Newly Declassified FBI Docs Reveal Predictive Data System

An anonymous reader writes 'Newly declassified documents show that the FBI is developing a data-mining system to uncover terror sleeper cells. Among the 1.6 billion records in the National Security Analysis Center — tens of thousands of travel records, including hotel and airline records. Other revelations in the documents uncovered by a Wired.com FOIA request show that the feds want to expand the system for use in cyber-crime investigations, and it's already been used to scrutinize helicopter pilots and Philly cab drivers. The system has eerie resemblances to DARPA's once-banned Total Information Awareness program."

Sounds familiar...

[Captain Centropyge]

Reminds me of the crap the DHS is pulling with gathering travel information... [slashdot.org]

Today, do something out of the ordinary

[HalAtWork]

Do something out of the ordinary, once or twice a day. Deviate from your normal routine in very absurd and unusual ways for no apparent reason.

False Positives

[codeAlDente]

And how much does it ruin your life if you come up as a false positive?

A better solution

[Anonymous Coward]

A better solution: offer anybody who's a member of al Qaeda $10 million to knock it the fuck off.

Judging from the CIA's released estimates of membership, we'd wind up a couple billion ahead at that rate. That's what I call a "free-market solution"!

Re:Today, do something out of the ordinary

[bertoelcon]

I actually did something like that on facebook by giving incorrect feedback on ads and becoming a "fan" of stuff I hate just to see how hard it was to screw up the recomendations.

It is actually harder than you might think.

Re:Give up?

[wellingj]

This is the US. The 50 citizens who were previously considered 'cooks' will take careful aim with their semi-automatic weapons before the cops realize the violence they have instigated. It happened in 1776, I hope it doesn't happen in the US again, but it very well could.

Don't worry... I'm probably already on the list. But you won't be, you displayed 'civility'

Re:Today, do something out of the ordinary

[martas]

wouldn't help, unless you did something unusual regularly, i.e. usually did something unusual. for example, if a lot of people regularly acquired material necessary to build a bomb, created fake identities, got on a plane with said bomb and identity, and then didn't blow up the plane, now that would confuse them. your suggestion would just slightly increase noise in the data. hiding trends through noise is much harder than hiding trends through bias, i.e. things that look like threat in many ways, but aren't.

Re:Give up?

[Anonymous Coward]

I would agree - except that, for no reason I can fathom, I was added to the no-fly list. Considering the nature of my work, I was probably a false positive (unless I'm doing some really cool stuff when I sleepwalk). Given sufficient data and algorithms to process it, you *will* have some percentage of false positives. At some point, there will be adverse consequences for those people. Knowing the creeping nature of this kind of capability, imagine the effects: getting pulled over more frequently because my license plate turns up as a potential terrorism suspect. How do you think the cops will treat me? Additional screening at the airport and other security inspection opportunities. Failure to obtain or maintain a security clearance - making it nearly impossible to hold a job in certain areas of the economy.

Yes, this thing will crush itself under its own weight, eventually, just as the no-fly list finally did (that's how I got removed, when they finally went back through it). And at some point, perhaps, it will be refined into a data prediction tool requiring manual analysis of the results, the cost of which will be shocking. In the meanwhile it will ruin many lives.

What's a good number for false positives on a detection algorithm? 1 in a million? That's 300 or so Americans, undeservedly suffering. What's a likely number, once you correct for the really important thing - no false negatives (actual terrorists who are undetected, which will result in the algorithm being changed to flag a detection more frequently)? Lots more than that - *lots* more. And as the detection set size grows, the number of people linked to those people grows as well, an exponential rise in the number of "persons of interest" screwed by the system.

Those of you who have had to fight with an institution - a school, a large business, a government - know how hard it is to win that fight. A lot of innocent people will pay dearly for this.

Re:so ?

[TapeCutter]

"I am sure there has been at least one nearly successful action in the US since 2001 that is utterly classified because it would tend to cause a panic"

Why would they hide panic inducing information while simultaneously filling the airwaves with panic inducing disinformation?

Sounds Like Vestigal AI To Me...

[Xin Jing]

I remember hearing a comment back after the 9/11 attacks that the FBI database couldn't be searched like Google provides it's search queries. From that standpoint of modernization and capability, I say cheers to the FBI for making such a rebound (smells like Carnivore) 8 years later. Interestingly, or rather unsurprisingly, "The FBI declined to comment on the program."

Now on to the AI accusations.

"That could change if the FBI gets it hands on the data sources on its 2008 wish list. That list includes airline manifests sent to the Department of Homeland Security, the national Social Security number database, and the Postal Serviceâ(TM)s change-of-address database. There are also 24 additional databases the FBI is seeking, but those names were blacked out in the released data."

The results of such a query aren't too far off from that of a true prototype AI, which in it's operationally completed state would provide the best prediction bang for the buck there ever was in the history of mankind. And how best to employ that fledgling AI but in law enforcement pursuit of known terrorist criminals.

Where were they, what did they do and where are they now?

Re:I've got an idea

[TheRaven64]

Even those of us in English-speaking countries run into problems. Want to refer to Euros, Pounds Sterling, or Japanese Yen? Sorry, none of those characters are accepted. You can type them, but Slashcode will mangle them (after submission, not always after preview) unless you remember to use HTML entities. Want to mention the app-switching feature on OS X? I'm afraid you'll have to mangle the name to Expose, because this is what Slashdot displays if you type it correctly: Exposé.

There were two reasons for blocking unicode (which used to work). One was people signing up with names that were almost the same as other users but with similar glyphs. That is trivial to fix; only permit ASCII for usernames. The other was people putting things like the right-to-left reading order character in the middle of their post, inverting the order of the rest of the page. That is easily fixed by a simple blacklist on the range of unicode characters that controls formatting.