Slashdot Mirror
Challenging The 'Unbeatable' Polygraph
George Maschke writes "Dr. Louis Rovner, a prominent California polygraph operator, has (through PR Newswire) issued a press release titled, 'Polygraph Unbeatable, Says California Psychologist.' All too often, such publicly-made claims by those with vested interests in the perpetuation of polygraphy (a make-believe science that offers make-believe security) go unchallenged. So, I've publicly challenged Dr. Rovner to support his claim and pointed out scientific research that contradicts it, as well as the examples of several notorious spies and a serial killer who have beaten the polygraph. See, A Public Challenge to Dr. Louis I. Rovner."
As I read it he isn't claiming the polygraph to be 'unbeatable' and concedes that some people may be able to beat it but most can't and certainly not by reading a book. More importantly, who cares? Last time I checked polygraphs are generally inadmissible by law.
Yes. I mean no. I mean... I think I'm not sure. Well but... Maybe... Well if you put it that way... erm. No? *BBBBBBBBZZZZZZZZZZZZZZZT* "What is your favourite color?" "Blue! ... No Red!"
"Aaaaaaaaaahhhh..."
I'm sure all the Slashdot readers who are notorious spies or serial killers will take heart at this!
Sheesh, evil *and* a jerk. -- Jade
Otherwise, I'd be good.
I support the FairTax www.fairtax.org
"Overall," says Dr. Rovner, "we are confident that polygraph tests have a 96% accuracy rate when done properly."
If that is true, then if you have 1 spy and 49 honest people, this polygraph will likely falsely accuse two honest people as being spies.
Avantslash: low-bandwidth mobile slashdot.
Most polygraphs work on the idea of remorseful feelings the subject will have if they report a lie in response to a question. Indeed they are beatable - most murderers and other criminals would not give off remorse when asked questions, thus the machine interprets the response as the "truth". This is the main reason polygraphs cant use these results in court.
Thus, when polygraphs are used, it's important it _not_ be the only tool used. For instance, when the USGov't investigates someone applying for a security clearance, they check everything in addition to using the poly. Credit history, school records, school/military diciplinary records, tax records, medical history, family medical history, they perform various psychological exams, they talk to the guy's friends and co-workers and supervisors, and so on. They ask questions about international travels, friends who are non-US citizens, etc.
This way, when someone "passes" a poly, there's evidence to back that up or refute that result. If the investigate report backs up a positive polygraph result and nothing negative is found (or the negatives are manageable), then the guy can probably be given that clearance. Otherwise - the red denied stamp gets pulled out. Indeed, someone can pass the poly and still be denied the clearance - such as a black eye on the credit report (espionage risk - if the guy falls behind on mortgage payments, he could sell secrets to whomever wants them) or a history of alcoholism (clumsiness with classified material risk - if the guy gets drunk while acting as a courier, he risks losing it).
ok, this is a little OT, but i thought it was fascinating enough that i'll post it anyway.
So a few weeks ago, I was driving back to school late at night and was listening to Art Bell (yes, its full of wackos but it's entertaining. Been listening since 7th grade)
Anyway, there was this guest on about polygraphs and plants, yogurt bacteria, eggs, food, etc.
Basically the guest said that if you hook up a polygraph to various "living" things, you can get some sort of reading off of them. If you put stress on/around the thing being monitored, it will react.
For example, if you hook up a polygraph to an egg, and have a dozen other eggs around it. If you take one of the eggs and put it in boiling water, the egg hooked up to the polygraph machine will go crazy.
With plants and yogurt. If you hook up a polygraph to a plant, and have a cup of "live" yogurt beside it. If the yogurt is disturbed (such as stirring up the fruit in the yogurt). This will kill the live bacteria in the yogurt and the plant would react.
Lastly, the guest said that you can't (for the most part) beat a polygraph with anything mjaor (such as if you murdered someone). Why? Because you conscience would get the best of you. The one exception is if you life was in danger. (he didn't elaborate much on what that meant)
And lastly, a link to the show
Is it true that more people vote for the winner of American Idol, than vote for the president? -Ali G.
There was a recent study where a small number of people were able to detect lies with a nearly 100% accuracy. To me, this is far more impressive than a polygraph's results.
What you are referring to is something that is called the "base rate fallacy". This mathematical fallacy occurs when you try to interpret the results of a test without taking into account the frequency of the thing being tested for in the population being sampled.
Taking the claimed 96% accuracy rate as a given, suppose that 1/10K people are terrorists. If I randomly polygraph 10K peple, I'll on average turn up 1 terrorist and 400 false positives. I can only be 1/4 of one percent sure in my result.
