Judging from this video, it might be easy to train a machine learning algorithm to distinguish lies, possibly even subtle ones, from truth. I'll write that down in my todo list and give an update when I get around to it...