraud. You’ve likely gotten calls when you make an “out of band” purchase. A search of buying patterns might have gotten a McVeigh for purchases of large quantities of fertilizer with an absence of other agricultural purchases. A little further checking would have put him under surveillance.

    On the last, I think you are wrong. My bags have certainly travelled on planes that I have not. I’ve never seen a plane halted and bags unloaded and searched because a traveller checked in but failed to board. Have you? As I noted, I did not recommend stopping TSA but that it should not the primary line of defense … but that it should in fact be seen as such. In order to be seen as such it has to be credible. I didn’t connect the dots there on what that means for what TSA should be doing, but I’m guessing with a moments reflection you could.

    Unless the “random crazies” or the people of category “e” don’t talk to anyone at all and are complete loners there’s going to be evidence for network analysis. The question is it the data stream too large to be tractable or not … or what it would take to become tractable … as well as how other intelligence information could be used to tighten the search parameters in a realistic way.

  3. Boonton says:

    I might be wrong about the bags, nonetheless I do believe there is a chain of custody for checked baggage so it is at least x-rayed and/or searched. I also believe that when a passenger is kicked off a plan for unruley behavior his baggage also has to be removed. The gist is what happened to the Lockerbie Scottland plane in the 80’s, a bomb in the checked baggage, can no longer happen today (we hope). Right off the bat if you intention is to blow up a plane you need to find someone willing to die himself. While that doesn’t eliminate the candidate pool, it does dramatically decrease it. Again we are talking about layers of security here, not a hypothetical Chinese Wall that is the only defense because it is unbeatable.

    Of your categories, “a” is not related to domestic terror and the rest are susceptible to intelligence analysis or “data mining.” Your assumption seems to be that the parameters of the search of fixed in stone and not adaptable to intelligence gained or changing events is somewhat a problem. As for locating a McVeigh type actor, credit card companies while allegedly not being able to detect defaulters (perhaps they do, perhaps not …

    “A” might also be suspectible to data mining. The point here is not that data mining can never be helpful, it’s that it isn’t the catch all that its defenders make it out to be. The big problem is that there are very few terrorists relative to the general population.

    A related example might be serial killers. They do have a particular profile, they seem to be overwhelmingly white men. I believe there’s one documented case of a white female serial killer and maybe one black male. Yet this is useless to give to a traffic cop as a ‘profile’. The cop could pull white men over every day for a 20 year career and never find a serial killer. While ‘white male’ is an excellent profile for most serial killers, so few white males are serial killers that using it too much would overwhelm your system with false positives that waste time, money, resources and ultimately harm more useful potential sources of info.

    For example, in this case I think it’s very important to note that the primary and only source of helpful info we had was this guy’s father who felt enough trust in the US to try to warn us that his son was up to no good. I think having even 5% of the Muslim population feel that way would be a thousand times more useful than your best hypothetical data mining system looking at credit card receipts and blog comment posts to try to guess who will be the next terrorist. This relationship type intelligence, though, seems to be what too many commentators on the right have no respect for and are far to willing to toss overboard with the usual “let’s anal probe every Arab guy who comes to an airport” rhetoric whenever anything happens.

    I’m a bit surprised at your newly discovered faith in government here. Look Google is probably the best datamining company ever to walk the earth. Netflix does pretty good too. Despite this stunning private sector effort, the fact is I still have to dig through the results more often than not to get what I want and Netflix only occassionally wows me with a really good recommendation.

    Unless the “random crazies” or the people of category “e” don’t talk to anyone at all and are complete loners there’s going to be evidence for network analysis.

    Let’s add to the ‘random crazies’ list the anthrax killer whose motive appears to have been devotion to the pro-life cause (although no one has ever explained why his first target was the National Enquirer). Maybe ‘network analysis’ might uncover cult type groups planning a terrorist attack but the ultimate problem here is that terrorism is called asymetrical warfare for a reason. It is about individuals or small groups using a very small attack to inflict a massive amount of damage. By definition you are trying to micro-predict the behavior of individual humans and not groups or averages of individual humans but all individual humans.

    While sociology doesn’t have a uncertainity principle like physics, one certainly exists. A lot of mass behavior remains unpredictable. Why some movies take off, some songs become hits, one funny YouTub video goes viral while another doesn’t remain unpredictable. A data mining system is not going to predict the next terrorist to strike but it can give him cover by shifting our eyes towards a predetermined ‘profile’ that a terrorist will exploit by avoiding. Type B terrorists already have an advantage here since they and their supporters spend hours debating which strategies work. It is known that Osama bin Laden stopped using satellite phones because he learned their locations can be pinpointed. The data mining approach, by definition, is ALWAYS backward looking. It always requires a series of terrorist attacks first in order to build a ‘profile’ of the next terrorist to strike using the assumption that the next terrorist will look like the last one.