On the other hand, suppose I know that 50% of the people working in an office are stealing supplies, but I don't know which. If I test 100 people, I'll get 4 false positives and 48 true positives. I can be 92% positive than any person who failed their polygraph steals office supplies.
The lesson is this: evidence can only be weighed in context. There will probably never be a single test that can determine the truth on its own.
Post may contain irony: discontinue use if experiencing mood swings, nausea or elevated blood pressure.
"I am lying." or "This sentence is a lie."
It's not true or false. ...maybe it would break...
She can tell when I'm lying 100% of the time....
It's better to be the foot on the boot than the face on the pavement. ~~ tkx Kadin2048
I've taken one polygraph in my life. I was 19 and full of that sort of moral superiority that comes from the false certainty of youth. I answered all the questions truthfully, especially the one about whether I'd ever smoked pot. I hadn't and thought anyone who did was a loser. In fact, I felt strongly about the subject.
Afterward, the guy puts his arm around me and tells me I passed and that one lie that I told about the pot wouldn't be held against me. He patted me on the back and sent me on my may.
One anomalous response was interpreted as a lie. A faulty technology had convinced a total stranger that I smoked pot when I never had. The report of that session went to my new employer who didn't fire me but did make the report available to another employee who happened to be my sister. To this day, she thinks I've experimented with drugs when I haven't. After all, what's my word balanced against a neat-o cool technology with all those scribbling pens and sensors and stuff, right?
Polygraphs are bunk. People who make their living in that industry are, by my definition, liars and should be shunned.
Yes, I know I'm only one data point. But sometimes it only takes one data point to know when a technology has failed and is not trustworthy in broad application.
This does interest me in that I am scheduled to take a "lifestyle" poly sometime in the next two weeks as part of a government screening process. I am not at all worried about the poly cathing me on a lie - I have nothing to hide. I more worry about getting a false positive. What does the research say about faulty readings? I've read a little bit on it, and it seems that the practice is too much and art and not enough a science. Can anyone alay my fears?
I took three polygraphs as part of a process to obtain a security clearance (no, I didn't get it.) I believe the effectiveness of the polygraph has little to do with the 'technology', and a lot to do with the theater surrounding the examination.
From Skepdic:
'It doesn't appease me that many defenders of the polygraph know it is junk science but defend its use because many people confess to crimes during interviews done before or after being given the test. The machine may not be able to detect lies accurately but, as Richard Nixon said, "it scares the hell out of people." The end justifies the means.'
I did not become a vegetarian for my health, I did it for the health of the chickens. --Isaac Bashevis Singer
Falls positives is what I am worried about. People being convicted because they were nervous and upset about being charged with something they didn't do.
MMO Quests are like orgasms:
You may solo them, I prefer them in a group.
Polygraphs pick up body reaction. Sadly to few are used and humans are to complex to truly be able to tell why a person reacts.
A simple test is a pedophile image. Both a pedo and a normal person would react with an increased heart rate. The pedo because he is excited, the normal person because of revulsion.
Only when you would start to measure things like blood chemistry and brain activity would you be able to do a true polygraph.
At the moment it is like trying to tell if someone if is lying in an interview by crossing their arms (said to be a typical defensive position). It might just be they are hiding a stain they suddenly spotted, are cold, trying to stop their arms from moving because they are expressive people and been told off about it, just plain nervous about job interviews, trying to hide their beer belly, are just listening to what your saying and really thinking about it.
Polygraphs are a tool, not 100% reliable but an indication for investigators in wich direction they could be looking for more solid evidence.
Those who riducle polygraphs forget one tiny little detail. Most criminals are not smart, prepared, cool blooded and calculated offenders. Just as often as not the polygraph is a bluff wich the criminal will fail just because they know they are guilty and will be found out.
Smart people beating a police polygraph is a total lie. Smart people are never even questioned by the police.
MMO Quests are like orgasms:
You may solo them, I prefer them in a group.
is in Scientology. Those individuals train for years to defeat a lie detector, even if they're not ready for it. The e-meter basically is a lie detector (it's a little hyper-sensitive on any reaction, as is shown from their "rock slam" of the needle bouncing like mad since they don't use the reduced bounce meters) that they train against for years to get to where nothing they say or do will carry a reaction (i.e. "floating").
Naturally, as was said before, you can defeat most polygraph tests with 30 minutes of training, or using the ability to answer the "wrong" question with the right answer for what they're asking you.
