ANDREW KANTOR   More Columns Read previous columns that go behind the tech headlines in Andrew's archive.

Television butchers technology. It has to. There's no time in your average crime drama, say, to go into the details of tracking a cell phone or running someone's DNA; what takes seconds on TV often takes hours in the real world. (I understand that they've only got 42 minutes to work with, so I don't sweat the time thing.) It's when they deviate from reality so much that my goat gets got.

For example, I gather I'm one of the few Americans who doesn't watch 24. I stopped after the first episode of the first season, when Jack Bauer (that's Kiefer Sutherland to you and me) demanded of his geek-in-residence, "I need every Internet password associated with this phone number!" And he got it. It just doesn't work like that.

One thing occasionally comes up on TV that I wasn't sure which category it belonged to—"accurate but glossed over" or "totally wrong." That was polygraphs. Lie detectors. On TV, the suspect is hooked up to a machine that traces a dozen or so lines on scrolling graph paper. When she tells the truth, the lines barely move. When she lies, the lines vibrate like a seismograph in an 8.0 earthquake.

My guess was that in the real world a lie isn't quite so obvious, but that a trained examiner would be able to spot it right away.

Boy was I wrong.

It turns out that polygraphy is not only an incredibly inexact science, but that reading the results of a lie detector is almost entirely subjective. In short, lie detectors don't work. But people's lives have been ruined by them.

The problem isn't that the machines don't record something—they do: heart rate, respiration, sweat-gland activity, and so on. But what the changes in those numbers mean is entirely up to interpretation.

Obviously there are two sides to this. On the one you have the American Polygraph Association (APA) and some law enforcement agencies. Neither of these are what you would call objective. The APA's existence depends on people accepting that lie detectors work, and as we've seen all too often, police and district attorneys are happy to have something that appears to provide evidence that can convict someone. (Lie detector tests are rarely allowed as evidence against someone in court — that should tell you something — but "failing" a test can sway public opinion, and jury members aren't hermits.)

On the other side you have, well, dozens of groups and organizations with names like AntiPolygraph.org and StopPolygraph.com. If you're like me, you might at first blush think these are fringe groups with their own (hidden) agenda, and that they aren't about to provide unbiased information. After all, there's always a conspiracy theorist to be found.

Except that also on the anti-polygraph side I found the American Medical Association, the American Psychological Association, and 60 Minutes. They all found essentially the same thing: Lie detectors show what the examiners want them to show.

In 1986, 60 Minutes demonstrated this rather dramatically. Using Popular Photography magazine as a front, the producers hired several polygraphers to help find someone who had, they were told, stolen hundreds of dollars of photographic equipment. (No such theft had taken place.) Each examiner was told that a different one of the 'suspects' was probably the guilty party.

Lo and behold, each polygrapher fingered the suspect they were told ahead of time was probably guilty. Oops.

An American Medical Association expert testified before Congress that "the [lie detector] cannot detect lies much better than a coin toss." Further, an article published in the Journal of the American Medical Association by the AMA's Council on Scientific Affairs said in part, "Though the polygraph can recognize guilty suspects with an accuracy that is better than chance, error rates of significant size are possible." Ouch.

A 1997 survey by the American Psychological Association found that psychologists feel that "The use of the polygraph (lie detector test) is not nearly as valid as some say and can easily be beaten and should never be admitted into evidence in courts of law." Eek.

In a briefing paper from 1996, the ACLU wrote: "Despite the claims of 'lie detector' examiners, there is no machine that can detect lies with any degree of accuracy. The 'lie detector' does not measure truth-telling; it measures changes in blood pressure, breath rate and perspiration rate, but those physiological changes can be triggered by a wide range of emotions such as anger, sadness, embarrassment and fear."

THREE KINDS OF POLYGRAPH TESTS Only on television do polygraphers simply ask "Did you kill your wife?". In the real world the tests are complex; different tests ask different kinds of questions. The most common form, and the one that no one of consequence seems to think is valid, is called the Control Question Test (CQT). It compares a suspect's reaction to relevant questions ("Did you strangle your boss?") with questions relating to possible prior, um, misdeeds ("Have you ever stolen anything from someplace you worked?"). Another kind is the Directed Lie Test (DLT), in which a suspect is told to deliberately lie before being asked the relevant question in order to see the differences in reaction. This is also of, shall we say, questionable validity. The only kind of polygraph test that shows any kind of accuracy is the Guilty Knowledge Test (GKT). Rather than ask "Did you shoot your husband," the GKT attempts to determine whether the suspect knows something that only the guilty party would know (e.g., "Did you know that the knife was thrown out the kitchen window?"). Unfortunately, it's the least valid of these tests, the CQT, that is most often used.

(In the interests of accuracy, I should point out that these comments are regarding the most popular lie detector test, called the Control Question Test. See the sidebar.)

What does the APA have to say about this? Obviously they have a strong, convincing argument demonstrating that polygraphs are incredibly accurate. Or not.

According to its Web site, the organization has collected the results from 80 research projects published since 1980. Of those, 12 showed polygraphs to have 98% average accuracy, 11 showed 92% accuracy, 41 showed 80% accuracy, and 16 showed 81% accuracy.

In other words, of these 80 studies the APA itself mentions well more than half indicate the tests are wrong 19 or 20% of the time! How would you like to be the one out of five who loses his job because of an inaccurate polygraph test? One out of five — most of us would call that unconscionable.

(And job losses due to polygraph-like tests are nothing new. In the 1950s and '60s, several organizations in Canada — including the military, police, and civil service — tested applicants by measuring their perspiration, pupils, and pulse rate when shown pornographic images. Although subjects were told these tests were to measure stress, this cousin of the polygraph was actually designed to 'detect' homosexuality. Nicknamed the Fruit Machine, it cost a lot of people their jobs before it was finally eliminated.)

The APA goes on to say, "While the polygraph technique is not infallible, research clearly indicates that when administered by a competent examiner, the polygraph test is one of the most accurate means available to determine truth and deception." Switching to my Language Guy hat, this is a sentence you can't argue with. Considering that there is no real way to accurately determine truth or deception, saying that the polygraph is "one of the most accurate" ways is like saying "Reading tea leaves is one of the most accurate ways of divining the future." It's easy to compare yourself to zero.

People are more willing to believe that a polygraph is actually detecting lies because, unlike tea leaves, there's science involved—there are electrodes and graphs and machines. Technology just seems trustworthy. Unfortunately it's not, because lie detection isn't based on the technology itself so much as on the interpretation of what the tech says. And human interpretation is always subjective, as 60 Minutes proved.

Here's my advice: If you're ever asked to take a polygraph test, say "No way." If you have to — if it's a job requirement — relax and tell the truth. It won't make much of a difference.

Andrew Kantor is a technology writer, pundit, and know-it-all living in Columbus, Ohio; he's also a former editor for PC Magazine and Internet World. Read more of his work at kantor.com. His column appears Fridays at USATODAY.com.