ASCOPE Analysis

[pvebber]

I am telling you the Rings will rule

I'll get back to you later tonight on the other thread, but I've never questioned that systems - theory inspired solutions are not very valuable

Back to cell phone forensics - and large scale data mining in general:

What gets scary is on the one hand you have cell phone forensics stuff that undoubtedly has great value when applied to know bad guys, but raises serious questions when applied more broadly to "us law abiding citizens" - particularly when we are not aware of what data our cell phones are collecting:

http://technolog.msnbc.msn.com/_news...phone-tracking.

What constitutes "public information" that law enforcement can acquire without a wrrent, and where does that run afoul of privacy concerns when we are not aware what information devices we carry are making public about us?

This gets really dicey when one considers the likely application of programs like the one in Michagan to collect lots of data on lots of people.It doesn't take rocket scientist to figure out that they are likely doing social network analysis on this data to understand social networks and be able to rapidly create lists of "persons of interest" when a crime occurs.

For those not aware of how the predominant types of social network analysis software works, they typically use "Bayesian Netwok Analysis" that uses proabaility based assumptions about the connections between network nodes and determines the correlation and probability that a member of the network is linked to another person on the network in a particular way.

What is potentially troubling about this is the often arbitrary nature of those assumptions and the proabailities associated them. This was a major part of the downfall of HB Gary executive Aaron Barr in his attempt to profile and expose members of the hacker group Anonymous. Barr used Bayesian Network tools to correlate and find connections between chat aliases and real names and locations. Bayesian methods use a form of "guilt by association" that considers it more probable that one web presence is connected to another based on the temporal and content correlation of web activity.

These methods have a high degree of accuracy , but suffer from a high degree of false positives as well. Barr likely accurately identified some membersof Anonymous with this technique, but also falsely accused a significant number of girlfriends, roommates, officemates, or just random people who happend to use certain leywords in chat at the wrong time.

The extension of this to llink those identified with "co-conspirators" through more coventioal social network analysis leads to a another failure mode that is guareenteed to implicate innocents who happen to be assocaited with the someone they really don't know as well as they might think.

Police are taking techniques that worlk well in the real world - if you are hanging out in a vacant lot with a group of known gang menbers, what are the chances you are part of that gang too - to the web where we often interact with people in online games, social network stes or other web 2.0 services in an extremely narrow context that are difficult for Bayesian analysis to differentiate from nefarious associations.

I support Slap and his brothers in LE to the 9s when it comes to employing these techniques to catch known (or legitimately suspected) bad guys.

I have a HUGE problem however with the idea of the LE community collected large volumes of social network data, particularly in cases where law abiding citizens are unaware of just what data is collected, and using that data to implicate people in criminal activity without any judicial oversight.

There is a huge potential of abuse of number crunching large amounts of data using error-prone statistical techniques that will in effect produce seemingly compelling cases of circumstatial evidence that is statistically guarenteed to result in innocent people being innapproriately implicated in criminal activity.

There are a million stories in the naked city everyday, thus a 1 in a million story will probably occur during each one.