Lie detectors can predict whether an individual is lying with an 80-90% accuracy rate.
In fields where lie detectors have been adopted, they are having success. In 2007, Harrow Council in the United Kingdom introduced software that measured whether or not a person was telling the truth from undetectable tremors in their voice to help combat benefit fraud. In the early stage of the pilot, the software caught 126 claimants who were deceiving authorities and illegally claiming benefits.  Lie detectors clearly work in the legal system. They are able to determine truthful statements from lies and should be employed in criminal trials to determine guilt.
Anything less than 100% should not be admissible in court. When the consequences are so high, the restriction of an individual’s civil liberties, there can be no room for error. Even if lie detectors are 95% accurate, as some practitioners claim, in the best-case scenario, 5% of suspects would have their names sullied over a crime they did not commit.  In polygraph testing, false positives are also more common than false negatives. In one study, an accused person who was innocent had a 50% chance of displaying a guilty reading, reducing the fate of an innocent suspect to little more than a coin toss.  In addition, almost no studies on the effectiveness of polygraph tests have tested their accuracy on children, those with mental disabilities, people speaking their second language, or even those of below-average intelligence (test subjects are usually university students of above-average intelligence). There simply isn’t enough evidence to say they can detect lies to a high level of accuracy. 
[P1] Lie detectors work to a high level of accuracy. [P2] Therefore they should be admissible in court.
[Rejecting P2] With the stakes so high, anything less than 100% accuracy that could influence a jury's decision should not be admissible.