Automatic Lie Detector

Coming soon to airports:

Tested in Russia, the two-stage GK-1 voice analyser 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 general, I prefer security systems that are invasive yet anonymous to ones that are based on massive databases. And automatic systems that divide people into a “probably fine” and “investigate a bit more” categories seem like a good use of technology. I have no idea whether this system works (there is a lot of evidence that it does not), what the false positive and false negative rates are (this article states a completely useless 12% false positive rate), or how easy it would be to learn how to fool the system, though. And in all of these trade-off discussions, the devil is in the details.

Posted on November 21, 2005 at 8:07 AM26 Comments


Steve L. November 21, 2005 9:40 AM

I’m not all that familiar with how lie detectors work, but I’ve always wondered if you are just nervous about the whole process in general won’t that set off the detector? I mean say they ask you are you planning to hijack the plane – I may get so nervous that they think I might actually hijack the plane that I’ll set the thing off. Or am I off base here and they don’t work like that. To me that would be my biggest concern in using this type of device.

jammit November 21, 2005 10:03 AM

In an effort to make the machine more reliable, would it be ok for me to answer “yes” to every question? If it asks me “are you a bad guy?” and I answer yes, will the test taker accept the machine saying I’m lying and let me go, or am I going to “dissappear”?

Craig Hughes November 21, 2005 10:30 AM

Bruce, couple of observations on your comments:

  1. “Automatic systems which divide people into probably fine…categories seem like a good use of technology”. I thought one of your biggest peeves was automated systems which automatically eliminate people from needing secondary screening?

  2. “A complete useless 12% false positive rate”. Surely a high false positive rate will actually in this instance work in favor of screening out actual bad guys, assuming that the false reponse isn’t easily controllable by the attacker, since the higher the false positive rate, the more this lie detector test acts like a random sample. Though perhaps what you mean is that you think a 12% random sample is too big (assuming that in real life the actual positive rate is close to or equal to 0)? Presumably this would depend on what ends up happening to those 12%, though with a response rate that high of course, the humans operating the machines will just end up ignoring the machine’s results.

  3. If you have trained personel using this machine, then doesn’t this to some degree end up becoming a bit like the famous El Al “interview” test? Assuming that the questions asked by the machine/operator are random and not scripted.

RSaunders November 21, 2005 10:33 AM

Voice stress analysis has been a popular parlor trick for ages, now repackaged by the likes of the National Institute for Truth Verification ( as an investigation technology.

However, the National Academy of Sciences study on the topic (available online at concluded, “Overall, this research and the few controlled tests conducted over the past decade offer little or no scientific basis for the use of the computer voice stress analyzer or similar voice measurement instruments as an alternative to the polygraph for the detection of deception. {page 168}”

These folks need the link to the cartoon in your prior blog entry.

Ben W November 21, 2005 10:38 AM

I’ve heard a rumor about lie detectors. Bear with me a moment.

They rely on physiological stress (either galvanic response or voice tremors) – and so anything you do to cause stress while they are testing will immediately lead to a false positive.

The rumor is that by squeezing ones largest muscles when answering a question, pretty much all measures of physiological stress skyrocket. It just happens that one’s largest muscles are in the buttox, for which rapid tensing isn’t as noticable.

Any thoughts on this rumor?

Ben W November 21, 2005 10:44 AM

Also, regarding #1 by Craig Hughes;

This is hardly an automatic system. It involves a trained person placing something on your head and talking to you.

An automatic system would be a “I am officially secure” identification card or PIN number, issued only to those who qualify (or pay) for it. An automatic system relies on predefined data, and largely ignores one’s current state.

another_bruce November 21, 2005 11:04 AM

there’s no such thing as a lie detector, there is only a “polygraph” which is not based on science. further info can be found at the polygrapher is, in essence, just another interrogator using trickery to fool and intimidate a subject. yes, it’s possible to fool a polygrapher in return, one poster mentioned flexing the butt (caveat: some polygraphers sit you down on a special chair that can detect this). there’s also the thumbtack in the shoe trick, the biting the cheek trick, many of which can be detected on close observation, but no amount of close observation can detect me artificially summoning waves of guilt and angst to muddy the polygraph baseline when i contemplate the dark day i accidentally ran over my high school girlfriend’s irish setter. in the recent “runaway bride” case, the fiance was negotiating the terms of his polygraph with law enforcement, he wanted it to be public, but that was a show-stopper for the cops. science does not fear to show itself, if it did there would be no science.
voice stress analyzers, no better than conventional polygraphs, and possibly worse in the format described, headphones and yes-or-no questions. these words can be spoken so quickly that there is no time for the voice to tremor, conversely, to take the system down in advance of a terrorist attack, decoys can manufacture tremors too, enough so that the system managers will tire of apparently false positives and turn off the machine.
my voice is not quavering because i’m a terrorist. it’s quavering because the foxy marketing rep in the low cut dress just bent over to pick up her ipod and flashed me!

