Slashdot Mirror

This is called Computer Voice Stress Analysis (CVSA). It has been largely discredited as being less effective then polygraphy (I think).

I wish I could provide a more authoratative link, but it's difficult. The entire field of lie detection is so buried in political bullshit that it's almost impossible to tell what is and is not effetive. Every study is from someone on the take, and every cited study is funded by the citer. For example, take this from the American Association of Polygraphers with, surprisingly, a comprehensive listing of why CVSA doesn't work.

As I understand it, from people I talked with involved in security at my previous government job, pretty much all lie-detectors and methods is 90% psychological and 10% actual. In other words, having the subject believe it works is where 90% of the effectiveness comes from. For specific situations (ie, did you kill Jim-bob), polygraphy seems to be far more effetive then CVSA. For general-purpose ("Have you ever done anything bad?"), all forms of lie-detector are suspect, at best, and very much a voodoo-science. CVSA's lure comes from the fact that it's cheap and easy to train people to use, and less intrusive (requires less calm enviornment). However, it's far less effective then polygraphy, and it's primary function is to give the interregator a psychological advantage, and no more.

Amir Lieberman, the developer of the system, is also responsible for the previous rash of questionable truth detector software, which happens to be still available. It did recommend training, and it was widely sold for its ability to work over the phone. It even has a sequal. (warning, Not compatible with Opera. Probably not Mozilla.)

Namesysco doesn't claim very high accuracy for the Truster software. "The voice analyst achieved an overall accuracy rate of 78% for truthful subjects and 61% for deceptive subjects." In other words, only 10% more liars were caught than flipping a coin, while 22% of innocent subjects were considered lying.

The American Polygraph Society does not have a much rosier view of the situation. They have concluded that Computerized Voice Stress Analysis, and specifically the Truster software, has only a "chance-level detection of deception,"

And actually, the dead giveaway to the scam should be from the lion's mouth himself. "Our products were originally for law enforcement use ? we get all our technology from Nemesys-co ? but we need more development time [for that application]" In other words, "our products don't work and can't be sold unless you slap a 'for entertainment purposes only' label upon them. Our products are to 'entertain' airport security."

Good catch.

Polygraphing is a given hot topic, there are zealots in both proponent and opponent camps. I find it diffucult to find an objective source of information on the topic and its accuracy.

Antipolygraph.org has a link here

and the American Polygraph Association has a link here

George W. Maschke
AntiPolygraph.org

All bets are off when it comes to grand juries. :) Their investigatory pwoer is amazing, as Ken Starr demonstrated. But at least they only do indictments.

I just rechecked re admissibility and am unhappy to see that in the last few years federal courts are reconsidering the per se exclusion of polygraph tests, largely because of a Supreme Court case called Daubert. States have a variety of rules, most very hostile to polygraph.

I didn't know about the Act -- thanks (I see that it has other exceptions). There has been a lot written about the increasing use of the polygraph as a gov't screening measure, despite its multiple failures. I'm concerned that little effort seems to be going into measures that really do advance security.

"This is the first technology that allows lying to be measured or lying to be detected without any contact with the subject whatsoever instantaneously, in real time," said lead researcher James Levine, an endocrinologist at the Mayo Clinic in Rochester, Minn. "You don't need to hook them up to anything -- you don't need any sophisticated experts to analyze the data."

Remember, stress is a matter of the body, but a lie is a matter of the mind. They're correlated in many people, but by no means identical. Just think, do you know any smooth-talking liars (i.e. ones displaying minimal stress)?

Sig: What Happened To The Censorware Project (censorware.org)