Slashdot Mirror
World's Most Accurate Lie Detector
An anonymous reader points to this "interesting article about the world's most accurate lie detector. It seems they are getting real close to Voigt-Kampff. Watch out fellow replicants."
← Back to Stories (view on slashdot.org)
I'm the one, it was me all along, I put the rubber snake on Mrs. Montgomery's chair in Grade 3!
I invented New Coke!
I shot J.R.
...and I came up with the idea for Pop-Under Advertising!
I throw myself on the mercy of the /.court
It's already more accurate than polygraphs. The dirty little secret about that, of course, is that a polygraph machine needs to be 'interpreted.' Polygraph operators are more or less dowsing for the truth. Remember all those keyboards back in the 80's that let the autistic comminicate with the rest of the world? Same kind of thing.
What's scary is that this test can be used without someone knowing. A regular lie detector is hard to sneak on someone.
I read in a one of those women magazines ( I was getting my hair cut, okay?! What else was I to do?) various ways to tell if people are lying. It was the usual bag of tricks - see which direction their eyes go. (Something like if they look left, then they are accessing the left hand side of the brain which is the 'artistic' side - so they are making something up. If they look right then they are trying to do memory recall.)
Anyway, they said that experts had looked at videos of clinton telling everyone that he hadn't had sexual relations, and that the experts had all said he was a very bad liar, and it was really obvious..
Dunno how true this is...
"He would also like to see it used during TV interviews with politicians, so audiences could tell whether they were being spun a yarn."
That would be illegal within a month.
If it turns out to work, it will be banned for use on TV during politicians' speeches. (The first amendment will be found not to apply.)
Expect to see software that can edit out the facial gestures, in real time. Politicians will only release footage that has been filtered by it in their own offices.
Software that detects that filtering will be banned.
Yeah, good idea, but its not necessarily the fact that what you're stating is a truth, but that you have a very firm belief in what you're saying that constitutes a truth. (and vice versa)
They often mistake nervousness (embarassment that you've been asked to step aside for an "interview" in the airport), vs. outright fibbing.
I have observed that many of the tests of these devices involve studies where the subjects are encouraged to fool or beat the machine and they base their results on that. However, the subjects are probably already familiar with the tests and are (I imagine) not in fear that something bad will happen to them if they fail.
In the case of the proposed settings (airports, police station), failing this lie detector test could lead to Bad Things. I wonder if these machines pick up the facial features that arise when "the subject is making an attempt to provide a satisfactory response" and THIS mental activity causes the facial feature changes the trained AI responds to.
I suspect the AI was created by evolving a neural net based on human trials (2-layered? -they even mention that it monitors 24 channels; this wouldn't take too long to evolve). Hence it is trained to the testing conditions, and these premises of what they are testing, and/or the methodology of the training may be incorrect.
Maybe the question really is: do we care that the device may be responding to mental stress vs. truth/lie telling? Some might consider it more useful that it just measures stress (which in a question and answer session can be caused by fabricating responses). It could indicate when someone should have a "bad feeling" about someone or their claims. At the same time I feel that if my allegations are true, this device also should (in addition to standard polygraphs) not be used to generate evidence for trial proceedings.
Anyone with more insight into this care to add?
THIS THING CAN TURN ON A DIME, MACROSSZERO STYLE ALSO FUCK BETA, ~NYORON
The problem is that people tend to trust machines, for some bizarre reason. Just because the machine isn't subject to emotional bias doesn't mean it's output is worth anything. Polygraphs are barred from use in the courtroom -- because of their unreliability AND tendency to unduly influence the proceedings -- yet the government is using them with abandon as we get tough on terrorism by treating citizens like suspects.
Polygraph administration and analysis turns out to have huge subjective factors despite attempts to appear otherwise. Some people are especially accomplished liars; it doesn't hurt to be a sociopath who really doesn't have any emotional reaction.
Perhaps this system will have some use -- let's remember that even the standard Voight-Kampff got beat! (why did Deckard press on with 2x more Q's?) -- but I hope it and other techno-gadgets don't lessen emphasis on good old-fashioned security and common sense.
CmdrTaco: Describe in single words. Only the good things that come into your mind. About CowboyNeal.
... I'll tell you about CowboyNeal.
Me: CowboyNeal,
"Can of worms? The can is open... the worms are everywhere."
