I see on Footnotes that a new free software package, LiarLiar, has been released that claims to use “Voiice Stress Analysis” to detect lying. The blurb is carefully written to suggest that it’s something short of a lie detector – “Stress can be one indicator of whether or not a person is being truthful in the statements and communcations that he or she is making.” But the screenshots and the app’s name are clear – this is a lie detector. And it’s pretty clear that lie detectors in general just plain don’t work. It’s pseudoscience.
In particular, according to a review by the National Academy, there’s no good evidence that voice stress analysis works to detect deception. As with polygraphs, that means that the problem of false positives (people wrongully accused) and false negatives (people getting away with lying) will swamp any actual usefulness the tool can offer. To the extent that people actually believe what the software is telling them (and why else would you use it?) this clearly could do more harm than good.
Image editors can be used for counterfeiting, p2p applications can be used for trading illegally copied software, and encryption (such as gnupg) can be used to encrypt terrorist communications. Should we stop research and development in the above applications because they can do more harm than good?
Where did you get the idea that I think anyone should stop these guys from doing anything? I said no such thing.
I’m not suggesting the software should not be developed, and I’m not suggesting research into voice stress analysis should be stopped. I’m just trying to point out that anyone who actually believes that these things can detect lies is a fool, and will do themselves more harm than good. And any claim offered by the developers of LiarLiar (or any other software, free or commercial) regarding its ability to usefully distinguish truth from falsehood should be viewed with skepticism.