Saturday, February 08, 2003

In the humor column, we find that in Powell's recent UN presentation, he fondly praised a "recent" British government report on the threats from Iraq. Turned out that much of that report was blatantly plagiarized, and many years old, to boot. Can't see how this helps Powell's international credibility, but just shows again how strongly the Bushies believe what they want to believe.

Recently read a summary of an interesting criticism of the Total Information Awareness program. I think it was by someone at the University of Texas, but Dubya never did like the school that much. How dare they reject his law school application just because he was a dumb booby? Anyway, obvious enough that I can very easily reconstruct it from memory, so scarcely merits a detailed citation, though I'd have been glad to provide it.

He compared the terrorist detection problem to the medical diagnosis problem for rare diseases, and made the VERY generous assumption that TIA could somehow become 99.9% accurate in the diagnosis of "terrorist". He also made a rather optimistic assumption about the number of possible terrorists to be found, and showed that because of the false positive errors, you'd still wind up with 1,000 innocent people for each terrorist identified. In numeric terms, he estimated 250,000 false accusations. That's simply because there are LOTS of innocent people and relatively very few terrorists, so even a highly accurate test will fail in this way. The summary skipped over the next part, but filling in by guessing, I think he actually projected that the effect of those false accusations would be to create many new terrorists. That's because of the nature of this "disease" and because the diagnosis is not an statistically independent event here. We're already assuming that the system works well, so that means that those 250,000 candidates did share many characteristics with the actual terrorists. The false accusations are at a minimum going to result in investigations, and here is where the independence is violated. Being investigated as a possible terrorist when you actually are similar to a terrorist is quite likely to be the triggering event to push the candidate over the line. In too many cases they're going to find out that someone has been asking questions about them, or maybe in extreme cases they'll get fired or otherwise punished because of the suspicion itself. Since the system has presumably correctly identified them by their similarities, any perturbation is liable to shove them over the line. And heaven forbid that the actual terrorists should somehow get a copy of the list. Talk about your "golden" recruiting hot list!

No comments:


About Me

My photo
As a blogger from before there were blogs, I've concluded what I write is of little interest to the reading public. My current approach is to treat these blogs as notes, with the maturity indicated by the version number. If reader comments show interest, I will probably add some flesh to the skeletons...