Lie detection via facial cues / body language strikes me as the sort of thing best done with a neural network: thousands of tiny noisy cues each with very weak

Lie detection via facial cues / body language strikes me as the sort of thing best done with a neural network: thousands of tiny noisy cues each with very weak correlations that need to be combined with solid statistics. Humans can't process this many cues at once and bias drowns out the signal, but a smart NN hooked up to a powerful camera is another story. The 'active' method in the article is useful, but has a limitation: you need to be able to ask questions in real-time. Collating the answers and automating truth evaluation would be a pretty interesting AI problem. It should also be possible to have an AI formulate the statistically optimum questions to ask, Akinator style. P.s. That would be the killer app for Google Glass, right there. Would also increase chance of being thrown out of pubs by 1000%, but hey.