The DoD gave me TS-poly SCI/Counterintel/COMSEY
ME... a foreign cartoon character.
I've got an upcoming briefing with Navy Intelligence brass about my upcoming Yankee White investigation.
Ha ha ha... Suckers.
THIS THING CAN TURN ON A DIME, MACROSSZERO STYLE ALSO FUCK BETA, ~NYORON
A novel by James Halperin, sets out a fictional account of what would happen to society IF everyone could know if anyone else was telling the truth, or not.
Imagine politics in that brave new world!
What would happen, sir, if you answered the question "So are still beating your wife?"
On the surface, the press release looks pretty good to me -- the fact of the matter is that only a very small amount of the population has the disicpline, self awareness, and controll to intentionally not get caught lieing on such a test. With access to a machine and a skilled operator and many hours/days/months, etc of practice most people could learn through biofeedback techniques how to do it.
;) and that it is more likely to generate a false negative than a false positive... I.e. It can call people liers who are telling the truth. Furthermore, a pyschopath (in the psychology sense) with no sense of morals or empathy will generally show a very flat response -- they simply don't feel like lieing is some incorrect and their CNS does not respond accordingly. Additionally, some people always look like they are lieing even on the most banal questions like asking a blue eyed person if they have blue eyes, just because you are asking at all. Simply put, if you are ever asked to take one to prove you are innocent, Don't! It's not worth it.
The real problem (and the reason why it is generally inadmissible in court) is that the polygraph measures physiological responses to stress... It has been shown in the past that it is easy to manipulate the results through the questions (which, of course, a skilled operator would never allow/do
Once again, a self serving piece of propaganda by an operator who wishes to protect their revenue stream by making their work seem valid. This is one of those articles that is basically correct scientifically, but is intended to be misread in a misleading fashion -- just because it is difficult to intentionally beat a lie-detector test, does not mean that the test is meaningful, valid, or reliable.
More Caffeine. NOW
not exactly. 96% statistical accuracy means that if you have a population of 49 honest people and 1 spy (or H honest people and S spies for a total population of H+S) that it will pick out the S spies 96% of the time.
.00000939 or 9.39 per million
to have what you suggested, the test would have a 96% "POWER" (or a 4% beta error for n=50).
Alpha error (Type I error) = Probability(X returns false | X is actually true) => false negative
Beta error (Type II error) = Prob(X returns true | X is actually false) => false positive
The actual probability that a 96% accurate polygraph machine would do what you describe (1S + 49 => 2S that are really H) is:
(.96^47)(.04^3) =
This says that it got 47 people right (47H) and 3 wrong (2H + 1S). THAT is what a 96% accuracy means.
Do I believe the doctor's claim that the machine is that accurate? Hell no. I think the doctor that created the claim has a failed understanding of whata statistics actually mean and is giving a stat for something else entirely.
-Ab
Nothing fails quite like prayer.
(.96^47)(.04^3) = .00000939 or 9.39 per million
.18, and the probability of having at least 3 positives is .32. That's a 1 in 3 chance that you are falsely accusing 2 or more people.
What distribution are you using?
The model it looks like your are describing is binomial with probability of success 0.96. If that's so, the probability of having exactly 3 positives is
Education is a better safeguard of liberty than a standing army.
Edward Everett (1794 - 1865)
It is the binomial, but you aren't applying it properly.
.04.
.96 chance the machine will return H(onest) and .04 it will return S(py). If the person is a spy, there is a .96 chance it will return S(py) and .04 chance it will return H(onest).
.96 * ... (44 more) ... * .96 * .04 * .04] * S[.04] .96^47 * .04^3 or 9.9 per million.
.04 * .04 or .0016 or 16 per 10000 attempts.
... well, we don't know. They may work better, they may work worse.
A 96% probability says that "We are 96% confident that the Lie detector will get it right." So p=.96 and q (or 1-p) =
To expand:
If the person is honest, there is a
The original poster postulated that if there was 1 spy and 49 honest people, that it would return 48 H's and 2 S's and chances were the 2 S's would be false positives that were really H's. This says that out of the 49 H's, it got 47 right and 2 wrong and of the 1 S it got it wrong. The probability is:
H[.96 *
Which simplifies to
What the original post got confused is what the 96% meant. It means that the machine gets it RIGHT 96% of the time, not that it returns H 96% of the time. Out of 100 tries, it is EXPECTED that 4 will be wrong, but not necessarily. it would not be unfair to believe 0 would be wrong or that 10 would be wrong. We don't know which type of wrong would occur (false positive nor false negative) because we wouldn't know the true state. The easiest way to avoid this is to administer the test twice. The probability of screwing up twice on the same person is
Keep in mind, that the actual probability is a confidence interval, (usually 90%, 95%, 99%, or 6sigma). Let's say it's 95%. So the 96% claim is actually saying, "We are 95% certain that the machine will give the correct output 96% of the time." This says that of 100 machines, 95 of them will work at the 96% level. The other 5
All of this is based on the Binomial approximation of the Normal. Vice versa, there is the normal approximation of the binomial. Both are accurate for large enough sample sizes (n>=36). If n36, then correction factors are needed for the bias and the confidence interval increases (which is a bad thing).
