Tuesday, February 14, 2006

A New Lie Detector?

A company called Nemesysco is now marketing the GK-1 Security Access Control System, a device that supposedly uses “Layered Voice Analysis (LVA) technology” to determine whether an individual is lying based on analysis of the person’s voice.

Well, they don’t claim it detects lies, as such, but they do claim that it will identify someone’s “overall emotional state” so that you can “single out potential harmful intentions”.

This is supposed to be a rapid screening system that you can use in places like airports. A subject answers three to five questions that take no more than a minute, and the system responds with a red or green result. I can only guess that a red result means you should take the person aside for more intensive questioning and searching.

While I don’t think such a system is necessarily a bad idea if it works, I have a couple of problems with it. For one thing, where did they find a broad sample of people who really had “potential harmful intentions” to use as the baseline for calibrating their machine? Do they have recordings of dangerous criminals that were made shortly before they committed their crimes? That’s about the only potential source of comparison data that I can conceive, but Nemesysco doesn’t say anything about how they determine what data in a person’s voice will “reveal their real intentions”.

They also say that one of three algorithms their device uses to detect threats is a “polygraph-like analysis”, which sets off my alarm bells. A polygraph test is a trick, not a legitimate test. The polygraph machine is, at best, a stress indicator, and the test can only work if the tester can ask a legitimate control question on which he knows the subject will lie. Furthermore, it’s pathetically easy to create false stress readings on a polygraph, so an informed subject can easily beat the test. If the GK-1 is using similar principles, I wouldn’t expect it to be any more reliable.

Maybe there is legitimate test data for this machine somewhere in the archives of Nemesysco, but I wouldn’t be inclined to trust the thing without seeing the results of some well-designed performance tests. Nemesysco’s website provides nothing of the sort, so I consider their product to be a dubious investment, at best. It may catch the spontaneous offenders, but I expect that determined intruders will find ways to beat the test, meaning that Nemesysco clients will be -- to some extent -- buying a false sense of security.

No comments: