The European Union is to deploy lie detectors at airports to question incoming travellers. Sure, they’re fancy and modern and new because they employ artificial intelligence. The problem with all of this being that lie detectors themselves don’t work. So, to the extent that these are lie detectors, this will not work obviously enough.
Further, the problem with lie detectors is not that they will work with only a little bit of refinement, or just that next iteration of technology. It’s that the basic underlying assumption itself is incorrect, that there are unfakeable or unmissable physiological signs when someone is lying. It’s that base assumption which is wrong. And the more professional and more trained the person being questioned the more it is wrong. The outcome of this being that the really bad guys, the ones you really want to catch, are precisely the people who pass lie detector tests and thus don’t get caught by their use:
Passengers at some European airports will soon be questioned by artificial intelligence-powered lie detectors at border checkpoints, as a European Union trial of the technology is set to begin.
Fliers will be asked a series of travel-related questions by a virtual border guard avatar, and artificial intelligence will monitor their faces to assess whether they are lying.
Those signs being looked for just aren’t there – at least among those who have been trained. Heck, even a good poker player can beat the things.
‘We’re employing existing and proven technologies – as well as novel ones – to empower border agents to increase the accuracy and efficiency of border checks,’ project coordinator George Boultadakis of European Dynamics in Luxembourg.
‘iBorderCtrl’s system will collect data that will move beyond biometrics and on to biomarkers of deceit.’
Yes, we humans are indeed pretty good at telling when someone is lying. But there simply is no one single nor combination of facial or any other physical signs which exist when someone is lying. Nor any other set whose absence shows they’re not lying. Thus any system based upon a certainty of catching such is bound to be wrong.
According to early testing, the system is around 76pc accurate, but the iBorderCtrl team say they are confident they can increase this to 85pc.
Neither number being useful enough.
They do say that if people are pulled by this first level of the system then off they go to another and more intense level conducted by actual people. But that’s not the problem. False positives will indeed be – largely enough – caught by such a system. The AI says “bad’un” the human level is able to clear them. False negatives though, they slip through the net entirely. A bad’un who isn’t picked out at that first level because they’re good at lying is then not checked at all. And given the ease with which those biomarkers can be simulated or avoided anyone truly a bad’un will have the training. Thus this AI system is entirely useless at the actual job at hand, catching the bad’uns.
But then, you know, the EU, white hot heat of technology and all that.