Hey - it missed 2/7 lies in the example that the article mentioned. That doesn't strike me as useful.
Used at an airport: 50000people/day * 28.57% false positives = a lot of messed up flights.
Maybe after they get to better than 99% will it really be useful.
No, I don't trust in god. He'll have to pay up front, like everybody else.
I wonder when banks will start to use this for loan screening. I think there are no laws to prevent them from doing that. A simple hidden camera behind a mirror.
Now this thing is very easy to use and operate (contrary to other known methods). And will probably cost peanuts. Which means with 87% accuracy 13% of people who are honest and good citizens will be denied loans, insurances, jobs, etc.
That's more than 1 in 10. Scary, huh?
But how this ends is easy to predict - through generations of breeding more successful people at looking thruthful will prevail and the indifferent faced bustards will perish in extinction.
*makes an honest face*
I know it's a joke, but the problem with this is distinguishing falsehood from lying. A lie detector tries to figure out if what you're saying corresponds with what you believe to be true (subjective truth), while physics theories deal with correspondence between the theories and reality. So sorry, but it doesn't look like it'll work.
Or maybe they could just apply it during a replay.
It seems they are getting real close to Voigt-Kampff. Watch out fellow replicants.
"Reaction time is a factor in this, so please pay attention."
I'm not a replicant, but after that movie, I'd still run if I heard that phrase. It's like Heisenberg's Uncertainty Principle. The test dealing with my response will, in turn, affect my response. That's what polygraphs base themselves on, that if you lie, it'll detect your nervousness due to the fact that the test looks "real"
Only in slashdot are posts of solidarity modded at -1 Redundant, while posts of antagonism are modded as -1 Flamebait.
For an illuminating article on how lie detectors work, and how they can be used falsely, see this website.
a perfect lie detector. I involved in a research which used mild electric shock when the lie detector returned confidence level below a certain theshold. We found that mild electric shock could really improve the accuracy.
:)
Later I found that if I inverse-proportioned the intensity of electric shock to the confidence reading of the lie detector the accuracy would sky-rocket.
Unfortunately, my research was called to stop when a prof. used an obvious lie on my prototype. The confidence level reach 0 and the detector output maximum electric shock which knocked the prof. 10 feets away from the desk.
Damn. it was that tyrannic dictatorship destroyed a great research like that!
Just what the world needs, another "lie detector". The fact a bunch of college types couldn't keep their faces straight while lying is hardly proof of anything but how stupid they are. I have no doubt a REAL "professional liar" (i.e.: scummy corporate lawyer, CIA goon, etc.) would be able pass the new "lie detector". I also have no doubt this fact will do nothing to stop the sale of this new fraud, given its obvious potential role in harrassing airline passengers and the profit involved:
"Plans are being drawn up to test the machine as part of airport security. Dr Bandar says that by asking passengers simple questions on arrival, customs officers would be able to determine who to search for drugs or other items of contraband."
"The research team have secured a patent application for the system, which they say could bring in billions of pounds for the university and the scientists involved."
Pathetic.
I doubt it is better than my wife...
Recently, however, there was a committee put together by the National Academy of Sciences to study the scientific validity of the polygraph and related lie-detection methodologies, both in the lab and out in the real world. If you want to read the report, you can find it online through the NAS's publishing website.
As for what the report says about micro-expressions (from page 164), it notes that previous studies of micro-expression detection were able to achieve rates of up to 75% accuracy, far better than chance. It goes on to note that such methods, at the time of the report's writing, were labor intensive but that recent work in automating the process held promise. It seems that this new AI work is showing that promise to be well founded.
While 75% accuracy is good, it is nowhere near what would be needed in a diagnostic tool. Even 95% accuracy isn't good. (Note, according to signal detection theory, we really should be talking about percentages of false positives and false negatives, but I digress; let's assume the accuracy works the same for both error types.) 95% accuracy would mean that for every 100 truthful people interviewed, 5 would be judged as liars. For every 1,000 people, 50. Inasmuch as such false positives can ruin lives, careers, marriages, reputations, etc., that rate is too high. Likewise, if 5 out of 100 liars slip through, that isn't a test I feel confident using for national security concerns. Even "good science" needs to be damned good to fill the shoes made for the "lie-detector".
-tcp