Even if we did develop a lie detector with high accuracy, how would we know?

There's good reason to believe that lie detectors might work much better in the laboratory, on test subjects telling inconsequential lies to experimenters and knowing they'll go home, than on real suspected criminals telling vitally important lies (or truths) for life-or-death stakes.

Nice one! Now that you mention it I remember seeing an article about FMRI being used in that role a while ago. Excellent example.