"But they are still a useful tool that can catch most liars most of the time."
There is a boatload of studies to show it, too.
Could you provide any references to this "boatload" of studies? I agree that there are plenty of studies that show polygraphs are not perfect, or even 90% effective (a rough threshold for use in court), but I am aware of NO studies that show they are less than 60-70% effective (where 50% = random).
Generally speaking, polygraphs have been shown to be less good at detecting lies than a friend is.
There you go. So a polygraph enables a stranger to tell if you lying as well as your friends can. Since I can tell if my friends are lying FAR better than I can tell if a stranger is lying, that is pretty damn good.