Something for that growing pile of bad ideas...
News.com: "Tested in Russia, the two-stage GK-1 voice analyzer requires that passengers don headphones at a console and answer 'yes' or 'no' into a microphone to questions about whether they are planning something illicit. The software will almost always pick up uncontrollable tremors in the voice that give away liars or those with something to hide, say its designers at Israeli firm Nemesysco.
'In our trial, 500 passengers went through the test, and then each was subjected to full traditional searches,' said Chief Executive Officer Amir Liberman. 'The one person found to be planning something illegal was the one who failed our test.'"
So, let's take one of the more stressful environments and have a robot determine if there are any stressed people lurking about. This is a bad idea. What about people who don't speak English? What if your flight is leaving the gate? What if you have bomb, are on Zanax, and you're okay with that? Will they test children? Deaf mutes?
Once a massive voice-stress analyzing infrastructure is in place at travel choke-points, will they record and retain answers? For how long? Will they be tied to your permanent record? Will this become your biometric id? What other questions will law enforcement inject into the system? Have you ever done drugs? Have you ever had an abortion? Are you a legal alien? Do you own a firearm? Do you ever get angry? Do you agree with the president?
Yech. Will this really make us more secure?
Probably not. A bit of Googling on the accuracy rates of voice stress analyzers gives a high of 98% accuracy, and a low of 'worse than coin toss'.
Okay, let's give benefit of the doubt. 98%. Check. According to JFK's own web site, 4,107,950 people traveled through the airport in August 2005. With a 98% accuracy rate, the system would have incorrectly identified 82,159 passengers - as either lying when they were not, or not lying when they were.
Assuming the false positive rate is much much higher than the false negative rate, 2,739 people per day would have been incorrectly identified as a threat requiring greater scrutiny, bag searches, and one-on-one questioning.
This decreases security by diverting sensitive human eyes, ears and time away from real threats, and points them at thousands of innocent people. Screeners adjust to the the system and assume most people are harmless (miss the extra 'the' in there?) - it becomes routine wand waving. And there's no one left to look at people who may actually be a threat, but who happened to game the robot and certify 'clean'.
Cash wasted. Lines jam. Security decreased. And that's the best case scenario.. Never mind the coin-toss option...
Update: Bruce Schneier has picked up the thread here (linking back to this post - Cheers Bruce!)
2 Comments:
The news article I've read on this state as much as a 12% false positive rate. That would be a huge number of required security actions. Makes the use of the GK-1 even more problematic.
My first thought was similar to what you state with what questions are allowed to be asked. I think that there are plenty of questions that could easily be questioned legally.
Then there is the thought of control of the information gleaned from this tool. What if you're found not to be a threat, does that mean that your name and information is retained in a government database? What is the allowed use of this information?
I've sent an email to Schneier's blog for his opinion of the topic. If he comments on topic, it should have some interesting commentary there.
9:55 AM
The false positive/false negative rates for "lie detectors" are all over the map, but the most reliable data indicates the proverbial coin toss. The bottom line is that most rely not on the technology, but on the effect it has of encouraging confessions.
Lie detectors are snake oil, and as you noted, will most likely make things worse, not better.
1:11 PM
Post a Comment
Links to this post:
Create a Link
<< Home