I think part of the issue is that errors are not evenly distributed. There are many more false positives (detecting a lie when the subject is truthful) then false negatives. Most people assume errors are random, in this case they are not. Of course, this is combined with the other unreliability of the measurements make it close to a waste of time.

[https://www.psychologytoday.com/us/blog/the-human-beast/2013...]