-Ab
Nothing fails quite like prayer.
"He's also never heard of people like me who have consciences but can act against them at will providing I can prove to myself that what I'm doing is correct despite feeling it is wrong. "
:-)
This is totally off topic, but what standard are you using when proving to yourself that your conscience is wrong? I'm sure the philosopher Hume would have liked to know, as this is believed to be impossible in moral questions
But I don't think you are wrong when you say you do that. Millions of people do it all the time, usually by accepting someone elses stated conscience as being more important than their own sense of right and wrong...
Just so we can continue, could you give an example?
xkcd is not in the sudoers file. This incident will be reported.
OK, we agree, if nothing else is specified, that "96% accurate" means: If the person is honest, there is a .96 chance the machine will return H(onest) and .04 it will return S(py). If the person is a spy, there is a .96 chance it will return S(py) and .04 chance it will return H(onest).
Now you calculate, for a population of 49H, the chance of detecting 47H + 2S(correct) + 1S(false):
Talk about "know your statistics". It should of course be: .96^48 * .04^2 * 49!/(47!*2!) = 0.25. Then there is the chance of detecting 48H + 1S(correct) + 1S(false), 49H + 1H(false) + 1S(false), and so on, and the outcome is that you can expect about 2 false positives and 1 correct positive.
The easiest way to avoid this is to administer the test twice.
That is assuming that the chance of a false positive is something in the machine instead of something in the person being examined.
Avantslash: low-bandwidth mobile slashdot.
is that there is this perception among the general populace that polygraph tests are (nearly) infallible.
HAND.
Talk about "know your statistics"....
.96 chance of being correct regardless of whether he's H or S.
.04 H).
We calculated 2 different things. You calculated 3 S (1 correct, 2 false). That wasn't the original poster's posstulate. His was 2 S (both false) which implies one of the H's is false as well. As for the factorial part, that isn't used because the order is irrelevant and the individual trials are not dependant on each other. The first guy has a
I was calculating the odds of exactly 2 false S and 1 false H. You calculated the probability of any 2 False answers (S or H, as long as they were false). Remember, the actual spy's probability is reversed (.96 S,
-Ab
Nothing fails quite like prayer.
It is the binomial, but you aren't applying it properly.
.96 * ... (44 more) ... * .96 * .04 * .04] * S[.04] .96^47 * .04^3 or 9.9 per million.
.04^3 means you will get exactly 47 successes followed by exactly 3 failures in exactly that order, which seems strange. However the probability of exactly that outcome is what you say, about one in 9 million.
Okay, well what is the probability distribution function for the binomial distribution?
The probability is:
H[.96 *
Which simplifies to
.96^47 *
If you get 1 failure, followed by 47 successes followed by 2 failures, your model does not permit that possibility. You require all 3 failures to come at the end. There are about 20,000 different ways to get 47 successes and 3 failures in any order (50 choose 47 - the binomial aspect for which the distribution is named).
If your argument were a coin, you would say the probability of getting 1 success (say a head) in two tosses is (.5)^1*(.5)^1 = 0.25. But we all know that sample space is { HH, HT, TH, TT } = 0.5 probability of exactly one head. However the probability of getting 1 head followed by 1 tail is indeed (0.5)^1*(0.5)^1.
Education is a better safeguard of liberty than a standing army.
Edward Everett (1794 - 1865)
As for the factorial part, that isn't used because the order is irrelevant and the individual trials are not dependant on each other.
Whoa, missed this earlier. I think you have a major mistake in there: The factorial part IS USED because the order is irrelevant and the individual trials are not dependant on each other. The factorial part accounts for all possible combinations of 47 successes and 3 failures.
Consider the simplified example of three people taking the polygraph test and you'll see you must use combinatorics to arrive at the proper probability.
Education is a better safeguard of liberty than a standing army.
Edward Everett (1794 - 1865)