Data Mining In Law Enforcement

jcatcw points out a blog entry by Scott McPherson, CIO for the Florida House of Representatives. McPherson condemns the state of data sharing and data mining in law enforcement, saying that the US causes itself a great deal of trouble by focusing more on "antiterror armor and nuke-sniffing devices" than a useful information distribution network. He discusses a few such projects, and how they could have directly affected the events of 9/11. Quoting: "One of those ingenious things that actually worked, Seisint founder Hank Asher's brilliant MATRIX system, remains mired in controversy and politics. Hank showed me MATRIX just a few short weeks after the 9/11 attacks. Using law enforcement data and commercial data, all of the commercial data available in the public domain, Asher's query produced [hijacker Mohamed] Atta's photo -- and about 80 others, many of them fellow 9/11 hijackers, many of them associates of the 9/11 hijackers. It was simple data mining and algorithms, and none of the information was obtained illegally."

Comments

Hold on a minute here

[goldcd]

so he managed to write some software that analyzed the internet - and managed to produce photos of some of the people that erm had already erm been identified. Surely (and maybe I've misunderstood something here) a 'result' would be identifying people likely to commit terrorist attacks, allowing enforcement agencies to monitor them and prevent them from commiting future attacks. (and no - this doesn't mean off-shoring every muslin who downloaded the Jolly Roger Cookbook).

Re:Hold on a minute here

[k1e0x]

Yeah, he wrote software that detects terrorists after they have committed a crime.. Its key component searches google news. heh.

But really. Lots of people *may* commit crimes. Computers may decide you are likely to rob a bank tomorrow, that does not mean you will. We need to make sure the law is always about what you do not what a computer projects your going to do. The day we jail people who *might* be about to commit a crime is the day we put people in jail for their thoughts.

Re:Hold on a minute here

[Alpha830RulZ]

If you assume for a minute that the author of TFA is smart enough to figure out if this was a google search or not, this is probably pretty interesting. I'm going to, perhaps naively, assume that the data mining approach was done as a reasonable experiment of a mining approach on some set of data, and arrived at a set of names that should be interesting to check up up. I'll further assume that he properly restricted his training set of data to only data that was available before 9/11.

If that is the case, this is a pretty impressive set of results. Being able to identify, say, 5 of the attackers, and to have a number of the other hits be known associates, when the training set likely consisted of at least 10's of thousands of names, is pretty fair accuracy. The false positive rate is pretty fair, as well, especially when you contrast it to the No Fly list, which has numerous false positives, and no known successes in identifying anyone of interest.

There is likely some sort of clustering algorithm behind this, and the math behind those is pretty solid. Before you dis this, or even get excited about privacy issues, I'd suggest you check out a reference such as this

I'm not really concerned about data mining as a privacy issue, and I think it's a pretty legitimate approach for law enforcement. As a side note, I do data mining and predictive analytics for a living. It's objective, it's factual, and if the practitioner is knowledgable about it, it shouldn't be stigmatizing. Indeed, it would reduce scrutiny on the majority of the folks that would otherwise be tarred by having an arabic surname and swarthy skin.

It would have the potential to be vastly more effective, and vastly less expensive than the path we are on now. One reason that we might not be using could be that we -have- used it, and didn't find anything. That's the thing about objective data mining, if there is nothing there, it'll tell you that. I don't think, for our current administration, that it's a desireable outcome to find that there is nothing to worry about. If that happened, the populace would be less fearful, and less easy to control.

Take this one step further, and apply this bit of thought. It has been shown time and again that the TSA is incompetent, and that any motivated terrorist could get a weapon on board a plane. It is further obvious that our ports are porous, and that soft targets abound. We have seen no triumphant pictures of the authorities frog marching attempted terrorists away, no success stories of how these measures have saved our lives again. We have also seen no further attacks.

This strongly suggests to this practitioner that we have a near zero incidence rate of terrorists in the US; that when a terrorist attempts an attack, he succeeds, and that the lack of attacks suggests that the attack rate is close to zero.

Data mining would be a useful tool to calibrate this theory.

Hindsight is 20/20

[garcia]

Wow, really? You were able to identify after the fact? Great! Real useful -- that and the fact that it's much easier to find that information when you are looking for a specific result. If this guy had come out and said, "hey, I was able to find those people before the fact," then I'd be impressed.

Re:Hindsight is 20/20

[Chris Burke]

Yeah, I've got a mother-fucking perfect Suicide Bomber detector. It never fails. 100% specificity, 100% sensitivity. Here's how it works (it's patented, so my lucrative business is not in danger by sharing my methods):

I stand around a marketplace in Baghdad. When a guy runs up to a crowd, screams "Allah Akhbar", pulls a string on his coat, and fucking explodes all over the place, I point at the spot where he used to be, and say "That was a suicide bomber".

And before you try to horn in on my business, know that I've already sold the DoD enhancements to my algorithm that covers cases where the bomber doesn't scream "Allah Akhbar", or where the bomber is a she not a he, or where the explosives are in a car not a coat. Or combinations thereof.

But seriously, it says that "his query" produced Atta's photo (and 80 others only some of which apparently had anything to do with 9/11). What exactly was this query? "9/11 hijackers"? "terrorists named Atta"? "Arabs who've been pulled over"? So Atta's driving citations means it was theoretically possible for someone to pull his name up. The question is, why would they have done this? What would have motivated someone to perform that query, and how exactly does data mining driving citations lead to the important conclusion that Atta was a terrorist?

The article makes good points that data sharing between law enforcement agencies is a good thing, and helps with such rather mundane things as finding fugitives who skip out on parole, or people who don't show up for court dates. But that MATRIX nonsense is yet another attempt to cash in on post-9/11 anti-terror funding bonanzas. Which, now that I've gotten my slice of the pie, I'm against. :)

Re:Or not

[shmlco]

Well, as in many things it would seem that there's a loophole or two involved. While there are many restrictions placed on government in terms of data collection and data mining, there are few placed on individual businesses who do the same thing (think credit agencies). As such, there's little stopping the government from simply contracting out its needs to private companies.

Bad news actually

[iamacat]

The same techniques will likely be effective for identifying most effective protestors against current administration, or people that can be most effectively exploiting sexually, financially or politically. In fact, terrorists generally cover their tracks much better than innocent civilians.