It doesn't mention how mind-bendingly wrong the conclusions of a piece of software would be that had been developed using observations of human behavior, analyzed and codified by other humans, to try to isolate the specific behaviors present in the one part of human behavior no human has ever shown he or she is all that good at figuring out: when someone else isn't telling the truth, the whole truth, or something in addition to the truth.
Software is a lot more objective than human perceptions, because it's more stupid. The algorithms that allow analytical software to reach conclusions are written by humans codifying data that are entirely subjective, then trying to verify those subjective judgments and reduce the error rate by piling on more tests built on more series' of subjective judgments.
However learned the explanation of theory or practice sounded in some smoke-filled underground bunker at DHS when this idea was pitched, no matter how many graphs were involved or how many white coats the scientists wore while swearing they could do this, no one can do this.
Reuters illustration
