Homeland Security panel at PC Forum

Just saw a presentation at PC Forum from SRD, a company that specializes in rooting out the non-obvious relationships between card-cheats, mafiosi and high-rollers and casino employees and high-rollers.

The presentation was very impressive. In one instance, SRD investigated all 20,000 employees at a casino, along with all the customers at the casino, and discovered a couple dozen cheats who were either playing at the casino — or employed at it. They did this by applying algorithms that fuzzily matched personal information from the "insider" and "crooked" lists.

SRD has taken an investment from In-Q-Tel, the CIA's venture-capital arm, presumably to apply this technology to catching terrorists.

But this begs a question: how can you apply this to homeland security? Do we open an investigation into every American and cross-reference it against every bad-guy? How can we possibly square that with the Constitution?

And what's more, if the result of this investigation is not a criminal charge — i.e., if you get added to a no-fly list (or worse, if you get declared an enemy combatant and denied a court appearance) — how do you clear your name? If the basis for suspicion is that the word-fequency histograms of your public utterances have n points of similarity with our profile of a terrorist, how do you rebut the accusation? What if you never even find out what the basis of this accusation is?

The In-Q-Tel representative, Gilman Louie, is talking a good line, arguing that government shouldn't have access to commercial data, that there is no advantage to throwing more and more and more marketing and other commercial data into an already overloaded analyst's queue.

Esther Dyson says that she doesn't believe that the "Chinese walls" that keep the Feds from digging through personal info are really all that effective.

Gilman Louie from In-Q-Tel says that the real problem is local law-enforcement, who use techniques like sorting all records for Arabic last names — technology makes bad policy worse. Unless we beef up the audit and accountability side of the house, it will get very scary.

I just put my questions to the panel: How can you square investigating everyone with the Constitution? How can an algorithm's oracular pronouncement stand in for due process? Gilman Louie's response was very good. He said that he believed that profiling is the wrong answer to the wrong question, that it should be a tool that's applied after human judgement, not before. He compared profiling to the internment of Japanese-Americans and averred that profiling is a dangerous tool for racial or ideological discrimination.

Though he didn't say it, there's good research to suggest that profiling doesn't work.

That said, profiling is alive and well in America. CAPPS is in use in all US airports. CAPPSII is coming. DARPA's spending money like drunken sailors on profiling technologies.

I think Gilman Louie was being straight, but I also think that's he's not representative of the people who are calling the shots at Homeland Security — and that's bad news for all of us.