Joe Buck November 21, 2005 11:30 AM

This should be obvious, but it wasn’t to at least one commenter.

The reason that a 12% false positive rate makes the system unusable is that one out of eight passengers will get flagged, and 99.999% of them will be completely innocent. Worse, it will selectively pick out the most nervous flyers (since “lie detectors” are really fear/stress detectors). Within two weeks the security agents will be ignoring and cursing the device, those passengers repeatedly flagged by the device will complain, loudly, and the airlines will push to junk it.

The other problem is that it’s also going to have a false negative rate: sociopaths have no problem passing lie detector tests, and people can be trained to pass such tests. If we could be certain that the 12% picked out by the device would include all the bad guys, maybe the hassle would be worth it. But we can’t be.

Bruce Schneier November 21, 2005 11:52 AM

“I thought one of your biggest peeves was automated systems which automatically eliminate people from needing secondary screening?”

I dislike automated systems that do that badly, which most do. I might have been too subtle in my original comment….

Shura November 21, 2005 11:54 AM

Is that the same lie detector where agents looking into buying one hooked it up to the salesman, asked whether it was working, and got “lie!” when he said “yes”?

I remember reading a story about that recently, and I do recall it was an Israeli firm who had built the thing, but I’d have to look it up…

Arturo Quirantes November 21, 2005 12:52 PM

Imagine the test goes like this:

Were you born before 1945? Answer:no

Did you belong to the SS or other Nazi associations during 1939-45? Answer: yes

As I was born after 1945, the first answer is true (the guard can also look at my passport, or my face). However, is answer #2 a lie? Technically, yes; but as it contradicts answer #1, it cannot be taken as a true answer. Right, it is not correct, but clearly it is not intended to be taken as a true answer.

On second thought, I can imagine the passengers planning a denial-of-service attack: everybody agrees to answer “yes, I’m going to kill the president” and see if the airline people really dare to leave everybody in the ground. Silly? Well, after six hours waiting at the lounge, anything can happen…

bunk November 21, 2005 1:21 PM

AFAIK polygraphs do not have a very impressive track record. Polygraphs are rarely admissible in court. Aldrich Hazen Ames, who was a double-agent for the CIA successfully passed two polygraph tests. I doubt that a device that only measures “tremors” in a voice will come close to being as effective as polygraphs.

Roy Owens November 21, 2005 1:34 PM

The guy wants development money. So does that other guy with the new model dowsing rod.

Since the Patent Office recently slipped through an antigravity machine, and one powered by ‘hydrinos’ (‘shrunken’ hydrogen atoms) he can get some patents out of this and make money licensing.

Koray Can November 21, 2005 2:53 PM

Holy cow! If this thing works, why just limit its use to airports ? Let’s install these on every corner and make everyone state if they are going to commit a crime tonight… Cheating spouses, card counters, tax evasion, etc. can all be caught by this ! If it’s good for the airplane, why not !?

What happens if I show a ton of stress, but carry nothing on me to prove that
I am going to commit anything ? Are they going to deny me based on just my stress levels ?

Ari Heikkinen November 21, 2005 4:59 PM

Seems to me like a good idea to profit from the seemingly infinite money wasted for counterterrorism.

Moderator November 22, 2005 2:26 PM

Chris, trackbacks generated through a third-party service tend to get held for review, since they look like spam pings sent through a compromised PC. They’ll appear on the site as soon as Bruce or I has a chance to check the filter.

I’ve added you to the whitelist, but you’ll probably run into the same thing with other blogs running Movable Type 3.2.

Roger November 22, 2005 11:13 PM

I agree that in general, polygraphs are junk science, little more a prop to help an interrogator frighten his suspect into confessing. I am skeptical of the false positive rate being as low as 12%.

Having said that, a key point which several posters seem to have missed is that this device is not being used to replace searches, nor to add a security check in places there were none. For those sorts of applications a false positive rate of 12% or even 0.1% would indeed make it a white elephant; however it is actually being used only to refine the process of sampling passengers. For this application, a 12% false positive rate is very far from useless. If the system designers understand Game Theory, then so long as the sum of the false positive and false negative rates is less than 100%, it adds some positive information to the process and they can use this to optimise their strategy. How much it improves things depends on several factors but turns out to depend most strongly on the fraction of passengers searched. The smaller that fraction is, the more dramatic are the improvements from the additional imperfect information.