    My objection here is not to data mining but to the unrealistic faith people like to put in it. Exercises where critics say the dots should have been connected always happen in hindsight and never are examined next to how many false leads and dead ends would be generated by using that ‘dot connecting’ method.

  4. Mark says:

    Boonton,
    Serial killers are the example of a terrorist organization that would indeed be hard to track, a lone person acting carefully in complete secrecy. However, terrorists are inherently political and ideological, which lends them to networks and groups. Also, they tend to want to

    A related example might be serial killers. They do have a particular profile, they seem to be overwhelmingly white men. I believe there’s one documented case of a white female serial killer and maybe one black male. Yet this is useless to give to a traffic cop as a ‘profile’

    This is indeed a good example of why a single dimensional search parameter is particularly useless. Create a stupid search and you will indeed “pull over” people for 20 years not finding anything.

    I think having even 5% of the Muslim population feel that way would be a thousand times more useful than your best hypothetical data mining system looking at credit card receipts and blog comment posts to try to guess who will be the next terrorist.

    I agree, which is why I talk about search parameters that include the entire intelligence apparatus as a whole and not just one facet.

    You talk about profiles. I wasn’t.

    The data mining approach, by definition, is ALWAYS backward looking.

    I disagree.

    Exercises where critics say the dots should have been connected always happen in hindsight and never are examined next to how many false leads and dead ends would be generated by using that ‘dot connecting’ method.

    Actually what I was saying is that we have no idea how many attempts have been foiled and stopped.

    Look at the recent example. His dad made a report. That would mean that the search parameters would flag anything connected to him. Which would have already been in place, because of his connections to ex-Gitmo individuals. Ultimately other connections, like his buying explosives and so on would make him a “hot” suspect. So … when he got an airport a “TSA” operative should be giving him a “random” much more intensive search. That way the nominal TSA front gets the credit for the stop.

  5. Boonton says:

    However, terrorists are inherently political and ideological, which lends them to networks and groups. Also, they tend to want to

    Only type A and B terrorists. The anthrax killer did not seem to talk to anyone and even his targets seem only loosely related to his ideological cause of anti-abortion. Likewise Tim McVeigh seems to have done little chatter on cell phones or web sites, perhaps hatching his plan with his buddy over a beer in the backyard. Individual people are capable of vast amounts of craziness. You could end up with a terrorist blowing up Catholic Churches because he considers himself super-pro-life and thinks the Catholic Church isn’t doing enough! To cover the field you must assume terrorists will be hyper-rational, hyper-irrational and in between.

    Furthermore the weakness of networks is also their strength. While networks might find it hard to keep all their communication ‘off the grid’ (think of the movie Goodfellas and Ray Liota’s bad habit of talking on the phone even though he knew better), they also work better than ‘lone wolfs’ at reverse engineering security systems, coming up with novel ideas to short circuit profiles and so on.

    This is indeed a good example of why a single dimensional search parameter is particularly useless. Create a stupid search and you will indeed “pull over” people for 20 years not finding anything.

    Granted but many of the other serial killer profiles are still of limited use (think of a profile that can be assembled from databases, not in depth psychological studies of individuals). There is no simple profile that can sweep through the millions of white men and come up with a list of a thousand or so who may contain the few hundred of at large serial killers. Likewise we have a similiar issue with Muslim terrorists. We aren’t looking for tens of thousands per million or even hundreds….all told we are maybe looking for ten per million Muslims in the whole world and those ten per million are responding and adjusting to our search methods.

    You talk about profiles. I wasn’t.

    I think they are interchangeable here. Basically you are asking computers to search huge databases and detect patterns that either seem to match previous terrorists (with minimal ‘false positives’) or try to proactively detect ‘terrorist behaviors’ (say a person who buys fertilizer and bulk quantities of perioxide from a hair product supply house). The first type of profile is more sophisticated than the radio talk show call in guest “all Muslim men” profile but it is still essentially backward looking assuming tomorrow’s terrorist will look like todays. The second is almost certain to generate a lot of false positives which means you must investigate your leads very gently (perhaps the guy is taking up farming and has a wife studying hair styling) and consign yourself to the fact that you’re net is going to let a lot of fish slip through before it does its job and catches someone.

    Look at the recent example. His dad made a report. That would mean that the search parameters would flag anything connected to him. Which would have already been in place, because of his connections to ex-Gitmo individuals. Ultimately other connections, like his buying explosives and so on would make him a “hot” suspect.

    Unless I’m mistaken the only thing that was an actual data point was his dad making a report less than a month before he got on the plane. He brought his own explosives? And you buy explosives in Yeman from a place that puts your name in a computer database!!!! I suspect even a well designed data mining system would have simply set him aside for a more intense pat down….which would have had a high chance of missing the explosives stuffed in his crotch. Of course you could make the ‘no fly lists’ much more sensitive but then you open up the reverse problem of Al Qaeda or other trouble makers swamping the system with false ‘reports’ in order to inflate the no fly list and make it unworkable.