"Get your facts first, and then you can distort them as much as you please." (Mark Twain)

Friday, October 08, 2004

No Flight of Fancy 

Google is a wonderful tool - enter a few keywords and get a list of Web sites that feature those key words. The probelm is, you're never sure if you entered the right keywords, the ones that will lead you directly to the one page you most need. I mean, Google could take a very inclusive approach to search engine design, allowing pages that may or may not be responsive to your needs to come up among your results. For example, Google could be designed to include words that sound like your keywords among the results. This sort of algorithm would be more thorough - the likelihood that the one best page is among the results returned would be increased - but that thoroughness would come at a high cost. Specifically, such an algorithm would clearly return more irrelevant, or marginally relevant, results. In other words, you might get a little more wheat, but you would absolutely get a lot more chaff. Paradoxically, the ultimate result would be that finding your best page would actually become less likely, even though it is more likely that your best page is among all those results somewhere.

If you've used Google, it's very likely that nothing I've said here is news to you, or is very controversial.

And so I can only conclude that no one at the Transportation Security Administration has ever used Google. From the Washington Post:
The federal government's 'no-fly' list had 16 names on it on Sept. 11, 2001. Today, it has more than 20,000.

The list, which identifies suspected terrorists seeking to board commercial airplanes, expanded rapidly even though the government knew that travelers were being mistakenly flagged, according to federal records. The records detail how government officials expressed little interest in tracking or resolving cases in which passenger names were confused with the growing number of names on the list.

More than 2,000 people have complained to the Transportation Security Administration. Airlines, at one point, were calling the agency at least 30 times a day to say that they had stopped a passenger whose name was similar to one on the list, but after further investigation was determined not to be a terror suspect, according to a TSA memo.

More than 300 pages of documents related to the no-fly and related lists were released late Thursday night by the TSA and the FBI in response to a federal court order. The American Civil Liberties Union had filed suit on behalf of Jan Adams and Rebecca Gordon, two peace activists who wanted to know why their names had turned up on a no-fly list.

The documents reveal early symptoms of what are now known to be flaws with the watch lists. Travelers who were flagged by the lists said they now foil the system by altering how their names are spelled on their tickets -- adding their middle initials, full middle names or titles, for example....

The TSA acknowledges that the system for checking passenger names for suspected terrorists needs fixing, and it plans to overhaul it in a new program called Secure Flight. The Justice Department declined to comment.

The false matches "underscore the need we have to get more information on passengers to adjudicate those that are not a risk," said Department of Homeland Security spokesman Brian Roehrkasse....

Passengers are falsely flagged by the lists in such large numbers because of the kind of technology airlines use to compare the reservation lists to the watch lists, according to experts in name-matching technology. Each airline conducts the matches differently. Many major carriers use a system that strips the vowels from each passenger's name and assigns it a code based on the name's phonetic sound, according to the Air Transport Association.

The name-matching technology is "too simplistic for a very complex problem," said Jack Hermansen, co-founder of Language Analysis Systems Inc. in Herndon, a company that has a competing name-matching technology that factors in a name's cultural origin. "It's these accidental matches that cause the big problem."

The phonetic-code concept is traced back to a program called Soundex patented in 1918, which was used by Census Bureau officials to help sort out names that sounded similar but might be spelled differently. The name "Kennedy," for example, would be assigned the Soundex code K530, which is the same code assigned to Kemmet, Kenndey, Kent, Kimmet, Kimmett, Kindt and Knott, according to genealogy Web sites that use the technology. Today's systems are more sophisticated than Soundex, but they grew from the same origins, experts said.

"The reason this technology is used is you're really trying to protect against typing errors," said Steven Pollock, executive vice president at TuVox Inc., a company that sells speech-recognition software. "When someone types in a name, the problem and the challenge is people will spell names incorrectly. . . . Names are definitely the toughest things to get [right], no doubt about it."

But the phonetic coding systems tend to ensnare people who have similar-sounding names, even though a human being could tell the difference. Earlier this month, for example, Rep. Donald E. Young (R-Alaska), said he was flagged on the "watch list" when the airline computer system mistook him for a man on the list named Donald Lee Young.

It seems to me that having an airline security system that is broken and doesn't work, that inconveniences travellers unnecessarily (at who knows what cost in lost productivity), that makes it even harder for struggling airlines to compete and make a profit, and that induces people to rely on it when it brings no actual, verifiable improvement in safety, is not much better than having no airline security system at all. It's almost like we screened passenger baggage, but left airfreight loaded on the same aircraft completely uninspected. Wouldn't that be silly?




Post a Comment