For a simple example, compare searching 12% of passengers completely at random to searching 50% those who fail the voice stress test and none of those who pass. Even this crude strategy roughly TRIPLES the odds of detecting contraband *, whilst simultaneously halving the total number of searches and their attendant delays, costs, discomfort and inconvenience. Much more sophisticated dynamic strategies are possible, incorporating information from a variety of sources such as Customs officers’ instincts and experience.

  • This holds over a fairly wide range of reasonable looking parameters. To do the calculations we need to know several other parameters, e.g. the false negative rate. Unfortunately they do not state what the false negative rate is, but to simplify the argument I assumed this is under equal error conditions, i.e. false negative is also 12%. Fortunately — and surprisingly — for false negative rates much under 50% the result is not especially sensitive to the exact value, i.e. the cartels have to make their “mules” extremely smooth liars in order to get only modest improvements in their ability to slip through the net.

Arturo Quirantes November 23, 2005 5:08 AM

@ Roger:

Your points are valid and good. Unfortunately, the politicians who ordered this -and other- detectors didn’t probably think if this. They want quick, easy, foolproof solutions. And so do the 18-year-old-turned-into-security-guard. They want a magic box that beeps when, and only when, the bad guy arrives. Unless they learn that magic boxes don’t exist, detecting devices will probably do more harm than good. Of course, if all you want is a false sense of security (plus the votes you’ll get out of it), then fine.

Clive Robinson November 23, 2005 5:19 AM

@Ben W
@Steve L

All lie detectors that use external indicators (breathing rate, skin resistance, pulse rate etc) can be cheated with training. For instance sports people involved with things like shooting (guns and arrows 😉 or other precision activities learn to slow their pulse rate and breathing to improve their accuracy.

So yes (Ben W) flexing “your butt” does change the readings, however, most of the more sophisitcated lie detectors look at the timing from the question to the physiological response, apparently a concious action takes longer than a subconcious action therefore (in theory) the two can be told apart by training questions.

It has been said a number of times on “investagative programes” that the Scientologists trained their own people with their own version of a lie detector. It appears they now sell these (along with the L Ron Hubbard novells) as a way of dealing with stress in the modern world…


On another point lie detectors that use external factors only work if you yourself actually belive you are telling a lie, and that it has significance to you, if either point fails then so does the lie detector, as you exhibit no external stress indicators.

Added to that you need to eliminate other forms of stress that might be co-incident (such as the tester making a movment or a light flashing or a fly buzzing in the room etc etc) with questioning.

A little while ago I posted a URL to one of Bruce’s blogs about using brain scanning as a lie detector, apparently the results from this do detect sociopaths and other types who do not show external stress factors.

It works by showing what parts of the brain react and in which order during and after the question. As it’s very new technology I guess only time and testing will show how good or bad it is.

Ben Lie November 30, 2005 11:05 AM

For example simply asking “are you a terrorist”(without any checks) has much smaller false positive rate. If they only focus on false positive rate, they has no clue what they are talking about, and by all signs it’s same as snake oil cryptography.

Heck, there isn’t any reliable fingerprint locks (most can’t distinguish fingers of different persons at all). And it is relatively trivial thing. And it is easy for purcharger to check if lock at least somewhat works. What can you seriously expect from thing that can’t be really checked and whose false positive and negative rates is unknown (you can only speculate)?

Also, no matter how clever and accurate you can be, if you don’t really know false positive and negative rates, thing is completely useless.

Yes, in naive misapplication of game theory to real world, you take values of false negative and positive rates as absolute truth that applies to everything, and then use nice formula to compute how much this improves your safety. But it is misapplication of game theory.
You do not and can not know real false negative rates you get with people who are willing to do something special to get through.
It’s just like applying numbers theory to determine secureness of Caesar’s cipher: there is 26! possible keys so Caesar’s cipher is very strong security technique.

andre March 7, 2006 6:56 PM

is there on the market divice which detects fear and stress based on blood pressure heart beat or so on without talking and aswering questions.something with integrated alarm so it can be adjusted or rather calibrated to differantes levels.can you please let me know.Thank you. andre

Leave a comment


Allowed HTML <a href="URL"> • <em> <cite> <i> • <strong> <b> • <sub> <sup> • <ul> <ol> <li> • <blockquote> <pre> Markdown Extra syntax via

Sidebar photo of Bruce Schneier by Joe MacInnis.